Atomic spectrometry update: review of advances in the analysis of metals, chemicals and materials

Robert Clough a, Andy Fisher *a, Bridget Gibson b and Ben Russell c
aSchool of Geography, Earth and Environmental Science, University of Plymouth, PL4 8AA, Plymouth, UK. E-mail: afisher@plymouth.ac.uk
bIntertek Sunbury Technology Centre, Shears Way, Sunbury, Middlesex, UK
cNational Physical Laboratory, Nuclear Metrology Group, Teddington, Middlesex, UK

Received 13th September 2023

First published on 6th October 2023


Abstract

This update covers the literature published between approximately June 2022 and April 2023 and is the latest part of a series of annual reviews. It is designed to provide the reader with an overview of the current state of the art with respect to the atomic spectrometric analysis of various metals, chemicals and materials. Advances have been made in several areas. For cultural heritage scientists, it is worth noting that, after many years of inserting analytical data into chemometric tools in an attempt to elucidate provenance, several free to use algorithms that take an inexperienced analyst through the entire process have been developed. It is envisaged that such algorithms will also be developed for forensic samples. In other areas, there has been a big increase in papers reporting methods for analysing coals, with many of these applications employing LIBS. This increase is presumably a result of the use of coal increasing rapidly in some countries. The use of LIBS for analyses is still growing in many areas. For ferrous and non-ferrous metals, the development of more reliable calibration strategies for LIBs has been a common theme. The standoff ability of LIBS has also been put to good use in the nuclear industry and for the detection of explosives. Waste management of assorted materials, e.g. polymers, metals, etc. has also used LIBS analyses, often used in conjunction with chemometric data analysis tools.


1 Introduction

This is the latest review covering the topic of advances in the analysis of metals, chemicals and materials. It follows on from last year's review1 and is part of the Atomic Spectrometry Updates series.2–5

Several review papers are pertinent to many fields of industrial analysis and are therefore reviewed in this section rather than the specialised sections below. An example is the review by Khan et al. who reviewed the use of laser induced breakdown spectroscopy (LIBS) as a trace metal detection technique.6 The review contained 204 references and covers numerous topics. Included in these are ways to increase the sensitivity of LIBS analyses. Such methods include double pulse LIBS, nanoparticle enhanced LIBS, LIBS coupled with laser induced fluorescence (LIF), resonance enhanced LIBS, spark discharge-LIBS and magnetic field-assisted LIBS. All of these were discussed and examples from the literature given. Techniques coupled with LIBS were also discussed, including LIBS-Raman spectroscopy. A brief section covered many of the mathematical algorithms that have been used to treat the data. The advantages and disadvantages of the different forms of LIBS, i.e. single pulse, orthogonal double pulse, collinear double pulse, femtosecond LIBS, resonance enhanced LIBS and several others were also tabulated. The majority of the review was, however, focussed on different applications. Many of these were tabulated for easy reference and covered sample types such as metals, foods, plants, biomaterials, archaeological samples, explosives and waters.

Another review, entitled “Laser induced breakdown spectroscopy imaging for material and biomedical applications: recent advances and future perspectives” with 160 references was presented by Gardette et al.7 The “imaging” part of the title indicates that the review focusses on the 2-dimensional surface mapping of samples rather than bulk analysis. This paper also gave an overview of the requirements of the system (laser types, focussing elements, spectrometers, detectors, etc.), discussed the chemometric methods required for data processing and then gave examples of applications. The application sections included: materials analysis for manufacturing applications, oil industry and corresponding catalysts, nuclear materials, pharmaceutical industry, forensics and other geological and biological-based materials (plants, animal and human tissues), etc. A conclusion and future perspectives section was also supplied. The authors envisage the use of LIBS imaging to continue expanding its repertoire of applications by being able to detect halogens. Also envisaged was the further development of tools to aid quantification.

Reading through this update, the reader will note that many of the LIBS applications present methods of sample classification. This can be useful for recycling purposes, help forensic services distinguish between different explosive residues, different gunshot residues, etc. A critical review of recent trends in sample classification using LIBS was presented (with the aid of 244 references), by Brunnbauer et al.8 One of the problems associated with the use of LIBS for classification purposes is that the scientists who collect the data are often analytical chemists and therefore, are not completely conversant with many of the chemometric tools that are available to aid in the data analysis. Therefore, they are unaware of what is possible to do with them. This review gives a very good presentation of the whole data analysis process. Included in this was a schematic diagram of the process from data acquisition, use of training sets, data pre-treatment, variable selection, model construction, validation and the use of test sets of data. Each of the boxes on the schematic diagram was then discussed in the text, giving examples of the methodologies that can be used. For instance, in the data pre-treatment section, subjects such as background correction (spline interpolation, Lorentz fitting, etc.) were discussed. Similarly, different types of normalisation (normalisation to the total emission intensity, normalisation to an internal standard and normalisation to the standard normal variate) were also discussed with examples given. A small section on de-noising was also presented. Other sections discussed the classification methods linear discriminant analysis, partial least squares discriminant analysis, support vector machines, K nearest neighbour, random forest, artificial neural networks and soft independent modelling of class analogy (SIMCA). Each of these was discussed giving a description of how they work, their capabilities and some examples. Overall, this is a very good review for anybody moving onto the area of classification. Reading it enables an understanding of the processes, the processes of what happens rather than inputting data and then “something happens” and the samples are classified.

A brief overview of calibration-free-LIBS was presented by Zhang et al.9 Although the review is very much focussed on the mathematics behind the technique, examples of applications can be found throughout the text. These included cement, soils, tissues, tungsten materials and assorted alloys. This review cited 88 references and discussed the fundamental algorithm, modifications to it and self-absorption correction.

Cheung has produced a review of laser induced plume fluorescence that contained 70 references.10 The review covered the period 2005–2021 and comprised three main sections: the conceptual basis, the applications and the de-noising algorithms that may be used. The applications section is conveniently tabulated and includes examples for ceramics, metals and alloys, pigments and inks and aqueous colloids. The advantages of the technique were discussed including the very low detection limits (ng g−1 level) and the fact that both atomic and molecular analytes can be determined simultaneously. This latter advantage is dependent on the excitation wavelength being in the vacuum ultraviolet region of the spectrum and also requires that the analytes are in a plume of material such as that produced by laser ablation (LA). The review concluded with some future perspectives.

One final paper of interest that is potentially applicable to several different sample types was presented by Willner et al.11 It is well known that LA-ICP-MS can be problematic with regard to calibration since certified materials that are extremely closely matched in terms of matrix composition, hardness, density, boiling point, etc. are needed. Since these are often difficult to locate, the accuracy and precision can be compromised. Willner et al. developed a method that enabled materials to be prepared using a standard addition protocol. The final solution containing 0.5% polyethylene glycol, 5 μg g−1 Eu internal standard and the desired concentrations of analytes all in 30% isopropanol could then be sprayed onto the surface of the material to be analysed. The details of the spraying system and protocol were described in the paper but basically, the standard was sprayed through a mask so that some areas of sample were covered whereas others were not. The procedure was initially tested on a Kapton film (a polyimide). Homogeneity of the areas sprayed was tested using an acid digestion followed by ICP-MS analysis. For subsequent LA-ICP-MS analyses, excellent correlation between the observed signals and the deposited amount was observed, with R2 values of greater than 0.99 being achieved for Ag, In, Pb and Zn in the Kapton film. The Pb in the film was shown to be very inhomogeneous, with many areas having a content <LOD but other isolated spots having high concentrations. The method was then extended to silicon, silicon carbide, glass, copper and aluminium samples, demonstrating its applicability to numerous sample types.

2 Metals

2.1 Ferrous metals

As always, this has proved a popular area of research with several different analytical methods being represented in the assorted applications. However, one of the main techniques used has been LIBS where a large number of Chinese research groups continue to attempt to improve calibration strategies, reduce varying interference types and develop new applications. It was noted that there are several research groups that have developed methods using LIBS followed by assorted mathematical algorithms on the spectral data produced to determine the hardness of steels.

Since last year where there was a significant decrease in the number of papers on the topic of calibration strategies for LIBS-based analyses, there has been a revival during this review period. An example was presented by Peterson et al. who compared traditional univariate with machine learning calibration methods for the analysis of steel slag.12 They first used the univariate ratio method where the ratio of the analytes against one of the major components of the slag (in this case, Ca) is made. A graph of the data for Ti/Ca obtained using LIBS and the equivalent ratio obtained using XRF was plotted and an R2 value of 86.9% was obtained. The equivalent graph for Mg showed a correlation of 75.1%. After further calculations, the overall R2 value increased to 99.3%. They then moved on to an advanced multivariate analysis algorithm called Elastic Net, which allowed for several analyte lines to be used for the calibration functions. The authors explained how this algorithm works. Both methods produced data that were mass fraction ratios of analyte versus matrix. The actual mass fraction of each analyte was calculated so that the sum totalled 100%. The multivariate approach used 32 analytical lines to determine nine elements. As usual with this type of study, the large majority of samples were used to “train” the model and then the remaining samples used to test it. On this occasion, all but one sample was used to train the model and the remaining sample tested it. This was done in a loop, so every sample became the “test” sample in turn. The R2 value was 98.1%, which was similar to that obtained by the univariate mode. The only analyte for which the multivariate algorithm showed a significant improvement over the univariate one was for MgO. Precision was also similar for both methods.

Two papers by Belkov et al. discussed the use of partial least squares regression analysis combined with LIBS for the analysis of low alloy steels13 and both low alloy steels and cast irons.14 The papers had a slightly different focus, with one comparing three different methods of selecting the spectral variables13 and the other comparing the efficiency of various methods of spectrum processing.14 In the latter case, normalisation to the baseline, localisation of the spectral range and the addition of non-linear components of the spectrum allowed the accuracy of the regression model to be improved. The standard deviation of the results were improved by a factors of between 1.8 (for V) and 6.8 (for Si).

Another paper reporting analysis of low alloy steels, using LIBS in conjunction with a chemometric tool, was presented by Shen et al.15 These authors studied the effects of spectrometer resolution by simultaneously measuring the signals from the samples at different resolutions, using a dual channel spectrometer. This led to two different sensitivities being produced from the same sample so the Boltzmann plot method was suitable for calculating plasma temperatures for the low resolution (0.076 nm per pixel) spectra. However, because of instrumental broadening, Stark broadening could not be used to calculate electron number density. The high resolution spectra (0.01 nm per pixel) did not suffer from this problem. For these spectra the two-line method (intensity of Fe at 395.668 divided by intensity of Fe at 400.524 nm) was used to calculate plasma temperature. For the quantitative analysis of the samples for the Mn content, a support vector machine (SVM) was used to aid calibration. The results from the low resolution spectra were poor compared with those obtained using high resolution.

Calibration-free LIBS and calibration curve LIBS were compared with LA-TOF-MS and energy dispersive spectrometry for the analysis off high energy alloys (CoFeNiMo-based materials).16 Several different compounds were made in-house and the methodology used for this was described in detail. The LIBS experiments were undertaken using a Nd:YAG laser operating at a 532 nm, with an energy of 100 mJ per pulse. Detection over the wavelength range 250–870 nm was achieved using four different spectrometers, each with its own limited range capability. The calibration curve LIBS technique relies on the spectral line intensities and the concentrations of the elements present in the sample. The intensity of a line in a spectrum is proportional to the number density of the element, which implies that both emission line intensities and the element concentrations are proportional (IC). The signal of each analyte was normalised to the signal of one of the components of the material with higher concentration; in this case Mo at 379.83 nm. This was to eliminate some spectral errors. Calibration curves formed in this way had good linearity, with R2 values ranging from 0.974 to 0.998 providing LOD of between 2.92 mg kg−1 (for Mo) and 9.18 mg kg−1 (for Co). The calibration-free LIBS data were collected in the usual way, i.e. through assumptions that the plasma is in local thermal equilibrium (LTE), is optically thin, has spatial and temporal homogeneity, etc. and then using the well-isolated Stark broadened emission lines (free from self-absorption) of Mo to calculate the electron number density. The plasma temperatures were determined using the Boltzmann plot and Saha–Boltzmann plot. The results of the two techniques were similar and agreed well with the “expected” values for the materials made. Results were also comparable to those obtained using LA-TOF-MS and energy dispersive spectrometry.

Jun et al. used LIBS to obtain spectra from high-speed train wheel steel and then subjected the data to PCA for preliminary analysis.17 Using the three spectral correction methods of median filtering, baseline correction and multiple scattering correction as further pretreatment, a support vector machine (SVM) was used to provide a quantitative model. Overall, the model established using the pre-processed data of the multiple scattering correction proved to be the best. Providing an accuracy of prediction rate of 98.4%. It was concluded that the approach could be useful for discriminating between different metallographic structures of train wheel steel.

Curiously, several other groups have also published papers on the analysis of high-speed train wheel parts. An example was by Sheng et al. who analysed axle billets.18 Instead of LIBS, these authors developed a method they entitled “spark mapping analysis for large samples” (SMALS). The paper showed and described the instrumental setup which enabled samples of dimensions 1000 mm × 500 mm to be analysed. The methodology was also discussed in detail. The atomic emission spectrometer operating conditions were: spark frequency 500 Hz, discharge voltage 290 V and argon flow 8 L min−1. The instrument comprised a tungsten electrode with 90° angle and diameter 6 mm, a Paschen–Runge optical system with a PMT detector and a diameter of 500 mm, covering the wavelength range of 160–650 nm. The concave grating had 2700 lines per mm giving a spectral resolution of 0.74 nm mm−1. The spark crater had a diameter of 4 mm and a depth of approximately 1 mm. Calibration was achieved through the analysis of 32 low alloy steel CRMs enabling quantitation of nine minor components of the material (Al, C, Cr, Cu, Mo, Mn, S, Si and V). Every calibration had an R2 value of greater than 0.99 and provided LOD of between 0.0002 and 0.0066%, with S having the best LOD and Si the worst. The methodology was compared with μ-XRF and the data from both were comparable, indicating the accuracy of the methodology was adequate. The technique developed had the advantage of speed over micro-XRF analysis with over four millions spark data points collected and processed in five hours compared with approximately 1.8 million points collected in approximately 60 hours. The method was applied to the analysis of the train wheel axles so that elemental behaviour in them could be determined and any inclusions found could be analysed. Different elements had very different behaviours. The Al and S segregated to the central section whereas Mn, Mo and V tended to segregate towards the edge. The main inclusions were aluminium oxide, although some of manganese sulfide also existed. Both were found towards the central section.

Two reports from the same research group discussed the determination of the hardness of train wheel steels using LIBS.19,20 Conventional methods (e.g. the Vickers hardness test) use the depth of an indentation as a measure of hardness. However, this damages the sample and so the authors used the minimally damaging technique of LIBS for the task. The LIBS spectra from 12 standard steel samples were obtained and then numerous methods of pre-processing were compared. These correction procedures were: smoothing, multiplicative scattering correction, asymmetric least squares, adaptive iteratively reweighted penalized least squares, baseline estimation and de-noising using sparsity. To aid the reader, a brief summary of each of these was given in the paper. The model using multiplicative scattering correction combined with baseline estimation and de-noising using sparsity offered the best performance, providing a cross validation of correlation coefficient of 0.9909 and a standard deviation of cross calibration of 14.6935. The LIBS data contained an enormous amount of data that was either redundant or just provided background information. The authors eliminated these redundant data to simplify and hence speed up the calculations and to improve the model's performance. Several methods were tested: the uninformative variable elimination, competitive adaptive reweighted sampling and successive projections algorithm. As well as simplifying the data these variable limitation algorithms also created the partial least squares (PLS) and partial minimum support vector machine models. When all steps were taken into account, the competitive adaptive reweighted sampling-least squares support vector machine provided the best results with the standard deviation of cross calibration decreasing to 9.480. Overall, it was concluded that the approach of using LIBS in combination with the chemometrics offered a much more rapid, efficient and non-destructive method of hardness determination compared with the standard method.

A paper by Li et al. also determined the hardness of steels.21 In this example, the LIBS spectral data obtained from six GCr15 steels that had been tempered at different temperatures were analysed and a correlation between spectral line intensity ratio and hardness established. Principal component analysis reduced the number of data points to be analysed before a random forest model was developed. Initial results were disappointing, with the prediction accuracy for the hardness of only two samples being greater than 50% (69% and 52%, respectively) while many of the prediction accuracies were almost zero. The authors then changed their approach and instead of using PCA to reduce the dimensions of the original data, they used random forest only, to treat the data. At first they attempted to use the whole spectrum, but, in common with the previous papers, use of the whole spectrum led to significant amounts of redundant information being used to train the model, leading to decreased accuracy. They therefore allowed the random forest algorithm to choose its own variables which it did based on their importance. Once the final model had been developed, accuracy improved significantly with an average prediction rate being 96% over all of the samples. The speed of analysis and data processing (once the model had been formed) as well as the non-destructive nature of the analysis was again highlighted as advantages over the current methodology.

One of the main drawbacks of LIBS is acknowledged to be self-absorption effects from analytes that are present at high concentration. This problem has been addressed by Tang et al. who studied ways to minimise the Mn self-absorption effects during analysis of steels.22 To achieve this, they used a setup that enabled both LIBS and laser induced fluorescence (LIF) measurements to be made. This comprised a Q-switched Nd:YAG laser operating at 532 nm, with a pulse duration of 7 ns, at a frequency of 10 Hz and an optimised energy of 4 mJ per pulse. A second Q-switched Nd:YAG laser that was tuneable using an optical parametric oscillator operating with a duration of 10 ns and with a frequency of 10 Hz was used to illuminate the plasma formed by the first. Light was detected using a spectrometer equipped with an intensified CCD detector. The gate width of the intensified CCD was different for LIBS (1 μs) and LIBS-LIF (10 ns) measurements. The Mn lines at 403.08, 403.31 and 403.45 nm were used for detection whereas the primary line at 279.48 nm (set by the optical parametric oscillator) was used for excitation for the LIF measurements. Using these parameters and analysing seven micro-alloyed steel samples, the improvement in signal to noise ratio obtained using LIF was significant. This, however, did not give an indication of the level self-absorption so they then used the exponential calibration curve method to determine this. For LIBS-LIF a delay between laser pulses of 5 μs led to increased sensitivity compared with a delay of 1 μs. This was the opposite trend to LIBS alone. The R2 of the linear fitting calibration curve, self-absorption factor α, root mean square error of cross-validation (RMSECV), average relative error (ARE), and average relative standard deviation (ARSD) were all calculated. In general, the smaller the value of α is, the weaker the self-absorption and the lower the RMSECV and ARE values are, the greater the accuracy in quantitative estimation. For the three Mn wavelengths measured, α decreased by 88–90%, RMSECV decreased by 85–88% and ARE values were reduced by 90–93%. Average ARSD reduced by 29–33%. Overall, it was concluded that the use of LIBS-LIF can extend the range of the calibration to much higher concentrations without experiencing self-absorption effects.

The classification of steel samples is another topic that has received significant interest. Classification may be necessary for recycling purposes or any other reason that requires quick identification. Again, many of the applications described have used LIBS for the analysis. An example was presented by Bai et al. who used LIBS to analyse four grades of stainless steel (201, 304, 316 and 430) and then the random forest algorithm for the classification.23 The paper is in Chinese and so limited information can be gleaned. However, four analytes (Cr, Fe, Mn and Mo) were chosen and a total of 12 analytical lines monitored. The average recognition accuracy based on 100 classification experiments on 300 groups of data was 98.28%, with a processing time of 0.418 s. The processing time decreased to 0.019 s when a single group of data (one sample) is to be analysed.

Other examples of classification exist and they are all variations on the same theme. One such example was by Traparic and Ivkovic, who used LIBS and a machine learning algorithm to classify austenitic steel alloys.24 Again, data pre-processing was performed prior to the formation of a model. In this case, data pre-processing occurred using isolation forest algorithm analysis on MinMax scaled LIBS spectra over the wavelength range 200–500 nm. The Gini importance criterion was used to decrease the amount of data used to form the model and optimal model parameters were found by using the grid search cross-validation algorithm. A final random forest regression training of the model was then undertaken. Results from the model were compared with those obtained using linear regression with L2 norm and deep neural network. Performance metrics in terms of R2 and root mean square error indicated that the deep neural network was the best overall. However, this was because the random forest method developed had very good prediction results for Cr, Mn and Ni, but performed poorly for Mo.

Another application in this area was presented by Zeng et al.25 These authors concentrated on the removal of redundant information from the spectra to simplify the training of models and to improve their accuracy. They used both PCA and restricted Boltzmann machines for the task of data dimension reduction. A support vector machine was then used on the reduced data from both methods. It was concluded that the restricted Boltzmann machine–SVM combination was the better of the two because it produced an accuracy of 100% in only 33.18 s. This time was almost half of that required by the PCA–SVM method.

A further example was presented by Yang et al. who determined Mn and Ni in steel samples based on LIBS data, followed by a genetic algorithm–partial least squares combination.26 The experiment used 12 steel samples, 9 of which were used as calibrants to “train” the model and the remaining three to test it. For Mn an R2 value of 0.999 and root mean square error of prediction (RMSEP) of 1.347 were obtained. The corresponding values for Ni were 0.999 and 0.5254. The values for both analytes were better than for partial least squares alone.

Many of the papers discussed so far have treated the data to simplify calculations. A paper by Hu et al. reported methodology to improve accuracy by removing the continuum background signal.27 A “moving average corner cutting” method was described in full. Results from LIBS analyses where no model was used, moving average corner cutting and from corner cutting alone were compared. At first, they tested the models on simulated spectra where lines from the NIST database were used to construct a spectrum. The method was then applied to steel samples that are known to produce a large background signal. The R2 values of calibration for several analyte lines were all >0.99 and were the highest for moving average corner cutting compared with model-free and corner cutting alone. It was then also applied to the analysis of a titanium alloy using calibration-free LIBS. The mean relative error using the moving average-corner cutting algorithm was 2.21% which was better than that for model free correction (2.36%) and for corner cutting alone (2.32%). It was concluded that the model developed had the highest robustness and provided better accuracy for both calibration and calibration-free LIBS.

Two papers have reported the LIBS analysis of welds.28,29 In the former paper the on-line monitoring and defect detection of arc welding (tungsten–inert gas welding) was performed both on- and off-line. In arc welding defects can form if there is a temporary variation in the welding current, humidity level or if there is a lack of cleanliness. The ability to monitor the weld on-line as it occurs would therefore be advantageous. Similarly, the ability to measure completed welds would enable 2-dimensional maps to be constructed enabling defective ones to be spotted and rectified as well as identifying the composition of the seams. A schematic setup of the process was presented and this shows the presence of two spectrometers: one to detect light emitted as the weld is formed and one to detect light produced by the LIBS signal. Fibre optics transmitted the light from the welds to the spectrometers. The sample was placed on an XYZ stage enabling it to be moved. A total of 50 welding plasma spectra per second were recorded, giving rise to a spatial resolution of 0.12 mm (process speed = 6 mm s−1) for the conventional spectroscopic setup, while the LIBS setup acquired 10 spectra per s providing a resolution of 0.6 mm. Results were obtained from two sample types: EN 1.4301 stainless steel and AL 5183 (an Al-based alloy). Detection of Li and Na using the LIBS system is indicative of contamination, since they should not be present in the samples. This was demonstrated by analysing the metal plates before and after the application of some grease. The study was very successful and clearly enabled the monitoring of the welds both as they happen and afterwards. The authors envisaged future work would include the use of an algorithm to try and quantify the elements and to construct 3-dimensional maps of the welds rather than the 2-dimensional ones observed here.

The other paper, by Quackatz et al. analysed tungsten inert gas welds made on Duplex stainless steels using LIBS, XRF and EDS.29 The instrumental techniques employed enabled line scans across a weld made between EN 1.4162 and EN 1.4462 stainless steels to be made. The LIBS and EDS scans occurred 10 mm apart to ensure that they did not influence each other, it was stressed that the hardness of the steel determines the penetration depth of the laser during LIBS analyses. This is typically at the nm range. When one steel type is used, this is normally not problematic, but can be if two steels of different hardness are analysed. The penetration depth of electrons during EDS measurements is ∼1.5 μm, i.e. a much greater depth than for LIBS. Despite this, the elemental profiles were very similar and the LIBS had the advantage of being much faster. The XRF mapping indicated the presence of accumulations of Mn at individual locations. These accumulations were not visible during the LIBS analysis and this was attributed to the spatial resolution being much lower. Overall, it was concluded that LIBS was well suited for the detection of sub-surface elements because of the superficial ablation of the material and that it can be performed without pre-treatment of the samples, as well as in an ambient atmosphere. This makes LIBS very interesting for quality control.

Another LIBS application was presented by Cui et al. who have continued with their work using long–short double pulse LIBS.30 Double pulse LIBS is known to have advantages over single pulse. Many workers use two separate lasers to first vaporise the sample and the second to excite the vapour further. This usually leads to improved sensitivity. The authors use a setup (schematic given) that uses only one laser, that fires first for 60 μs to vaporise the sample and then again for 6–7 ns. The setup was applied to the determination of C in five ultra-low carbon-containing (9–89 μg g−1) standard steel samples. The authors first attempted the analysis in single pulse LIBS mode. However, because of surface contamination arising from grease, carbon dioxide in the air, dust, materials used for cutting, e.g. diamond or tungsten carbide, etc., this proved impossible. When using a SEM-EDS instrument, micron-sized black dots could be seen on the surface of some of these samples. Analysis of them indicated very appreciable concentrations of both Si and C (both >40%), whereas the Fe content was 9–11.5% when it should be 97 or 98% in the bulk sample. The use of the long-short double pulse version of LIBS, especially when pre-treatment pulses are also used enabled the sample surface to be cleaned and then analysed and proved to be mainly successful. The results were improved, but were in only reasonable agreement with certified values, e.g. 71.76 μg g−1 rather than 61 μg g−1. The LOD was 22.6 μg g−1. Precision was also not magnificent, possibly because the values measured were quite close to the LOD, with typical values being between 13.9 and 58.3% RSD.

Grunberger et al. presented an interesting paper in which the matrix effects observed during the analysis of steel using laser ablation-spark discharge (SD)-OES was compared with those from LIBS analyses.31 Thirty-six samples were analysed, all of which had an Fe content of >94%. The laser energy used for the LIBS was 55 mJ, which was the same as the total for the LA-SD-OES (5 mJ for the LA and 50 mJ for the spark discharge). For the LA-SD-OES, the craters formed were the same size for all samples and the Mn concentrations formed a calibration curve with R2 of over 0.99. The same was not true for LIBS which demonstrated a distinct dependence of plasma emission on crater size. The crater size appeared to be dependent on the Si content of the steel, with low Si steels having a LIBS crater volume of 0.001 mm3 rising to 0.003 mm3 for higher Si samples. This had a significant effect on the sensitivity of the Mn calibration, i.e. there was a matrix effect. The hardness and the grain size of the steel also had an effect. The different effect of the two techniques on the steel, despite the same energy being used, was attributed to different processes of sampling and ionisation.

The related technique of laser ablation-spark induced breakdown spectrometry (LA-SIBS) was used by He et al. to determine Al, Cr, Cu, Mn. Mo and V in steel alloys.32 The femtosecond laser was operated at a 1 kHz pulse repetition rate and at a wavelength of 800 nm with an energy of 7.5 mJ. Once the laser had vaporised some of the sample a spark was used to excite it further in a similar fashion to double pulse LIBS. The instrumental setup was described and a schematic diagram provided. Calibration curves for Al (394.40 nm), Cr (425.43 nm), Cu (324.75 nm), Mn (403.08 nm), Mo (379.82 nm), Ni (352.45 nm) and V (427.92 nm), were established through the analysis of a series of steel reference materials. These materials were: GBW(E)01249a, 010381a, 010383a, 010384a, 010459a, 010426, 010460, 010495, 010499, and 010503. Numerous parameters were optimised including: spark discharge voltage (3.0 kV), discharge capacitance (7 nF), laser pulse energy (1.5 mJ), The integration time on the CCD detector was 2 ms and the number of the number of repetitions per replicate was 500. Under these optimal conditions, the LOD obtained were: 8.7, 4.0, 3.1, 5.7, 12.6, 7.9 and 10.9 mg kg−1 for Al, Cr, Cu, Mn, Mo, Ni and V, respectively, representing improvements in sensitivity by factors of between 4 and 12 over LIBS under the same conditions.

A paper presented by Sanyal and Dhara described the direct, non-destructive determination of trace and major analytes in steel samples using micro-focussed bremsstrahlung radiation in XRF.33 Full bremsstrahlung excitation was used in combination with micro-focussed geometry and the detection limits of several analytes were reduced, enabling those present at 30–80 mg kg−1 to be determined directly. This made it possible for analytes such as Co, P, S and Si to be determined simultaneously with those elements present at percentage levels. The methodology was reportedly very simple, and yielded results that were precise and with good accuracy.

2.2 Non-ferrous metals and alloys

Many of the papers in this section are similar in nature to those in the Ferrous metals section. In terms of technique, LIBS is the most popular, with numerous authors attempting to improve calibration, not use calibration at all or to remove assorted interferences. Others describe reduction from thousands of data points to a much more manageable number using PCA or some other chemometric tool and then insert the trimmed data into another chemometric tool for classification of the materials. Those papers that describe classification, e.g. for recycling purposes or to detect counterfeits, employed a large variety of chemometric tools, e.g. random forest, K nearest neighbour, partial least squares discriminant analysis, etc. There are though, several other methods used for analysing non-ferrous metals and alloys and these are discussed in the sub-sections below.
2.2.1 Copper and copper-based alloys. Several papers from the same research group have sought to improve the signal stability/repeatability or to decrease matrix effects as these are regarded as being the main features hindering the widespread use of LIBS. Both of these problems were addressed in a paper by Long et al. who developed a method entitled data selection method based on plasma temperature matching.34 Matrix effects and signal uncertainty affect the LIBS signal by changing parameters such as plasma temperature and total electron number density at the signal-collecting temporal spatial window. The authors showed that if the temperature of the plasma varied by as little as 5%, the signal intensity varied between 36.7 and 77.5% whereas the electron number density had a much smaller effect, with a 5% variation leading to a signal variation of between 0.4 and 4.8%. The authors therefore deduced that if the temperature of different plasmas is kept constant, then signal intensity for the same concentration would remain constant, irrespective of the sample type, i.e. matrix effects would be minimised. The method developed therefore, picked a series of spectra from different plasmas based on their temperature to build a model. A series of 11 brass samples with Cu comprising between 56.62 and 95.9% and Zn between 4.02 and 41.76% were analysed. The Cu lines were used to calculate the temperature of the plasma and the Zn lines at 472.52 nm and 481.05 nm used to build the model. This resulted in a plasma temperature fluctuation between sample types of 90 K rather than 3500 K. Both univariate and multivariate (multiple linear regression) methods with and without the plasma temperature matching were used, with huge improvements noticeable in both. For the univariate approach, the root mean square error of prediction (RMSEP) improved from 3.30% to 1.06%, the determination coefficients (R2) improved from 0.864 to 0.986 and the precision improved from 18.8% to 13.5%. The multivariate approach offered very similar improvements, with RMSEP going from 3.22% to 1.07%, R2 improving from 0.871 to 0.986 and RSD improving from 26.2% to 17.4%.

In an attempt to improve signal repeatability, another paper by this group reported the building of a model to compensate for fluctuations in the total electron number density.35 This was another data pre-processing method that utilised significant amounts of mathematics. Fortunately, the authors take the reader through this. The model was tested on 29 brass samples and reportedly out-performed other normalisation methods (total area normalisation and segmented area normalisation). For the determination of Cu, the mean pulse to pulse signal precision improved from 5.1% to 1.03%; i.e. it was comparable to that obtained using ICP-OES. The RMSEP values for both Cu and Zn decreased significantly. It was concluded that the methodology developed would be useful for improving the repeatability of LIBS measurements.

The same research group has also investigated the effects of laser beam intensity distribution on signal stability.36 The experimental setup for measuring the LIBS signal and beam intensity simultaneously was described in detail. The methodology was tested on a copper and a silicon sample with the spectra and laser beam intensity distributions being collected for a period of 30 days. Day to day variations in laser beam intensity distribution were observed which led to signal intensity changing and, consequently, the signal repeatability being degraded. However, by inputting the LIBS data and beam intensity distribution into partial least squares regression, a model was constructed to modify the spectral intensity appropriately. In this way, the long term stability improved significantly, with the long term precision for the copper improving from 13.5% to 4% and from 10.7% to 6.5% for the silicon.

Another paper that has attempted to improve signal stability was presented by Li et al. whose approach was to employ spectral intensity as a means of focussing the laser onto the surface of the sample.37 Precise laser focussing is an important part of obtaining a stable signal and so the analysis of rough, uneven or curved surfaces can be problematic. Some instruments have positioning devices to try and assist with that. The method developed was hoped to be a solution for this problem. The instrumental setup was described in detail and a schematic diagram provided. The light emission from air was used for quantification. The measurement of Cu at 324.7 nm and O at 777.4 nm at the position where the O signal was maximum enabled an accurate and repeatable positioning of the laser above the sample. This was demonstrated by undertaking an experiment lasting 120 days. It was noted that other air species, e.g. nitrogen or hydrogen, could also have been used for the positioning. It was stated that the method could be used for field studies because only a relatively low energy laser is required and because there is no need for other positioning devices.

An interesting paper by Mateo and Nicolas assessed the mapping capability of linear correlation statistics for the LIBS characterisation of brass samples.38 The study focussed on the characterisation of materials that had been “weathered”, i.e. a corrosion layer had been allowed to form on areas of the surface of the samples. This was achieved by using four brass samples and then exposing half of each one to a mixture of acetic acid and hydrogen peroxide for different time periods. The LIBS setup involved a Nd:YAG laser operating at 532 nm with detection being achieved using an echelle spectrometer equipped with an intensified CCD detector. Correlation depth profiles (to study the depth of the corrosion), 2D correlation maps (to identify the area exposed to the chemical attack) and 3D correlation maps (obtained by combining the previous two types) were produced and compared with those generated using conventional LIBS. It is this 3D mapping that produced additional data to conventional LIBS. The linear correlation enabled layers at depth to be analysed and assessed as having been affected by the attack even though, using visual inspection, they may not appear to have been.

Different types of LIBS have also been used for the analysis of copper-based materials. Although calibration-free-LIBS is an extremely useful method, it requires significant amounts of knowledge of the sample/analytes prior to analysis (for instance, the transition probabilities of the analytical wavelengths used) and makes several assumptions, e.g. that the plasma is in local thermal equilibrium (LTE) and that it is optically thin. Iqbal et al. have used calibration-free LIBS with a commercial spectrometer to measure the composition of Devarda's alloy (an AlZnCu material).39 The intensity line profiles of these elements were used to calculate plasma diagnostics, e.g. electron number density (obtained from the Stark broadening of individual emission lines of the analytes) and excitation temperature (obtained using the Boltzmann plot technique). The excitation temperatures calculated using Cu and Zn were 8547 ± 5% K and 8100 ± 5%, respectively. The relative proportions of the material were calculated to be 57% Zn, 39% Al and 4% Cu. This was compared with the certified concentrations of 50% Zn, 45% Al and 5% Cu. Despite there being fairly significant differences between experimental and certified values, the authors concluded that agreement was “reasonable”.

A variant of LIBS called one point calibration LIBS was used by Li et al. for the analysis of copper–zinc–silver–gold alloys.40 This modification of calibration-free LIBS enables analytical lines to be used for which transition probabilities are not known. The paper described the mathematical basis of the method in detail. Experimentally, three pieces of sample were obtained from different manufacturers and then analysed five times each and the results averaged. Sample 1 was used as the calibration standard and the other two were the test samples. The atom lines for Zn at 468.014 nm, 472.216 nm and 481.053 nm were used for quantification. Excitation temperature and electron number density still had to be calculated for the calibration to be achieved and the authors explained how these were achieved. Results from the analysis of the alloy samples were compared with those obtained using SEM-EDS and were in good agreement, with the worst deviation being only 3.24%. The authors acknowledged that the plasma temperature and F factor had to be determined based on other species having emission lines with known transition probabilities. If a sample required analysis for which no components had known transition probabilities, then the technique could not be used. The likelihood of this occurring is very small though.

Resonance enhanced LIBS is another version of LIBS that offers improved sensitivity and minimal damage to the sample. Instead of using a Nd:YAG laser at its primary wavelength (1064 nm) or frequency doubled (532 nm), tripled, quadrupled, etc., a tuneable laser was used to ablate and excite the sample at a wavelength characteristic of one of the analytes required. An example was presented by Ma et al. who determined Pb in copper alloys.41 Resonance LIBS also exists in different versions: RLIBS-ground state atom excitation (RLIBS-G) where a main wavelength of atomic Pb is chosen (283.31 nm) and RLIBS-excited state atom excitation (RLIBS-E) (363.96 nm), In both cases, the Pb signal emitted from the sample was measured at 405.78 nm. The RLIBS-G was found to offer better performance than both the RLIBS-E and ordinary LIBS, with LOD of 0.001% rather than 0.020% for LIBS and RMSECV of 0.005% compared with 0.024%. The R2 was very marginally decreased though (0.98 compared with 0.99). Other versions of RLIBS were also tested. For the determination of In, RLIBS with normal stepwise line-atom excitation (RLIBS-NSL) was compared with RLIBS-G. Here, excitation for RLIBS-G was at 410.17 nm and for RLIBS-NSL it was 303.93 nm. In both cases, the emission signal for In was monitored at 451.13 nm. The RLIBS-G again offered better sensitivity. A final version entitled RLIBS-thermally assisted stepwise line atom excitation (RLIBS-TASL) was tested and compared with RLIBS-G for Si determination. Excitation wavelengths were 243.88 nm and 251.43 nm for RLIBS-G and RLIBS-TASL, respectively and the wavelength of detection was 288.16 nm for both. Once again, the RLIBS-G offered more intense spectral data. The reason why RLIBS-G was more sensitive than the other versions was postulated to be mainly because there are more electrons in the ground state, and consequently, more electrons are resonantly excited to the desired excited state.

2.2.2 Aluminium and aluminium-based alloys. The analysis of aluminium and aluminium-based alloys has offered similar studies to those for the copper-based materials. Much of the work concentrated on ways of improving sensitivity, de-noising the signals to improve accuracy and the use of chemometric methods to classify different alloys.

An example of the papers that have used de-noising methodology was presented by Xu et al.42 These workers used a fibre-based LIBS system and a novel wavelet threshold functions method of de-noising to aid the determination of Cr, Mg and Mn in aluminium alloys. Fibre-lasers have the advantage of producing very stable power over long-time applications, but because they have a very high repetition frequency, they produce much more noise from bremsstrahlung and multi-ablation. The overall effect is a reduction in accuracy. These authors developed eight methodologies on simulated spectra and then tested the more successful ones on aluminium alloys. The eight methodologies were all discussed in the text. The overall result was that based on signal to noise ratio, root mean square error (RMSE), smoothness, peak error and full width half maximum (FWHM) error, the accuracy of analysis was greatly improved using two of the methods. These were entitled modified threshold function (MTF) and double-exp. The mathematics behind the calculations were all discussed in the paper. A related paper by the same group also used background removal methodology based on modified iterative wavelets for fibre-based LIBS.43 In this paper, the methodology developed was compared to two conventional background removal methods and was found to provide the most accurate data. A third paper, again by the same group, discussed the use of fibre-laser LIBS for the determination of Cu, Mg and Mn in aluminium alloys.44

Random noise and interferences on LIBS spectra degrade the quality of the analytical data and so some method of removing them can be very beneficial. A paper by Yao et al. reported the use of spectral filtering to achieve an improvement in accuracy during the femtosecond LIBS analysis of aluminium alloys.45 Two algorithms were tested: the median filter algorithm (a nonlinear signal processing technology based on the statistical theory of ranking) and the Savitzky–Golay filter algorithm. The abilities of both were described in the paper. Operating conditions in terms of laser pulse energy (1.5 mJ) and platform movement speed were optimised (5 mm s−1). The latter had to be optimised to prevent the signal from decreasing caused by repeated ablation of the same position. This becomes problematic when the laser repetition rate is 1 kHz. The real optimum for the movement speed may be higher, but at speeds greater than 5 mm s−1 the motor began to overheat. The number of laser shots needing to be averaged before an acceptable precision was achieved was also determined. Precision improved rapidly up to 300 firings and then the improvement rate slowed significantly. A value of 500 was chosen for the experiments. Six Chinese standard reference materials were analysed under optimal conditions and the analytes Cr (428.97 nm), Cu (324.75 nm), Mg (285.21 nm) and Mn (403.44 nm) were determined. The filtering algorithms were applied to the spectra prior to quantification. Both algorithms decreased the sensitivity for all four analytes but improved linearity (R2) of the calibrations. Despite the sensitivity decreasing, the achievable LOD improved by factors of between 1.4 and 5.2 for median filter correction and between 1.2 and 2.5 for the Savitzky–Golay corrected data.

The same research group used a Ti:sapphire femtosecond laser (7.5 mJ, 35 femtosecond pulse width and a repetition rate of 1 kHz) to analyse four standard aluminium alloys for their Cr, Cu, Mg, Mn and Zn content.46 They used the one-point, multi-line calibration strategy and the technique of laser ablation spark induced breakdown spectroscopy for quantitation. This is where the laser vaporises and partially excites the sample and then a spark excites the plume further. The analytical setup was described in detail and a schematic diagram given. The spectrometer used had three channels capable of measuring the regions 200–317 nm, 317–415 nm and 425–500 nm simultaneously. Multiple lines of each of the analytes were used and several assumptions made; i.e. that the plasma was in LTE, it was optically thin, it was spatiotemporal homogeneous, etc. The theory and mathematics behind the multiple lines one point calibration were presented in the paper. One of the four standard samples was used for the calibration. A plot of certified value against experimentally obtained data yielded a graph with a correlation of 0.9981. The outcome was a method offering rapidity, simplicity and an enhancement in sensitivity.

A third paper by the same research group used calibration-free femtosecond LA-SIBS for the analysis of aluminium alloys.47 The same setup as in the previous paper was utilised and five standard aluminium alloys analysed. The sample moving speed was optimised for the same reason given previously. Data were imported to Matlab and a series of programs used to facilitate the removal of noise (median filter method) and the electron number density calculated for all five samples. The average electron number density from the five samples was 1.5 ± 0.3 × 1017 cm−3 and the plasma temperature was 6300 ± 300 K. Analysis of the standard materials yielded data in reasonable agreement with certified values with the average error for Al being <0.3%. Detection limits were lower than 0.1%. It was concluded that the method was rapid, sensitive and could be used to determine major, minor and trace analytes in multi-component alloys.

One of the limitations of LIBS is that it has relatively poor sensitivity. A number of approaches have been tested to try and improve this and one, by Awan et al., used a coating of either copper, gold or magnesium nanoparticles on the surface of the aluminium.48 The emission sensitivity of both Al and Na in the presence and absence of the nanoparticles was compared and the nanoparticles certainly led to an enhancement. The electron temperature was calculated using a Boltzmann plot and the presence of the nanoparticles led to an increase compared to the untreated aluminium surface. Conversely, the electron number density seemed relatively unaffected. The plasmas were tested using McWhirter's criterion and was deemed to be in LTE. It was concluded that nanoparticle enhancement of LIBS emission could be used to improve sensitivity and enable the determination of some analytes that would otherwise have been below the LOD.

Another paper by this research group used a machine learning-based calibration procedure for the LIBS analysis of aluminium-based alloys.49 Eight samples of known concentrations of Cu, Pb, Sn, Si and Zn were obtained and the line intensities measured and normalised against the aluminium matrix element in an attempt to minimise uncertainties. These intensity ratios of analyte against the matrix element were then used to train an artificial neural network (ANN). The results of the ANN analysis of samples were compared with those obtained using calibration curve LIBS and with the certified values. Those data obtained using the ANN were in excellent agreement with the certified values. However, the data from calibration curve LIBS were less impressive. The authors claimed the methodology could be used not just for alloys, but also for soils and other multi-element samples.

One paper discussed the classification of scrap aluminium alloy samples. Diaz-Romero et al. used LIBS followed by a machine learning approach and/or a deep learning approach.50 There were two LIBS setups: a gated lab-based system and the other an ungated industrial-based system. Both were discussed in the paper with the aid of schematic diagrams. The industrial setup was particularly interesting. It comprised a feeder belt running with a constant speed, a colour camera and a 3D camera that collect information about the sample object and the LIBS system which is synchronised with the 3D camera to apply a pulse while the piece falls from the feeder. A schematic diagram of the entire process, including the mathematical processing, was provided to aid the reader. This was particularly helpful because the process was complicated and comprised numerous steps. As a brief summary both systems had pre-processing steps, e.g. de-noising, baseline correction, feature extraction applied to the raw data. Once all the pre-processing steps are performed 145 features were picked and three different machine learning models (linear regression, support vector classifier and random forest) were trained and validated using repeated cross-validation. The most robust pre-trained model was then used to classify scrap Al from samples. These classified samples then underwent melting tests and spark analysis to prove that the sorting experiment had classified the samples correctly. In the final step, an end-to-end deep learning system was built to enable the classification in real-time at higher throughputs for which knowledge acquired in the previous steps were used. The best machine learning approach was random forest which processed one spectrum in 150 ms and provided a precision of 0.8 (where precision = true positives divided by the sum of true and false positives), a recall of 0.81 (where recall = true positives divided by the sum of true positives and false negatives) and an F1 score (where F1 = precision × recall divided by the sum of precision + recall, all multiplied by 2) of 0.80. The best performing deep learning approach was back propagation neural networks (BPNN)-Ghostnet (a type of convolutional neural network) which provided very similar metrics to the random forest, but did many more operations in a shorter time (200 spectra in 9 ms).

A paper presented by Ding et al. described the determination of Mg in 17 aluminium alloy samples.51 This was another example where LIBS spectra were obtained and then partial least squares (PLS) model established and its prediction ability assessed using random forest. The correlation coefficient and RMSE of the PLS model on its own were 0.6809 and 1.2042, respectively. When random forest was then applied these values improved to 0.8571 and 1.0918, respectively. The relatively poor performance was attributed to too much data being input, some of which was mis-leading or redundant. The variables were then screened and ranked in order of importance, with those variables that had an importance greater than 0.11 being selected. The random forest metrics then improved to a correlation coefficient of 0.9461 and RMSE of 0.9534. In addition to improving the performance, selecting the relevant variables only also led to a 91.67% improvement in the computational time.

2.2.3 Other metals and alloys. Several other metals and alloys have had applications reported in this review period. The analysis of titanium alloys has been described in two publications. A one point calibration LIBS application in which self-absorption was corrected was described by Hu et al.52 According to the authors, one point calibration LIBS corrects the Boltzmann plot of the unknown sample by using one known sample and obtains higher quantitative accuracy than calibration free-LIBS. Self-absorption still occurs though and this limits the overall accuracy. To correct for this the authors developed an algorithm that did not require the calculation of the self-absorption coefficient. The paper described the algorithm and then tested it by determining Al, Ti and V in two titanium alloys. It then compared the results with those obtained using “classical” one point calibration LIBS. Comparison with known concentrations showed that the “classical” version had average relative errors for the two samples of 8.78% and 9.28%. The developed method improved these values to 8.07% and 7.56%, representing only a modest improvement.

The other paper reporting the analysis of titanium-based materials was presented by Hou et al.53 These authors attempted to clarify why calibration-free LIBS is normally applicable only when using instrumentation that is gated. Since most systems come equipped with CCDs, i.e. are non-gated, the accuracy can be affected adversely. During the study a time-integration calibration-free model was developed that could be used with non-gated detectors. It was based on the usual plasma diagnostics such as electron number density and temperature. However, the line intensity at different times during the signal collection time window was estimated with self-absorption correction according to the temporal profile of the plasma properties, the model was validated using titanium alloys and the data were compared with those obtained using traditional calibration-free LIBS. The average relative error for the determination of Al and V decreased from 6.07% and 22.34% to 2.01% and 1.92%, respectively. This significant improvement led the authors to believe that their model could enable reliable calibration-free LIBS data to be obtained from non-gated detectors.

Another paper to discuss mathematical treatment of calibration-free LIBS was presented by John et al. who coupled it with multi-element Saha–Boltzmann plots during the analysis of Vitallium alloy.54 In conventional calibration-free LIBS, the Y intercept of the Boltzmann plot is used for estimating the concentration of analytes in the sample. Therefore, the accuracy of the measurement is very dependent on the slope of the graph being correct and, unfortunately, uncertainties in transition probability, statistical fluctuations and self-absorption can affect this. The more generalised method of calculating temperatures, e.g. the multi-element Saha–Boltzmann plot can, to some extent, help overcome this problem. The procedure for doing this and applying it to the LIBS spectra was given in an easy to follow step by step guide in the paper. A comparative study was made between conventional calibration-free LIBS, the developed method, results obtained using SEM-EDS and with the known weight percentage of the material. The method developed showed a significant improvement in accuracy compared with the conventional mode and had a similar accuracy to the SEM-EDS data. For instance the known weight percentage for the material was 65[thin space (1/6-em)]:[thin space (1/6-em)]30[thin space (1/6-em)]:[thin space (1/6-em)]5 for Co[thin space (1/6-em)]:[thin space (1/6-em)]Cr[thin space (1/6-em)]:[thin space (1/6-em)]Mo. The method developed yielded data that were 62 ± 3, 28.5 ± 1.5 and 5.3 ± 0.3 compared with the values obtained using conventional calibration-free LIBS of 76 ± 11, 20.1 ± 9.9 and 3.9 ± 1.1. A clear improvement in both accuracy and precision was therefore obtained.

Another factor that is known to affect the LIBS signal is that of temperature of the sample. This has been shown again by Zou et al. who determined the effect of temperature of gold samples and of artificial samples with differing Au content (0, 400, 600, 700, 800 and 1000 mg kg−1) on the Au signal.55 The analytical procedure, including the instrumental parameters, were provided in the text. An average of 300 measurements over 30 different spots on the gold foil were made to minimise unexpected measurement fluctuations. This number decreased to 200 measurements over 20 locations for the synthesised samples. Plots were made of emission intensity against sample temperature for three different Au wavelengths. All three showed an increase in emission as the temperature increased from 298 K to 573 K, but then dropped slightly at 973 K. Precision of the measurements also showed a distinct improvement at elevated temperature. The artificial samples also showed a similar trend, with a 280% increase in signal intensity, a 20% increase in signal to noise ratio and a 260% improvement in LOD (340.57 mg kg−1 to 95.53 mg kg−1) over the same temperature range. A theoretical study was then undertaken to identify the mechanism of enhancement. It was concluded that the increased temperature of the sample leads to an increase in the amount of sample ablated and an increase in electron number density.

Two papers have reported the analysis of magnesium alloys. In one by Fu et al. the magnesium alloys were cleaned using dilute nitric acid and then 0.05 g of it was acid digested using nitric acid with a small amount of hydrofluoric acid.56 The impurities Cl, P, S and Si were determined in the digests using ICP-MS/MS using nitrous oxide as the reaction gas. The Cl was determined as 35Cl14N+ at m/z 49, the P was determined as 31P16O+ at m/z 47, S was determined as 32S14N+ at m/z 46 and Si was determined as 28Si16O+ at m/z 44. Since all analytes were measured at a significantly higher mass than their isotopes, the sensitivity and their LOD were improved significantly. The LOD were 31.9, 4.43, 10.8 and 17.2 ng L−1 for Cl, P, S and Si, respectively. These were better than those obtained using either oxygen or hydrogen as the reaction gases. The accuracy was checked through standard additions and through the use of sector field (SF)-ICP-MS. Recoveries were in the range 94.0–106% and the concentrations were in agreement with those obtained using SF-ICP-MS.

The other application of magnesium alloy analysis was undertaken by Wagner et al. who used diffusive gradients in thin films (DGT) and LA-ICP-MS to visualise and monitor the corrosion process.57 The methodology was regarded as being a new approach to 2D spatio-temporal imaging that combined the time integrated Mg2+ flux imaging using DGT-LA-ICP-MS with real-time pH imaging using planar optodes. The DGT gel used was iminodiacetate and 10 μm beads of this was used to collect the Mg2+. A 2D distribution of Mg2+ was then constructed through the use of LA-ICP-MS on the resin bed. Meanwhile, the pH-optode reacted to local pH changes through a reversible change in fluorescence intensity. This change in fluorescence intensity was recorded using a digital single-lens reflex camera system, resolving 2D pH dynamics in near real-time at a temporal resolution of 60 s. The commercially available magnesium alloy WE 43 with no protective coating and also with a 3 μm thick polymeric coating (c-We 43) were used during the experiment. The materials were “corroded” using either 0.01 mol L−1 sodium nitrate solution or Hanks balanced salt solution (a simulated body fluid). The full experimental setup was described in the paper. The uncoated material showed intense Mg dissolution (11.9 ng cm−2 s−1) and a concomitant rise in pH during the first 15 min of exposure to the sodium nitrate solution. A much lower rate was observed for the Hanks' solution. The coated material was tested under the same conditions for 120 min and was significantly less corroded. The LA-ICP-MS offered excellent spatial resolution (10 μm × 80 μm) and had a very low LOD of 0.04 ng cm−2 s−1 for the sodium nitrate. The LOD was a factor of 10 worse for the Hanks' solution. This enabled leaching of the Mg through surface defects in the coating to be detected and their position identified.

A second paper also explored the corrosion process, this time of nickel-based materials.58 These authors used the non-destructive technique of grazing exit-X-ray absorption near edge structure (GE-XANES) spectroscopy for the task. The experimental setup was described in full, with the instrumentation comprising a GE-XANES instrument equipped with a pnCCD detector. This combination enabled a scanning-free, non-destructive method that was capable of depth-resolved analysis at the sub-micrometre range. This was especially useful for the analysis of layered samples or for samples that are experiencing corrosion. The setup enabled spatial and energy-resolved measurements and directly extracted the desired fluorescence line that was free from scattering events and other overlapping lines. The methodology was validated through the analysis of a layered sample of well characterised composition. It was concluded that this new method offers exciting possibilities for studying surfaces, e.g. of catalysts, corroded materials, layered structures such as some electronic components, etc.

A low power, low resolution LIBS instrument that was capable of quantifying alloying elements in nickel alloys was described by Choi et al.59 The Nd:YAG laser operated at 1064 nm, with ∼7 ns duration pulses, a 270 μJ per pulse energy and at the repetition rate of 1 kHz. Light emitted from the sample was detected using a miniature spectrometer that covered the wavelength range 200–650 nm and that had a resolution of ∼1 nm and a CCD detector. A series of standard nickel alloys samples was used during the study, including: Nimonic 901, Inconel 600, Inconel 625, Incoloy 800 and alloy UNS NO8825. Calibration was achieved using two different methods: the intensity based method and the ratio-based method. For the intensity method, at the measurement wavelengths of 520.4 nm, 438.1 nm and 547.7 nm for Cr, Fe and Ni, respectively, the LODs obtained were 2.04%, 0.92% and 1.41%. The ratio method (where the ratio of the signal from the analyte to that of a matrix element is used) provided similar LOD, but the precision was much improved, with RSD values of 0.65% and 1.28% being obtained for Cr and Fe, respectively. Having succeeded in analysing the samples, the authors then attempted to classify them. Two models were tested: K nearest neighbour and linear discriminant analysis (LDA). Classification success was 95% for K nearest neighbour and 98.3% for LDA, the authors then attempted to improve the success rate for LDA further by employing a two-step LDA model. Here, the two classes that were the most difficult to separate (i.e. the ones that decreased the classification accuracy) were modelled separately in the second step so that their differences can be exploited more effectively. This second step led to a classification success rate of 100%. The conclusion was that a low performance and low cost instrument can still be used successfully for both quantitative analysis and classification purposes.

Electrothermal vaporisation (ETV) of zinc samples into a two jet arc plasma for optical emission spectrometry was described by Kuptsov et al.60 A Fakel two-jet arc plasma operating at 15 kW with an angle of 70° between the jets and an arc current of 85 A was used with detection via a Paschen Runge style spectrometer with two channels – one from 190–350 nm and the other from 385–470 nm. Sample (20 mg) was placed in a graphite tube. This was heated to 95 °C for 50 s to remove any residual moisture and then vaporised at 2400 °C for 15 s. The resulting vapour was transferred in an argon flow to the two-jet arc. The resulting calibration curves had R2 values of greater than 0.99. Spike recovery experiments using the high-purity zinc sample “granulated zinc 6-09-5294-86” with 50 μL aliquots of liquid standard added were used as a validation method. Spike recoveries were element dependent, with values ranging from 88% (for B) up to 123% (for Al). Most analytes (11 of 17) had a recovery in the range 90–110%. Further validation was achieved through the comparison of data obtained with those obtained using conventional ICP-OES. Detection limits were impressive, with values ranging from 0.05 ng g−1 (for Sr) up to 30 ng g−1 (for Al, Co and Sn). Boron was the only anomaly, having a LOD of 100 ng g−1.

3 Organic chemicals and materials

3.1 Organic chemicals

As with any sample type, it is necessary to prepare and analyse suitable reference materials because they are still the best way to validate new methods and ensure accuracy of established ones. It is well known that SIMS analyses are affected by samples containing water. Zhang et al. have produced nine materials (four clinopyroxene and five orthopyroxenes) that have been analysed using FTIR and SIMS and have a known water content.61 Tests indicated that most of the samples were homogeneous, with a standard deviation of 18.6% (for 2 SD) for the analysis of both inter- and intra-fragments. The one exception was the clinopyroxene called diopside which showed some inter-fragment heterogeneity. The authors stated that although this particular sample could be used as a reference material, it should be used with some caution. The clinopyroxene samples had water contents of between 24 ± 3 and 774 ± 26 mg kg−1 whereas the orthopyroxene samples had a content of between 45 ± 3 and 534 ± 59 mg kg−1.

A review (with 179 references) of laser-based standoff detection of hazardous materials including explosives has been presented by Narlagiri et al.62 Techniques included in the overview included Raman spectroscopy, LIBS and LIF as well as an assortment of less well known techniques such as quantum cascade laser and external cavity quantum cascade laser-based IR, photoacoustic spectroscopy and terahertz spectroscopic imaging. The authors stated that the recent development of improved laser sources and detectors as well as the availability of databases of various hazardous materials has led to rapid evolution of the standoff techniques over the last few years. This means that they have improved from being relatively short-ranged and with low sensitivity for a limited range of materials under laboratory conditions to much longer range in ordinary atmospheric conditions and for a much larger range of materials with greater sensitivity. The review discussed all of these developments at length. Also discussed in the review were the different lasers that have been used. The advantages and limitations of the various techniques were also highlighted and future prospects discussed.

Two papers have discussed the use of LIBS to monitor laser-based paint removal.63,64 In the paper by Li et al. a LIBS platform that enabled the monitoring of paint removal from an aluminium-based aircraft skin was developed.63 A high frequency (kilohertz level) nanosecond IR pulsed laser was used for the paint removal as well as the formation of the LIBS plasma. After subtracting the spectra's continuous background and screening for the key features, a classification model of the three types of spectra (top coating, primer and aluminium substrate) was constructed using the random forest algorithm. A real-time monitoring criterion based on the classification model and multiple LIBS spectra was established and verified experimentally. Classification accuracy was 98.89% and the time required to classify each spectrum was 0.03 ms. The results obtained were in good agreement with those obtained using macroscopic observation and microscopic profile analysis. It was concluded that the LIBS cleaning/monitoring platform constructed offered core technical support for real-time aircraft skin cleaning. The other paper was by Choi et al. who used a high-powered Nd:YAG laser to clean thick paint layers from 304L stainless steel.64 Two types of two dimensional maps were constructed: the LIBS intensities of the elements Cr (at 520.84 nm), Fe (at 374.96 nm) and Na (at 588.99 nm) and the correlation coefficient distribution maps. The latter type was constructed using the Pearson's correlation coefficient. The two map types were compared with elemental distribution maps obtained using electron probe microanalysis (EPMA) and were in excellent agreement. Both of the above papers shows the applicability of LIBS, when used with multivariate analysis, for the application of the rapid, real-time monitoring of paint cleaning.

Another paint-related publication was presented by da Silva et al. who used microwave-assisted acid dissolution methods that had been optimised using central composite design for the determination of Cu and Sn in anti-fouling paints.65 Sample mass as well as the volumes of hydrochloric, hydrofluoric and nitric acids and hydrogen peroxide required for dissolution were all optimised. Once the optimal conditions had been established, the digests were analysed using ICP-OES. Results obtained using the microwave-assisted extraction procedures were compared with those obtained following a dry ashing procedure utilising a muffle furnace and with those obtained using LA-ICP-MS. Although all of the extraction procedures yielded good results, the best was the one that utilised a digestion medium of hydrofluoric acid, nitric acid and hydrogen peroxide. This is because complete dissolution occurred whereas using some of the other dissolution methods, some particulate matter remained. For these, the analytes must have been extracted because they yielded similar data. However, the particulate matter either needs the extra step of filtration or care should be taken to ensure it does not clog the ICP instrument's nebuliser. All procedures were discussed in the text, giving full experimental detail.

Two papers have discussed different aspects of paper analysis. In one in Chinese by Guo et al. food packaging paper was analysed using XRF spectrometry and the experimental data treated using principal component analysis (PCA), t-distribution random neighbourhood embedding and cluster analysis.66 In total, 44 different papers were analysed from several different sources. Using 80% of the samples as a training set for the artificial neural network, the authors attempted to classify the remaining 20%. Overall, they were reasonably successful, with an accuracy of prediction of 88.9% being achieved.

The other paper-orientated study was presented by Haekkaenen et al. who used LIBS and optical profilometry to assess the penetration depth of toners in three coated papers.67 The LIBS instrument comprised an ArF laser operating at 193 nm and with a fluence of 0.6 J cm−2. Light collected was transmitted via a fibre optic to the spectrometer which was equipped with a grating with 2400 grooves per mm and an intensified CCD detector. The penetration depth of the laser pulse differed depending on the layer of paper under analysis, with the toner being 150 nm, the coatings of approximately 350 nm and the depth in the base paper of approximately 1 μm. Using these data, it was possible to estimate the depths of each layer. Despite this, the width of each ablation was constant at 0.3 mm. The LIBS spectra revealed that C, Si and Ti were present in all toners, but the presence of Cu was indicative of the presence of a cyan toner, although it was also present albeit at lower concentration in the black toner. The method developed provided a unique and accurate means to study the extent of toner diffusion in coated papers.

An interesting application was described by Sun et al. who used a system combining LIBS and Raman spectrometry to analyse mothballs and the vapour given off from them.68 The LIBS was used to analyse two kinds of mothballs and to identify the characteristic emission lines that come from the volatile components. These wavelengths were for C (247.8 nm), CN (several bands), H (656.2 nm), Na (589.0 and 589.6 nm) and Cl (725.6 nm, 833.3 nm and 837.6 nm). The light from these wavelengths were then used to construct a machine learning algorithm based on PCA and support vector classification so that the spectra from air, from synthetic mothballs and from natural camphor could be analysed. The algorithm developed achieved a recognition accuracy of 98.33%, indicating that LIBS is capable of recognising airborne volatile substances originating from mothballs. The Raman spectrometry was used to supplement the LIBS data.

Li et al. used LIBS to calculate the heat of detonation of energetic materials.63 The overall premise was that the energy released is related to the intensity of the LIBS spectra and may be calculated by inserting emission data into a machine learning algorithm. The method involved attaching 15 mg of the material to double-sided tape and then attaching that to a glass slide. This was then mounted on an XYZ translation table. A laser operating at 1064 nm with an energy of 50 mJ was fired at it with a frequency of 1 Hz. A total of 100 shots were made at each sample ensuring that none of them overlapped. The emission intensity from the spectra then underwent a statistical correction to improve the quality and to extract the spectral features such as the emission intensity and emission shape correlation intensity. This was to decrease the negative effect of noisy variables. The resulting data were input to PCA-partial least squares (PLS). The paper discussed the statistical manipulation in great depth. However, when the data from 12 different explosives were used to train the model, the result was that it could predict the heat of detonation with great accuracy. It was concluded that the work could facilitate safe and rapid determination of the heat of detonation for small samples.

3.1.1 Pharmaceutical materials. There has been a large selection of different applications of the analysis of pharmaceutical products. Several have involved the use of LIBS analyses, whereas several others have involved the use of a chromatographic technique to separate different species or metabolites.

As with so many areas of analytical chemistry, LIBS has started to encroach on the analysis of pharmaceutical products. An example, by Wei et al. used LIBS followed by chemometric analysis of the analytical data to classify the different manufacturers of penicillin.69 This is important because it can help trace drug quality issues and can potentially identify counterfeit drugs. Twelve samples of three types of penicillin (phenoxymethylpenicillin potassium tablets, amoxicillin capsules and amoxicillin and clavulanate potassium tablets) obtained from 10 manufacturers were analysed in the study. The characteristic lines of the three drugs were identified and ranked in order of importance according to the decrease in the Gini index of the random forest algorithm. Analytes chosen were C, H, N and O. However, some of these analytes would also have a contribution to the signal from the discharge occurring in air. Some metallic analytes were therefore also chosen including Ca, Fe and Na. Three different classification algorithms were tested for their efficiency: linear discrimination analysis (LDA), support vector machine (SVM) and artificial neural network (ANN). All three classifiers performed excellently, with LDA having the lowest classification success rate of 97.50% for the amoxicillin capsules. The ANN was the most efficient with a 100% success rate for all three drug types. Unsurprisingly, the authors regarded this as a success and concluded that the methodology had great promise for this type of work.

Two other papers very similar in structure to the one by Wei et al.69 discussed above have been presented. Dwivedi et al. used optical and atomic force microscopy to analyse the surface of tablets and LIBS to analyse the chemical constituents and to quantify the hardness.70 Using a Nd:YAG laser and a spectrometer equipped with an intensified CCD the analytes C, Ca, H, Mg, N and O as well as CN were determined. The multivariate methods of PCA and ANN were used in an attempt to classify drugs made by five different manufacturers. The second paper, by Farhadian et al., again used PCA and ANN on LIBS data obtained from eight samples of three pharmaceutical classes (statins, beta blockers and benzodiazepines).71 Using a commercial instrument, they analysed the same spot on each sample 10 times and then measured the same sample in 10 different places. Each sample was therefore analysed a total of 100 times. The PCA reduced the data set to be input to the ANN, but also managed to separate the eight samples reasonably successfully. When using the simplified dataset, the ANN managed to classify the samples successfully and enabled predictions to be made of unknown samples.

Two papers have used size exclusion chromatography (SEC) coupled with ICP-MS to analyse pharmaceutical products. In one by Whitty-Leveille et al., ultra-trace metal–protein interactions in co-formulated monoclonal antibody drug products were determined.72 The study attempted to differentiate between the analytes (Co, Cr, Cu, Fe and Ni) that were interacting with the proteins and the same trace metals that were just free ions in the drug product. These analytes were chosen because they are the components of 316L stainless steel, the type most often used in biopharmaceutical applications (and hence, the most likely contaminants). Two monoclonal antibodies were formulated and stored for up to nine days in a scaled-down model to mimic metal exposure from manufacturing tanks. The materials were first analysed using conventional ICP-MS to ascertain the “total” metal content. This was achieved after an acid digestion of the materials. All analytes except Co showed a significant increase in concentration over the storage time period of 0–9 days, with Fe having the highest concentration; increasing from 8.2 ± 0.2 μg L−1 to 87 ± 2.3 μg L−1. Then, SEC-ICP-MS was used to separate the free ions from those associated with the drug. Sample was diluted to give a protein concentration of 1 mg mL−1 and then 5 μL was injected to the system. Relative quantification of metal–protein interaction was calculated using the relative areas under the peaks and weighting it to the “total” concentrations. Most analytes showed that the large majority remained as free ions – even after the nine days of storage. The exception was Fe where, after 7–9 days, approximately two thirds had become associated with the drug.

The other SEC-ICP-MS application was presented by Yamazaki et al. who quantified oligo-nucleic acids, a ferritin nanocage and their complexes simultaneously.73 The oligo-nucleic acid and ferritin were used as model compounds of nucleic acid drugs and a drug delivery system, respectively. The SEC separated the nucleic acid–ferritin complex completely from the free nucleic acids. Quantification of the nucleic acids and the ferritin was based on the number of P and S atoms, respectively, and was achieved employing an external calibration method using a series of elemental standard solutions. The sensitivity and accuracy of the method was assessed as being appropriate for evaluating the medicines already on the market. The mechanism of encapsulation was also studied and was found to possibly be regulated by both the average molecular size of the nucleic acids but also the surface charge related to the number of (deoxy-)ribonucleotides. These authors also thought that their methodology could potentially accelerate the development new therapies in the future.

The determination of drug metabolites and impurities is clearly an important area of research and Yao et al. have provided a study in which platinum chloride impurities in Pt-based drugs were quantified using HPLC-ICP-MS.74 The separation of inorganic Pt–chloride complexes (K2PtCl4 and K2PtCl6) was achieved in less than three minutes on a Hamilton PRP X-100 anion exchange column (15 cm × 4.6 mm) using 2 mmol L−1 hydrochloric acid as the eluent. It should be noted that a matrix elimination series of valves was placed between the end of the column and the ICP-MS instrument. This was required to ensure that the main drug did not reach the ICP-MS instrument because it would lead to saturation of the detector and require excessively long washout times to return to baseline levels. This could easily mask the contaminant signals. The LOD were 0.2 μg L−1 and 0.3 μg L−1 for K2PtCl6 and K2PtCl4, respectively. Precision for repeat injections of a 10 μg L−1 standard was better than 3% RSD and spike experiments yielded recoveries of between 94 and 102%. The methodology was described as being simple, rapid, sensitive and could easily be adopted as the method of choice for quality control of impurities in Pt-based drugs.

The detection of fluorinated and chlorinated compounds using ICP-MS has been difficult because of the extremely high ionisation potential associated with Cl and F as well as the polyatomic interferences that exist. Redeker et al. reported the use of a technique that used LC-ICP-nanospray-Orbitrap.75 These workers developed a system they entitled plasma-assisted reaction chemical ionisation. This is where the sample species were transported via an LC pump and separated using a reversed phase column at a flow rate of 50 μL min−1, nebulised and transported to a plasma in a flow of oxygen and then atomised and ionised as normal. The central channel of the plasma was then sampled using a quartz tube from a region that is sufficiently cool so that recombination of ions occurs forming molecules. In this way, HCl and HF were formed and these plasma products were mixed with a 1 mM solution of barium acetate that was introduced using a nanospray forming the ions BaCl+ and BaF+. Rather than having these detected using a conventional ICP-MS mass spectrometer, they were introduced to an Orbitrap mass spectrometer. This has far higher resolving power than any ICP-MS instrument and therefore easily overcomes the polyatomic interferences. Another advantage was that the instrument could readily be converted back to conventional electrospray ionisation enabling molecular information to be obtained. The LOD for Cl was in the region 8–12 pmol whereas for F the LOD were typically 5–12 pmol. The system was applied to the identification and quantification of drug metabolites. Spike experiments yielded a recovery of greater than 80%. The system developed found that minor metabolites represented up to 8% of the parent Flurazepam drug.

X-ray-based techniques have also found use for the analysis of pharmaceutical products. A paper by Zhu et al. described the use of a portable EDXRF instrument for the determination of 22 elemental impurities in the oral solid dose drugs Contezolid and Aztreonam.76 Also used during the study was cellulose onto which elemental standards were injected. The analytes were split into three groups: those requiring low energy (Co, Cr, Ni and V), medium energy (As, Au, Cu, Hg, Ir, Os, Pb, Pd, Rh, Ru, Se and Tl) and high energy (Ag, Cd, Mo, Pt, Sb and Sn). To overcome interferences, the non-linear least square method of fundamental parameters was applied. The paper discussed this approach in detail. The result was that the data met the accuracy, linearity and precision required by the United States Pharmacopoeia. The LOQ obtained also reached the desired level for the analytes based on the 10 g maximum daily intake. It was concluded that the method offered a novel, rapid, non-destructive, portable and sensitive way of determining the impurity analytes.

The other X-ray-based application was by Chuparina et al. who used both WDXRF and TXRF to determine Mn and Se in plant-derived materials that can be used as raw materials to synthesise new drugs.77 Manganese was reacted with the compound dihydroquercitin and Se was reacted with arabinogalactan. It was then necessary to check the concentrations of the two elements in the final products. Since no suitable CRMs are available for these materials, the authors resorted to preparing a calibration series using the synthesised materials. For the TXRF analysis, two sample preparation protocols were tested: acid digestion and dissolution using water. For the Mn–dihydroquercitin compounds the Mn content found using WDXRF was 13.5–17.4%. This was in good agreement with the data obtained using TXRF which were in the range 13.1–18.2%. The Se–arabinogalactan compounds had a Se concentration of between 0.5 and 12.9% as found using WDXRF and between 0.5 and 13.6% for TXRF. Again, these appeared to be in tolerably good agreement, with a statistical test (the t-test) on the results indicating that there were no significant differences. The results from both techniques were also in agreement with the calculated theoretical values. When TXRF was used to test for other contaminating elements, nothing significant was observed in the Mn–dihydroquercitin. However, there were some detectable peaks in the energy range 5.9–7.5 eV for the Se–arabinogalactan. Their presence was attributed to contamination.

3.1.2 Cosmetic samples. It is well known that one of the main drawbacks of LIBS analyses is that of calibration. This is because the calibrants need to be exceptionally closely matched to the sample to ensure that the amount of sample vaporised per laser shot is the same. Since this is often impossible, assorted chemometric algorithms have been used in an attempt to overcome this. An alternative has been proposed by Joca et al. who have developed a novel strategy for preparing calibration standards and samples using beeswax as a substrate.78 The method was tested by analysing lubricating oils and cosmetic samples (two red lipsticks, liquid lipstick and face foundation). Beeswax (15.0 g) was placed into a 50 mL polypropylene tube, and then lipstick or face foundation was added into the tubes to reach a concentration of 1.0 mg g−1 and 10 mg g−1, respectively. The tube was then heated in a water bath at 85 °C for 180 s. The melted mixture was then vortex mixed to ensure homogeneity. Then, using a pre-warmed pipette tip, 700 μL of the fused mixture was transferred to an acrylic support containing a 15 mm diameter hole, cooled in a freezer for 10 min before the solidified beeswax disk containing the sample was removed. Standards were prepared in a similar fashion but adding the relevant amount of organometallic standard to the beeswax. The LIBS analyses were undertaken on the disks with the analytes Fe, Mg, Si, Ti and Zn being determined. The internal standard used was either the C line at 505.50 nm or at 516.4 nm. Method validation was achieved using alternative techniques involving acid dissolution followed by either ICP-OES or ICP-MS analysis. Results from the proposed method were in good agreement (90–110%) with the alternative techniques. Calibration regressions for the proposed method were better than 0.95 and detection limits for the analytes fell into the range 4–24 μg g−1.

Elemental profiles and heavy metal content of natural tattoos (jagua) and dyes (henna) were obtained using ICP-MS analysis by Rubio et al.79 A total of 34 samples (some pastes and some solids) were purchased from a commercial website and were analysed for 11 analytes. Material was acid digested, first at room temperature because of the high organic content and then heated on a hotplate. After evaporation to dryness, the residue was taken up in aqua regia before being evaporated to dryness again. It was then taken up in nitric acid (2 mL) and heated before the addition of hydrogen peroxide. The digests were finally diluted to volume and were ready for analysis. The digestion took a total of approximately 30 hours. Although digestions of organic-rich samples can be challenging, 30 hours seems excessive. However, the results showed decent precision (1.2–9.9% RSD), acceptable recovery values from spiking experiments (87–118%) and, most importantly, that none of the 34 samples met the current European cosmetic regulations.

An ultra-sensitive method for the determination of As in eye shadow samples for juveniles was described by da Costa et al. who used dispersive magnetic solid phase extraction followed by flow injection analysis HG-AAS for detection.80 Samples were first acid digested using nitric and hydrofluoric acids with microwave assistance. The concentrations of 11 analytes were then determined using ICP-OES. However, the As content was too low for this analyte to be detected. Therefore, the authors devised a scheme to preconcentrate the As. Maghemite nanoparticles (γ-iron(III) oxide) were prepared in-house and characterised using XRD, FTIR and nitrogen adsorption. Both the adsorption process of the As onto the maghemite particles and the HG-AAS detection system were optimised using central composite design. Optimal conditions for adsorption were: pH 5.3, mass of maghemite particles = 184 mg and a stirring time of 41 minutes. Optimal HG-AAS conditions were: HCl concentration = 4 mol L−1, sodium tetrahydroborate concentration = 1.1% m/v and carrier gas flow = 145 mL min−1. A preconcentration factor of 50 was achieved which enabled a LOD of 0.9 mg kg−1 to be obtained. Spike recoveries were a little variable, with values of between 75 and 99% being achieved. Although the preconcentration procedure for the As was relatively simple, it was far from quick, with a further 20 minute period being required to remove the As from the particles. This meant that each sample took in excess of an hour to be prepared.

Another paper describing the analysis of eyeshadow was presented by Pawlaczyk et al.81 These workers obtained 94 samples representing different product types (matte or pearl), a range of expense, 12 different colours, eight manufacturers and four countries of origin. All samples were acid digested with microwave assistance using nitric acid and hydrogen peroxide and then diluted ready for analysis using ICP-MS. A small number of adult samples exceeded current legislative values (two samples had a Cd content greater than 0.5 mg kg−1 and one sample had a Pb content greater than 10 mg kg−1). This led the authors to conclude that the available products for adults were, by and large, safe. However, the results from the analysis of children's products were described as “alarming”, with concentrations of Cd reaching 4 mg kg−1 and Pb reaching 16 mg kg−1. A total of seven analytes were determined, with some, e.g. Ba and Bi reaching extremely high levels in some products (>2100 and approximately 1200 mg kg−1, respectively). Using only the Cd and Pb values for assessment, the country supplying the least contaminated products was Canada while Italy and Poland supplied the most contaminated. Some interesting trends were observed. The most contaminated samples appeared to be those that were pink, grey or “sea colour”. The matte samples had higher Cd, whereas the pearl ones tended to have higher Ba, Bi, Pb and Tl. Another interesting observation was that the more expensive eye shadows tended to have fewer contaminants. Multivariate statistical analysis (cluster analysis and PCA) were used and also showed some interesting trends.

3.2 Fuels and lubricants

The abstract contribution for this section has decreased this year possibly due to fossil fuels and petrochemicals being seen as environmentally unfriendly and many funding agencies being reluctant to be seen supporting this field. However, currently there is still a place in the economy for jet fuels, lubricants, gasoline and diesel and a whole host of other petrochemical products made from crude oil. It was disheartening to see a huge reduction in papers on alternative fuels particularly biofuels. These fuels are coming on-line and presenting problems not seen in traditional fuels both with their use and analysis. Industry is having to modify analytical methods to fit however very little of this work appeared in this year’s published literature. There is also, somewhat contrary to the present climate, an unexpected jump in the number of papers on coal analysis. However, nearly all the coal papers are from China and most are concerning LIBS calculations to improve the analysis of ash content and calorific value. This possibly reflects this country's continued and expanding use of this fuel.
3.2.1 Petroleum products – gasoline, diesel, gasohol and exhaust particulates. Very few papers were seen this year on this topic and only one justified mention, this was by Coelho et al.82 and described a method using palladium nanoparticles as a chemical modifier for Pb determination in leached extracts and petroleum waste by high resolution continuum source AAS. Hydrophilic palladium nanocubes and palladium nanospheres were synthesised with sizes varying from 3 to 22 nm. Positive effects in terms of sensitivity and thermal stability were found with the nanoparticle addition particularly using the 3 nm spherical-shaped nanoparticles. The optimised pyrolysis and atomisation temperatures were 800 °C and 2300 °C, respectively using 200 ng of 3 nm palladium nanoparticles. The limit of detection was 4 μg L−1 and the precision better than 18%. The method was successfully applied to determine Pb in leached extracts obtained from oily sludge and drill cuttings and the obtained concentrations ranged from <13 to 354.4 μg L−1.
3.2.2 Coal, peat and other solid fuels. Coal, despite the huge environmental issues around its use, is still a very strong contributor to this section. The first contribution, by Zhu et al., described an application of multi-point LIBS to investigate the mineral release of dispersed particle streams during coal combustion in an enclosed laminar flow reactor.83 A 1064 nm Nd:YAG laser was used to ablate the samples at atmospheric pressure. The electron temperature and electron density were calculated to evaluate the characteristics of the laser induced plasma along with the height from the burner surface. The particle temperature was measured by a two-colour pyrometer and the time-resolved behaviour of Na, Ca and Fe along with the combustion process were measured using LIBS. The electron temperature and electron density decreased with the flow of coal and became relatively stable during combustion. The Na release, mainly in atomic forms, decreased with the particle flow upon devolatilization. However, the Fe and Ca was rarely released upon devolatilization. During char combustion, the Ca had a similar behaviour to Na while the Fe increased and then remained constant. The temperature dependent kinetics of the Na and Ca release rate was found to obey an Arrhenius expression.

A contribution to this section was from Rajavelu et al. who described a study to demonstrate the feasibility of a fibre-based LIBS technique for the analysis of pulverised coal.84 A hollow core fibre was used to replace the air medium in conventional LIBS to deliver high power laser pulses to the target. This fibre had a transmission efficiency of 50–60%. The spectra captured using this technique proved that the addition of a hollow core fibre can upgrade conventional LIBS into a remote characterisation unit.

The next contribution was by Lu et al. who investigated the secondary breakdown of double pulse LIBS with different focussing geometries and positions.85 In this work petroleum coke samples were investigated and the plasma morphology, emission intensity enhancement, signal repeatability, plasma temporal evolution and plasma properties were systematically studied under different conditions. The focusing geometry and axial focusing position were shown to play an important role in the emission intensity and repeatability of the signal. With the focusing geometry set up to minimise spherical aberration, the emission intensity and repeatability were improved. The spectral intensity and repeatability depended strongly on the axial focusing position of the second laser pulse. The images of the plasma showed that secondary breakdown occurs almost on the boundary of the first formed plasma. The confinement effect and emission enhancement can be observed when the secondary breakdown occurs on both sides of the first plasma boundary. The results showed that there is an optimal focusing geometry and axial focusing position that could enhance emission and repeatability in double pulse LIBS.

The next paper in this section, by Ling et al., described a method for correcting the lens to sample distance change on the signal intensity in LIBS.86 A correction formula that fits the relationship between the obtained signal intensity and the deviation between the lens to sample distance and the focal length was constructed through a series of experiments based on 18 standard coal samples and validated with three types of unknown coal samples. Compared with the original signal intensity the relative errors between the corrected signal intensity and the signal intensity when the lens to sample distance is equal to the focal length decreased by a factor of more than ten for almost all the elements in the three samples. This indicates that the proposed method can be used to correct the signal intensity of the elements C, H and O in real-time online coal analysis and, if C can be determined reliably, it would provide a method of determining the calorific value of the coal.

Tian et al. looked to use combined LIBS and XRF to improve the repeatability of proximate analysis of coal in thermal power plants.87 The LIBS was used for the determination of C and H and XRF for the determination of the inorganic ash forming elements. This improved stability compared with using LIBS for the whole analysis. The combination of elemental lines in LIBS and XRF spectra and principal component analysis (PCA) with partial least squares was used to establish a prediction model and perform multi-elemental and proximate analysis of coal. Quantitative analysis results showed that the precision of C measurements was 0.15% RSD, with the RSDs of other elements being less than 4%. The standard deviations of calorific value, ash content, S content and volatile matter are 0.11 MJ kg−1, 0.17%, 0.79% and 0.41%, respectively indicating that the method has good repeatability in determining coal quality.

The next two papers both used a machine learning approach to evaluate coal ash content. The first, by Wen et al., used machine learning to rapidly evaluate coal ash content using X-ray fluorescence spectrometry.88 To evaluate the method, an XRF dataset containing 217 sets with different ash contents was constructed. The dataset was divided into a training set and a test set in the proportion 8[thin space (1/6-em)]:[thin space (1/6-em)]2. The processing package RandomizedSearchCV was used to optimise parameters during model training. Experimental results showed that a random forest regression model produced a superior prediction performance compared with other models. The contribution and role of each element to the ash prediction model was explained and investigated. Using ‘Shapley additive explanations interpretation’ which is a game theory approach, the nine most important elements were identified as being Al, Ca, Fe, K, S, Si, Sr, Ti, and Zr. These elements had the greatest contribution to the performance of the model. This suggests that interpreted machine learning models and XRF data could be a good alternative to conventional coal ash content prediction.

The second paper using machine learning was from Huang et al. and used a technique based on feed-forward neural networks and improved particle swarm optimization to predict coal ash content.89 The data set was obtained by testing the elemental content of 198 coal samples using XRF. The input elements for machine learning Al, Ca, Fe, K, Mg, Na, P, Si, Ti and Zn were determined by combining the X-ray photoelectron spectroscopy data with the change in the physical phase of each element in the coal samples during combustion. Mean squared error and coefficient of determination were chosen as the performance measures for the model. The improved particle swarm optimization and feed forward neural network model showed strong prediction ability and good accuracy for coal ash prediction. The effect of the input element content of the improved particle swarm optimization and feed forward neural network model for ash content was investigated. It was found that the K content was the most significant factor affecting the ash content. This study is useful for real-time online, accurate, and fast prediction of coal ash.

The last two papers in this section also used neural network applications. The first by Liu et al. investigated automatic coal rock recognition by LIBS combined with an artificial neural network.90 Automatic coal rock recognition is of considerable theoretical and practical significance for unmanned coal mining. The samples in this study were analysed using LIBS. Spectral data were optimised, and dimensionality reduction was performed using partial least-squares discriminant analysis. Then, 10 selected wavelength lines were used to construct a simplified spectral model. The artificial neural network was based on a simple spectral model and was designed to classify the coal and rock. The results demonstrated that LIBS combined with an artificial neural network has a high recognition accuracy rate providing a rapid and accurate coal-rock recognition method for unmanned coal mining.

The last paper in this section was by Chen et al. who looked into the correction of moisture interference during the LIBS analysis of coal by combining neural networks and random spectral attenuation.91 In this study regression models of volatile content were established based on low humidity samples to then analyse samples with higher moisture content. The random spectral attenuation method was used to reduce moisture content interference. The calibration spectra were replicated and multiplied by random attenuation coefficients to introduce information about the interference. All the simulated attenuated spectra were then used to train the artificial neural network quantitative model. Compared with direct modelling without random spectral attenuation, the coefficient of determination was improved from −3.4291 to 0.7102 and the root-mean-square error was reduced from 1.8709% to 0.4786. The results showed the ability of the method to detect samples containing moisture without additional sample pre-treatment which improves the speed of the LIBS analysis.

3.2.3 Oils – crude oil, lubricants. In this section only 5 papers were worthy of note. The first, by Silva et al., studied a miniaturised liquid–liquid extraction method for Ca, K, Mg and Na from crude oil prior to their determination using FAAS.92 The type of extraction solution, sample mass, heating temperature and time, stirring time, centrifugation time and the use of toluene and a chemical de-emulsifier were evaluated. Results were compared with those obtained after high-pressure microwave-assisted wet digestion and FAAS determination. No statistical difference was observed between the reference values and those using the optimised conditions for liquid–liquid extraction. These optimised conditions were: 2.5 g of sample, 1000 μL of 2 mol L−1 HNO3, 50 mg L−1 of chemical de-emulsifier in 500 μL of toluene, 10 min of heating at 80 °C, followed by 60 s of stirring and 10 min of centrifugation. The limits of quantification were 5.0, 1.5, 0.50 and 1.2 μg g−1 for Ca, K, Mg and Na, respectively and the RSDs were better than 6%. The proposed miniaturized extraction method presents several advantages such as ease-of-use and higher throughput. The use of a diluted solution for extraction also reduces the amount of reagents used and consequently laboratory waste.

Shishov et al. investigated the use of a deep eutectic solvent based on choline chloride for the extraction of metals from oil samples prior to elemental determination using ICP-OES.93 In this work deep eutectic solvents based on choline chloride and different hydrogen bond donors were systematically investigated for separation of single and multiple charged metals from oil matrices. The effect of deep eutectic solvent precursors was shown to play a critical role in the extraction of different metals. Carboxylic acids, polyols, sugars and urea and its derivatives were all studied as hydrogen bond donors. The deep eutectic solvent containing choline chloride, lactic acid and water provided the highest extraction recovery. Under the optimal conditions the LOD were in the range 0.02 to 17 mg kg−1.

Altuwaijari et al. developed a new sample preparation approach for the extraction of metal ions from crude oil prior to their quantification using ICP-MS.94 The extraction procedure was based on dispersive solid phase extraction in which dithiooxamide particles were produced in situ and used as the adsorbent. To achieve high extraction efficiency the sorbent was first dissolved in an organic solvent and then injected into the diluted sample. The dithiooxamide was then re-precipitated in the sample solution as tiny particles. The adsorbed ions were desorbed using a few microliters of choline chloride[thin space (1/6-em)]:[thin space (1/6-em)]5-amino-8-hydroxyquinoline deep eutectic solvent under sonication and analysed directly using ICP-MS. Limits of detection of 0.003–2.32 ng g−1 and LOQ of 0.009–7.56 ng g−1 were achieved. Relative standard deviations of less than or equal to 4.3% for intra- and inter-day precisions and acceptable extraction recoveries of 66–91% were obtained. Seven crude oil samples were analysed and 10 metal ions were determined successfully.

Two papers used LIBS for analysis. The first, by Joca et al., looked at the use of beeswax as a substrate for the preparation of standards for LIBS analysis of cosmetic and lubricant samples.78 For the lubricant oil samples, the best results were obtained by using the C2 emission band at 516.4 nm as an internal standard for determining Si and Zn while the C atomic emission line at 505.50 nm was best for Mg quantification. For Fe quantification the C2 emission bands at 516.4 and 545.7 nm both yielded similar results. For all elements stated there was an agreement in results of between 90 and 110% comparing LIBS with ICP-OES and/or ICP-MS. All the calibration curves showed a coefficient of determination higher than 0.95 and detection limits were between 4 and 18 μg g−1 for Fe, 4 and 6 μg g−1 for Mg, 8 and 24 μg g−1 for Si, 4 and 14 μg g−1 for Ti, and 4 and 8 μg g−1 for Zn.

The last paper in this section was by Xu et al. and described a hybrid method combining discharge assisted LIBS with wavelet transform for trace elemental analysis in liquid targets.95 Although LIBS is a highly promising detection technology for the quantitative determination of trace elements in liquids, there are several challenges to overcome. This is because of the fast plasma quenching, liquid level instability and limited laser-energy absorption making rapid real-time quantitative detection of trace elements with high-sensitivity in liquid targets very difficult. In this paper a feasible hybrid method of discharge-assisted LIBS with wavelet transform de-noising was proposed for trace metal element analysis in oil pollutants. Compared with conventional laser-induced breakdown spectroscopy this method has the capacity to increase signal intensities of trace metal elements by one order of magnitude. For Ca the signal to noise ratio is increased by 16-fold and the detection limits of Al, Ba, Ca, Cr, Fe, Na and Zn were lowered by a factor of between 2 and 24 compared with the original conventional LIBS level. Using a standard addition method the recoveries of Ba and Fe were in the range of 101.81–105.45%. This work provides an alternative, economical and reliable method for rapid real-time quantitative analysis of trace metal elements in various industrial applications associated with oil pollutants.

3.2.4 Alternative fuels. Three papers proved interesting in this section this year. The first by Liu et al. described a method using LIBS for local equivalence ratio measurement in opposed jet methane–air flames.96 In this method LIBS was applied to both non-reacting and reacting opposed jet flows to evaluate the measurement accuracy of local air–fuel ratios. Images and emission spectra of the laser-induced plasma were investigated simultaneously using both an intensified charge coupled device (CCD) and a spectrometer with a high degree of spatial and temporal resolution. The influences of the camera delay time, exposure time and laser pulse energy on the LIBS measurements were investigated. The spectral intensity ratios of H/O and C-2/CN on air–fuel ratio was quantified over a wide range of conditions extending from pure air to pure fuel. It was found that H/O and C-2/CN intensity ratios depended on the mole fraction of methane in the ranges of 0.0–0.8 and 0.3–1.0, respectively. The presence of a flame within the laser beam led to significant measurement deterioration relative to the corresponding non-reacting flows. This was corrected by increasing the laser pulse energy and applying a data processing method. This correction method was able to reduce the equivalence ratio measurement uncertainty to within 10% for mixtures with methane mole fractions lower than 50% and to within 15% for mixtures with a higher methane mole fraction. The LIBS measurements of air–fuel ratio in non-premixed flames were finally compared successfully with CHEMKIN (a software package whose purpose is to facilitate the formation, solution, and interpretation of problems involving elementary gas-phase chemical kinetics) program simulations. This demonstrates the ability of LIBS to accurately measure the spatial gradient of the air–fuel ratio.

The next paper in this section, by Varga et al., was an inter-comparison exercise on fuel samples for the determination of bio-content ratio using 14C accelerator mass spectrometry.97 Renewable components in the fuel industry, such as biofuels, are generally made by biological processing of recent organic materials. Based on classical analytical techniques, such as chromatography methods, it is difficult to distinguish the fossil (petroleum based) and biogenic (recent) component of the mixed fuel samples because the physical and chemical properties of these materials are quite similar. The carbon content of the fuel materials is generally high, so the measurement of the carbon isotopes composition, for the purpose of bio-content analysis, can be representative of the whole sample bio-content ratio. The method of determining the bio-based carbon content in liquid fuel samples is standardised but different laboratories use different protocols for sample preparation and perform the measurements using different instruments. Inter-comparison between the laboratories is therefore necessary to prove precision and accuracy and to demonstrate that the results are comparable. In this study three C-14 accelerator mass spectrometry laboratories were compared. Five samples were used in the measurement campaign, including two biocomponents (fatty acid methyl ester, hydrotreated vegetable oil), one fossil component (fossil diesel), and two blends (mixtures of fossil and bio-component with 90[thin space (1/6-em)]:[thin space (1/6-em)]10 mixing ratios). The results presented by the laboratories were comparable, and all three laboratories could determine the bio-based C content of the samples within 1% relative uncertainty which is acceptable in the scientific, economic, and industrial fields of bio-component determination.

The last paper in this section is a review article by Mere et al. containing 52 references on arsenic determination in the petroleum industry.98 The presence of As in natural gas and liquid hydrocarbons is of great concern for oil companies. In addition to health risks due to its toxicity and environmental issues As is responsible for irreversible poisoning of catalysts and clogging of pipes via the accumulation of As-containing precipitates. To address these problems robust methods for the determination of As and its compounds in oil streams are required. This review outlines the sampling techniques, sample preparation methods and As determination techniques developed during recent decades which are commonly used in the oil industry and in the new feedstock energy domain.

3.3 Polymers and composites

Although many aspects of polymer analysis have been covered during this review period, the two standout topics have been the waste management and the analysis of microplastics. Both topics have had numerous applications dedicated to them and will be discussed in detail below.

The need for certified reference materials has not diminished because they are the best way of assessing accuracy and method validation. A paper by Gao et al. reported the production of polymer reference materials possessing high homogeneity that had been produced using 3D printing.99 Solutions of Cd, Cr and Pb were gravimetrically doped into polyacrylate resin and then mixed. The 3D process then cured the samples. When tested, the materials were found to have retained all of the analytes and LA-ICP-MS was then used to determine the homogeneity. Using line scans the laser spot size could be as little as 50 μm for the Cd and Cr and as small as 14 μm for Pb and still no micro-heterogeneity was observed. The mass concentration of the analytes was determined using isotope dilution ICP-MS and were equivalent to the nominal values. It was concluded that this method of production offered significant advantages over traditional methods because it was quick and the products were more homogeneous.

Another paper reporting the preparation of reference materials suitable for the analysis of polymers was presented by Rogoll et al.100 These authors prepared a series of layered materials by dissolving metal acetylacetonates in the liquid monomer TerraGloss® UV Glanzlack 8/372 F NVK. The mixture was then placed on a glass sheet and allowed to wait for 10 minutes before being UV-cured at 365 nm for 10 minutes. A layer thickness of approximately 140 μm was obtained. For a single layer material, this was then removed from the glass sheet. For multi-layered materials, another coating was placed on top of the first and then cured for 15 minutes. The procedure continued until the requisite number of layers had been deposited. Several series of materials were prepared: some were single layer, single element materials, one material had a single layer but contained the four elements Cr, Fe, La and Zn (each at 1000 mg kg−1), a calibration series of seven materials containing Fe at concentrations between 200 and 2000 mg kg−1 and finally, a series of multi-layered materials containing alternate layers of Fe at 1000 mg kg−1 and blank layers. The materials were characterised and tested for homogeneity using a series of analytical techniques including ICP-OES, LA-ICP-MS, hand-held LIBS, μ-XRF and confocal μ-XRF. The materials were easy and quick to make and had excellent homogeneity at the micro-scale.

3.3.1 Reviews. Reviews and overviews are always useful because they offer an insight into the latest developments and work undertaken in a particular topic area. There have been three reviews pertinent to the analysis of polymers published during this review period. The use of spectroscopic techniques to aid waste management has been reviewed by Adarsh et al.101 The review contained 120 references and focussed on the techniques of LIBS, IR, Raman and LIF. All four techniques offer rapid analysis times and, when used singly or in combination and in conjunction with chemometric techniques, can offer high accuracy of identification of the polymer types. The review gave some theoretical background to the different techniques and showed schematics of the instrumental setups required. Tables were provided for the different techniques enabling easy reference for the reader. Another table compared the advantages, disadvantages and capabilities of the different techniques. A section giving insights to future prospects was also provided.

Another review, this time by Alghamdi et al. discussed the use of XRF as a tool for monitoring compliance with limits on concentrations of halogenated flame retardants in waste polymers.102 Both brominated and chlorinated organophosphate flame retardants were discussed in the review which contained 131 references. The first half of the review was spent discussing the legislation behind the manufacture and use of the materials and providing tables giving information on current legislation in different countries and examples of where and in the sample type that such compounds have been found. The second half of the review discusses the use of XRF for the analysis and includes a useful table discussing the advantages, disadvantages and actions that can help mitigate the disadvantages. A small section giving recommendations for future studies was also provided.

The third review of relevance to this section was presented by Baidurah who, with the aid of 109 references, discussed the methods of analysis for biodegradable polymers.103 The methods of analysis discussed included many of the microscopic (SEM, TEM), chromatographic (gas chromatography and derivatives), thermal (differential thermal analysis, thermogravimetric analysis) and other spectroscopic (NMR, FTIR and XRD) ones and so was by no means specialised for atomic spectrometry. However, there was also a section on XRF analysis. The review is possibly a good place to begin for a researcher just starting in the area, but it did not go into much experimental detail of the techniques.

3.3.2 Sorting of polymers for waste management. As mentioned above, one of the main areas of research has been that of waste management/polymer sorting. As well as the review by Adarsh et al.101 described previously, there has also been several other applications. Many of these have a similar structure, i.e. LIBS analysis of the materials followed by chemometric analysis of the data obtained. Included in this number was a paper by Tarai et al. who used picosecond-LIBS for the analysis and then principal component analysis (PCA) on the data to discriminate between seven different plastics types that had been collected from the garbage.104 The analysis was undertaken without first cleaning the materials and each sample comprised a total of 100 laser firings, with each spectrum being the average of 10 shots. The analytes used for the discrimination were: H. N, Na and O, although it was noted that polyvinylchloride also contained Ca. Analysis of the data using PCA indicated that more than 99% of the data were represented in the first three components and that a 3D score plot of these showed clear differentiation between many of the plastic types. The PVC samples were a very distinct group separated well from all the other polymer types. The polyethylene terephthalate (PET) samples also showed good grouping and as a whole could clearly be discriminated from other types. The other plastics also showed good grouping, but there was some overlap so discrimination was not straightforward. Despite this, the authors still concluded that the methodology showed promise.

A similar approach was adopted by Nie et al. who used LIBS incorporating a Nd:YAG laser operating at 532 nm to analyse six different plastic types of five different colours and then evaluated the data using several chemometric techniques.105 The polymers tested were: polyethylene, polypropylene, polyamide, PVC, acrylonitrile-butadiene-styrene (ABS) and polyoxymethylene (POM) and the colours were different shades of: black, yellow, blue, white and brown. Several algorithms were tested including support vector machine (SVM), PCA and neighbourhood component analysis (a method developed from the K-nearest neighbour (KNN) algorithm). A nice description of what each of these is and how they work was given in the paper. Using SVM alone, the different types of plastic could be discriminated with an accuracy of 97%. Attempts to discriminate between polymers of different colours proved more problematic using only SVM, with PVC being the worst at 82%. Attempts to improve the discrimination accuracy for PVC were then made using the other algorithms in combination with SVM. The PCA-SVM increased accuracy to 86% and NCA-SVM improving it further to 91%. Unsurprisingly, this led the authors to conclude that the chemometric methods when used in conjunction with LIBS could differentiate between different polymer types even when they were coloured.

Adarsh et al. used both LIBS and Raman spectroscopy to obtain the experimental data for both atoms and molecules.106 The same instrumentation was used for the LIBS and Raman data collection and this was described in detail in the paper. Data collection for the two could not be simultaneous because the operating conditions required were very different. The same spectrometer and the same Nd:YAG laser operating at 532 nm was used for both, but the LIBS required a pulse energy of 7 mJ and a collection time of 10 ms. The Raman spectroscopy required a much lower energy (i.e. to prevent the sample from ablating) and used a longer collection time, typically 2 mJ for 1 s. For the Raman, there was a trade-off between speed (higher energy laser pulse requiring a shorter collection time), but at the risk of the signal being masked by the LIBS spectra or a much lower energy pulse using a much higher collection time leading to lower sample throughput and greater noise. The authors therefore undertook an optimisation study to obtain the conditions used. Output from the techniques was interrogated using PCA. This managed to differentiate between two groupings of polymers: laboratory made (polypropylene, polymethyl methacrylate, nylon-11, polycarbonate and polylactone) and real world “post-consumer” plastics (polypropylene, high density polyethylene and polyethylene terephthalate). The PCA managed to make a clear distinction between all five polymers in the laboratory group and all three in the post-consumer group. It was concluded that this system that combined the techniques is clearly a cost-effective and successful method of polymer classification.

The final waste management application used energy dispersive X-ray fluorescence and scattering (EDXRFS) to determine analytes in waste polymers as well as used lubricating oil.107 Normal EDXRF can only determine elements with an atomic number greater than 10. The new technique has the advantage of being able to determine both low and high z elements. A weak 109Cd excitation source was used to analyse three types of polymer powders (polypropylene and low and high density polyethylenes) that had been moulded into 2.5 cm pellets. Used lubricating oils as well as virgin ones that had been spiked with analytes (B, Ca, Fe, Mg, Na and Zn) were subjected to thermal degradation at temperatures of between 100 and 400 °C for 24 hours and then also analysed. Once the EDXRFS data for the polymers had been obtained, they were input to PCA and soft independent modelling of class analogy (SIMCA). This managed to differentiate polypropylene from the other polymers and density measurements separated the low and high density polyethylenes. For the oil analysis, data were input to partial least squares modelling which used the Zn signal and the scatter peak to successfully determine the viscosity of the oil. A correlation between the viscosity and the thermal degradation was also established. The methodology was deemed ideal for industrial use because of its speed and non-destructive nature.

3.3.3 Micro- and nano-particulate polymers. The other main area of research is that of the analysis of micro-plastics. These have long been suspected to act as “sinks”, i.e. preconcentration hubs for toxic trace metals and organic molecules. They therefore have the potential to be toxic to marine life and for other animals further up the food chain. When in the environment, many micro-polymers develop a biofilm. This biofilm can confound the analysis of the particle itself and so some workers remove it prior to analysis. However, this runs the risk of changing the nature of the micro-particles and possibly losing many of them during the cleaning process. Porizka et al. described the use of LIBS to analyse pristine micro-polymer particles as well as aged ones that had developed a biofilm.108 Five types of polymer were tested: polyamide, polyethylene, polyethylene terephthalate, polypropylene and PVC; and these were all “weathered” under controlled conditions in either freshwater or wastewater. Biofilm was allowed to develop and the micro-polymers were then characterised using optical and scanning electron microscopy, Raman spectroscopy, LIBS and LA-ICP-MS. Regardless of whether or not a biofilm was present, the LIBS analysis (along with PCA analysis of the data) managed to classify the polymer types successfully. The LA-ICP-MS was used to monitor the adsorption and desorption of trace analytes from the surface of the polymers. This combined approach therefore yielded significant information and, the authors hoped, could be applied to the analysis of micro-polymers in tissue samples.

Aynard et al. presented a paper in which styrene/acrylic acid copolymer particles were prepared and then characterised using SEM and AFM as well as having their surface area measurements determined.109 The particles had carboxylated surface groups and either a smooth or a “raspberry-like” surface. Using Cu as a proxy for all trace elements, the authors then stirred a suspension of the particles with a Cu solution for a period of between 24 and 144 hours at room temperature, the particles were then separated by centrifugation and washed with pure water to remove as much Cu solution as possible. Three wash cycles occurred before the particles were re-suspended and an aliquot analysed using XPS and TOF-SIMS. Both the XPS and the TOF-SIMS have a depth-profiling capability and so could determine the Cu concentrations at the surface and at the core of the particles. The XPS peak for O at the beginning of the experiment at the core differed markedly to that after Cu exposure. This was attributed to the formation of O–Cu indicating that some Cu reached the core of the particle rather than just being adsorbed to the surface. The Cu content at the surface stabilised after 24 hours, presumably due to saturation. However, the concentration in the core continued to rise throughout the duration of the experiment. This confirmed the particles ability to accumulate Cu by both adsorption and absorption.

Chen et al. developed a method entitled “in situ preconcentration for electrolyte atmospheric liquid discharge optical emission spectrometry” and then applied it to the determination of Cd on micro-plastic particles.110 Micro-plastic particles were suspended in 1% nitric acid to remove the Cd from the surface and then centrifuged to separate the particles. The supernatant liquid was then placed in a well on a graphite rod and evaporated to give a preconcentration. A discharge between the graphite rod and a tungsten rod then occurred resulting in the Cd emitting light which was detected by a CCD. Other than the centrifugation step, the entire process was contained in a portable instrument which would enable in situ measurements to be made. The paper described the optimisation of the process and the figures of merit (LOD was 1.5 ng mL−1).

Single particle ICP-MS is a technique that has developed over the last 10–15 years and has usually been applied to nanoparticles and is capable of determining their composition as well as particle size distribution. The technique was applied by Gelman et al. to the analysis of polytetrafluoroethylene (PTFE) micro-particles.111 The authors wanted to determine the F component of the plastic. However, since the first ionization potential of F is greater than that of the Ar plasma gas, it cannot be achieved directly. Instead, they achieved F detection through the molecule BaF+ at m/z 157. Using microsecond dwell times (as is the norm for single particle analysis compared with the ms dwell times for normal analyses), a series of spikes were detected in the time resolved scan with the height of the spike being proportional to the size of the micro-polymer particle. The particle size LOD was 0.93 μm and concentration LOD was 8.5 × 105 particles per L. The authors should be mindful that over a certain size, particles will be discriminated against by the nebuliser/spray chamber assembly. There will therefore be an upper LOD for particles as well. However, the mean size of the particles analysed in this study was 1.2 ± 0.1 μm. This should be well below that upper limit and was in good agreement with data obtained using dynamic light scattering (1.2 ± 0.2 μm).

3.3.4 Other applications. The use of an atomic spectrometric analysis followed by a chemometric analysis of the data to classify polymers is not confined to the waste management field. It may also be applied to the analysis of polymers that have been found in the marine environment. This could potentially enable researchers to determine how long they have been there and possibly even identify how they got there, i.e. who the polluter was. An example was presented by Giugliano et al. who used LIBS followed by PCA analysis of the data to classify beached marine polymer pellets with an accuracy of greater than 80%.112 The LIBS analysis could distinguish the C–O backbone of some polymers from the C–H of others. When combined with PCA, it also identified a correlation between the surface roughness of the pellets and the yellow discolouration. A preliminary study into the metals adsorbed to the surface of the pellets was also undertaken.

The weathering of polymers is also an indication of how long they have been in the environment. Sommer et al. described the use of LIBS to study the weathering-induced oxidation of polystyrene; some of which had been treated with the anti-oxidant Irgafos.113 The LIBS was used to examine the spectral features of oxidation, i.e. monitor the species CN, C2, H and O; with O being monitored at 777.3 nm. Since LIBS can be used for depth-profiling, it is possible to “drill” through the polymer monitoring the signals to determine the extent of the oxidation. The method is quick and can be used quantitatively. Interestingly, the presence of different concentrations of Irgarfos delayed the onset of oxidation but, after 2000 hours, the amount of oxidation was similar to the untreated polymer. Although this study was undertaken in a laboratory-controlled environment rather than on real marine plastic litter, the approach would be the same and could give valuable insights into the weathering process and speed.

Several other applications have been presented during this review period. One by Magana-Maldonado et al. used an alkaline methanolysis of PET samples and then determined the Sb extracted using hydride generation introduction to a microwave plasma-optical emission spectroscopy instrument.114 The extraction process was discussed in detail and involved adding methanol, 5% potassium hydroxide, 0.5% sodium methoxide and a small amount of calcium carbide to approximately 10 mg of PET. This was then heated at 60 °C for 24 hours, cooled, 500 μL of water added, mixed, 200 μL of 8 M HCl added, centrifuged, the supernatant washed with 3 × 1 mL of 1 M hydrochloric acid, the washings combined and diluted to 25 mL using 1 M hydrochloric acid containing 10% methanol. Calibration standards were prepared in exactly the same way, but using 10 mg of PET known to contain no Sb. The sample was spiked with the relevant amount of Sb at the beginning of the extraction process. Data obtained from two emission wavelengths (217.581 nm and 217.919 nm) were exported from the instrumental software and uploaded to the Unscrambler software which then performed partial least squares regression. All extractions were completed in triplicate and method validation was through the method of spiking experiments which yielded recoveries of 93.8–99.3%. Although the experiment appears successful, the requirement of a complex 24 hour extraction procedure is hardly beneficial for a busy laboratory. The HG-MP-OES methodology was also applied to the analysis of water samples which, strangely, suffered worse recovery values of 68–102%.

According to Gilon et al. distinguishing between leather, synthetic leather and polymers is getting increasingly difficult.115 They therefore used a handheld LIBS instrument to distinguish the three sample types. The LIBS analysis conditions were optimised and the analytes determined included the tanning agents used for leathers (e.g. Cr or Ti), molecular bands (e.g. C2, CN and CH) and other atomic species (O and H). Analytical data were interrogated using chemometric tools such as PCA and cluster analysis. The different sample types were separated into four clear groupings representing the polymers, synthetic leathers and real leathers that had been tanned using Cr or by Ti.

The analysis and classification of polluted silicone rubber insulators using LIBS followed by an assortment of machine learning techniques was described by Sanjana et al.116 Pollutants included copper sulfate, carbon-based contaminants such as coal and fly ash, and Ca-containing materials representing materials such as cement and calcium phosphate fertiliser. Using the data obtained from the analysis of seven different groupings of samples, several machine learning techniques were used and their performance compared. Examples included linear discriminant analysis, decision tree, K-nearest neighbours and various gradient boosting techniques. The best performing algorithm was called light gradient boosting, which had a classification accuracy of 97.43% and had completed its calculations in what the authors described as “a reasonable computation time” of 5.1 s.

4 Inorganic chemicals and materials

4.1 Catalysts

This continues to be an extremely popular area of research with many hundreds of papers being produced. The vast majority report the development of the catalyst followed by its characterisation, often using XRF or some other technique capable of the analysis of solid materials directly. These routine papers may be interesting from a catalysts perspective, but bring no novelty with regard to the atomic spectrometry. Consequently, they will not be discussed in this review. Those papers that do offer some novelty, e.g. those that undertake on-line analysis, will be discussed because they offer potential advantages for the industry in terms of increased throughput or greater efficiency.

The slow oxygen evolution reaction (OER) requires a high overpotential to achieve relevant current densities (>2 A cm−2) even with a high loading of iridium oxide. Using a simple commercial 1,5-cyclooctadiene iridium chloride dimer precursor, Krivina et al. synthesised submonolayer-thick IrOx on the surfaces of conductive metal oxides.117 This was in an attempt to make every Ir atom available for catalysis and reach the ultimate lower limit for Ir loading, i.e. to reduce costs. The reaction on Sb/SnO2 and F/SnO2 conductive oxides is surface-limited and a continuous Ir–O–Ir network provided improved stability and activity. The IrOx was covered with a thin layer of acid-stable TiOx using atomic-layer deposition. The effects of TiOx on the catalyst's performance were assessed using ICP-MS coupled in situ with an electrochemical flow cell and ex situ using XPS. Tuning the binding environment of IrOx with TiOx addition enhances the intrinsic activity of the active sites while also accelerating the dissolution of the catalyst and the metal-oxide support. The interaction between the support, the catalyst and the protection-layer dissolution with OER activity was shown and the effects of annealing to densify the TiOx protection layer on stability/activity was highlighted. These ultrathin supported Ir-based catalysts do not eliminate the long-standing issue of the catalyst and support instability during OER in acids, but do provide new insight into the catalyst–support interactions and may also be able to assist in the elucidation of the OER mechanism.

Another paper to investigate catalysts for the OER was presented by Ko et al.118 The introduction of nickel to ruthenium oxide catalysts is a promising approach to enhancing the efficiency of the OER. However, since nickel has a poor activity, the mechanism through which enhancement occurs had not been fully understood. The authors prepared a ruthenium nickel oxide electrode through a modified dip-coating method and then characterised it using XRD, XPS and near edge X-ray absorption fine structure (NEXAFS) spectroscopy. The electrode offered excellent OER performance in acidic media and for carbon dioxide reduction in neutral media. An in situ/operando XANES experiment was conducted using a home-made electrochemical flow-through cell along with an on-line ICP-MS method. These studies enabled an understanding of the role the nickel plays in OER enhancement. The nickel transforms the electronic structure of the ruthenium oxide and produces a larger number of oxygen vacancies by distorting the oxygen lattice structure at low overpotentials. This increased the participation of lattice oxygen to the OER. Having elucidated the mechanism of enhancement, the authors speculated that it could be used in the future to design better and more efficient electrocatalysts.

Another paper to investigate the OER was presented by Zlatar et al. who, rather than attempting to improve the activity or selectivity, focussed on the stability of the catalysts.119 An on-line ICP-MS method was used to study an Ir single atom catalyst and a highly dispersed ruthenium catalyst on a support of indium tin oxide. First, commercial iridium oxide and ruthenium oxide nanoparticulate catalysts were tested and their high activity but low stability confirmed. They then prepared catalysts using surface organometallic chemistry. These were then characterised using ICP-based methods for determining the metal loading and then cyclic voltammetry to determine how much of it is available for catalysis. The electrochemical and dissolution properties were then determined using a scanning flow cell coupled with an ICP-MS instrument. Based on the dissolution insights gained from this, the observed Tafel slope increase was linked to a change in the OER mechanism from an adsorbate evolutionary mechanism to a predominantly lattice oxygen mechanism. The latter results in increased dissolution of the support and further destabilization of Ir and Ru. The authors then provided guidelines and solutions to some of the future challenges which may arise while assessing the stability and activity of single-atom and highly dispersed catalysts.

Improving the electrocatalyst stability was also the focus of the paper presented by Kreider et al.120 An online electrochemical flow cell was coupled with an ICP-MS instrument so that the impact on the composition and reactant gas on the multi-element dissolution of manganese chromium antimonate electrocatalysts could be determined. The presence of the Sb framework stabilised the Mn at the potentials used for the OER. It also stabilised the Cr for the OER and the oxygen reduction reaction (ORR). Dissolution of the Mn and Cr from the manganese chromium antimonate catalysts was governed by the ORR reaction rate. This was demonstrated by almost completely stopping the reaction by conducting experiments under nitrogen. Using a potential of 0.3, 0.6 and 0.9 V versus a reversible hydrogen electrode for 10 min, a total of 13% of the Cr was dissolved compared with only 1.5% for the Mn. This left the surface relatively Mn-rich. Operando XAS was applied and indicated no change in the Mn K-alpha edge at comparable potentials. This could be because the change in Mn oxidation state is too small to be detected, or that the layer is too thin for it to be measured using XAS. The equivalent experiments conducted using the on-line ICP-MS measurement indicated a dissolution rate of the Cr of 20 ng s−1 cm−2 for the first 5 min, but then decreased to less than 5 ng s−1 cm−2 over the course of the 30 min experiment. A similar profile was observed for the Mn, but at much lower concentrations whereas the Sb showed no dissolution. This paper also stated that the deeper understanding of the processes occurring within the battery would enable activity and stability to be maximised.

An octahedral bimetallic nanoparticulate catalyst (PtNi/C) doped with Rh (1.4%) and Mo (0.4%) in its surface was prepared, characterised and tested for activity for the ORR reaction in a study by Hornberger et al.121 The material was prepared using a solvothermal method using benzyl alcohol as the solvent and reducing agent in a sealed pressure flask. The full methodology was presented in the paper's supplementary files. Characterisation of the materials was undertaken using identical location scanning transmission electron microscopy (IL-STEM-EDX), operando wide angle X-ray spectroscopy (WAXS) and scanning flow cell ICP-MS. Doping the nanocatalysts with small amounts of Mo and Rh suppressed the detrimental Pt diffusion. Consequently, the material had an exceptionally high activity and stability. Unsurprisingly, it was concluded that the material prepared could potentially become a new family of shaped platinum alloy catalysts.

Another paper to attempt to improve stability was presented by Han et al.122 Water electrolysis cells and fuel cells contain platinum catalysts and unfortunately, this is well known to become oxidised and to dissolve from the cathode after time, hence decreasing the activity. This work described the preparation of a PtW/C catalyst rather than the conventional Pt/C one. The prepared material was first characterised using TEM-EDS and ICP-OES. The metal dissolution behaviour as a function of the applied potential was determined using in situ/operando ICP-MS coupled with a homemade electrochemical flow cell. The details of the operando ICP-MS system were not given in the paper, referring the reader to a previous paper. To identify the origin of oxidation tolerance for the PtW/C catalyst, ex situ XPS and in situ/operando XAFS analysis were conducted for the pristine catalyst, the catalyst after the hydrogen evolution reaction (HER) (−0.3 V vs. reversible hydrogen electrode, 15 min) and the catalyst after both HER and OER (1.5 V vs. reversible hydrogen electrode, 15 min). Based on these experiments, it was shown that W suppresses the oxidation of the Pt resulting in low cathodic Pt dissolution of the PtW/C catalyst. However, cathodic W dissolution was observed for the PtW/C catalyst. Inevitably this meant that as the W decreased, the oxidation of Pt started to increase again. However, the activity of the catalyst was as high as the normal Pt/C catalyst and the stability/active life was extended. Future work was envisaged to involve identifying a material that suppresses the Pt dissolution whilst not being oxidised itself.

Wang et al. reported the preparation and characterisation of Ni-based mixed-metal oxide catalysts for the dry reforming of methane and carbon dioxide into more useful materials (hydrogen and carbon monoxide).123 The paper guided the reader through the preparation of the catalysts, their characterisation using XRD, nitrogen adsorption, thermogravimetry, ICP-OES, TEM, XPS and XAFS and described the catalytic process. Several of the catalysts were produced, all of which were NiZnAl, but with Ni loadings of 10, 20 and 30%. The catalyst with a Ni loading of 20% was deemed optimal because the nickel particles were small (∼5–6 nm) and had by far the greatest nickel surface area. This catalyst gave conversion efficiencies of 69% for methane and 54% for carbon monoxide at temperatures of 30–60 °C. Similarly, the production efficiencies for hydrogen and carbon monoxide were 43% and 46%, respectively. These figures were comparable to the thermocatalytic process that occurs at 650 °C. Use of this new material would obviously offer a significant cost saving in terms of energy required. A further improvement was made by using a non-thermal plasma method to assist the catalyst. Again, the paper took the reader through what this was and how it was applied (an electrical discharge). When used in combination, the non-thermal plasma and the catalyst reduced the activation energy of the process by 50%, thus enabling its use at much lower temperature. The stability of the 20% nickel material was impressive, with only a 1% drop in activity over 600 min. Both XAFS and XPS indicated that metallic nickel acted as the active centre of the catalyst.

4.2 Building materials

The biggest single topic in the building materials section is the determination of Cl in concretes/cements or assessing the damage done by its presence. One such paper was presented by Fernandez-Menendez et al. who developed an improved LIBS technique that enabled improved Cl determination using molecular LIBS in a noble gas-enriched environment.124 Under normal circumstances, the broadband emission from these molecules are interfered with. For instance, the CaCl emission at 593 nm from cement or concrete would normally be interfered with by the CaO band between 590 and 620 nm and possibly Na at approximately 590 nm. The work presented in this publication overcame these interference effects as well as overcoming the problems of not having a Cl-free cement matrix and the variability of Na in the samples. The use of an atmosphere enriched in either Ar or He was tested and the Ar atmosphere improved the sensitivity of the CaCl emission as well as decreasing the intensity of the Na emission. The CaO peak was also greatly diminished because of the now depleted O presence in the plasma. The method was tested successfully using several real cement samples. The Cl content was between 0.23 and 1.5% of the sample.

Another LIBS method of determining Cl in cement pastes was described by Zhang et al.125 In this study, collinear dual-pulse LIBS was employed in which two nanosecond lasers with a total energy of 30 mJ were used to improve the sensitivity of Cl determination. This paper also used an inert gas atmosphere (this time He) at a flow rate of 4 L min−1, an inter-pulse delay between the two laser firings of 2000 ns, a gate delay of 800 ns, a pulse energy (mJ) ratio of 19[thin space (1/6-em)]:[thin space (1/6-em)]11 and a distance from sample to collection lens of 42.8 mm. Under optimal conditions, the signal to noise ratio for the Cl emission at 837.6 nm improved from 1.75 to 2.68 for a cement sample containing 0.706% Cl. The Saha–Boltzmann plot was used for calculating plasma temperatures and this was dependent on the laser energy, and other parameters of the LIBS system. It was also found to be related to the signal to noise ratio of the Cl line. Sixteen calibration standards were prepared by mixing cement pastes with sodium chloride solution of varying concentration. Two calibration strategies were tested: internal standardisation and partial least squares regression. The LOD when using the internal standardisation was 103.4 mg kg−1. The prediction performance was evaluated using the leave one out cross validation methodology and was 0.0910 and 0.0859 for internal standardisation and partial least squares regression, respectively.

Digital fabrication of concrete has increased over the last few years and although some problems have been resolved, there are others that have received virtually no attention. The production procedure leads to a layered structure with some of the layer interfaces being quite weak. These weak interfaces are called cold joints and they can be channels for aggressive agents such as chlorides. Bran-Anleu et al. used micro-XRF to monitor Cl ingress through 3D-printed, fine-grained concrete samples, each of which had been prepared with a different time interval between layer deposition.126 The results were compared with those obtained using neutron imaging of the moisture uptake. Cold joints formed with a 24 hour time interval between layer deposition were highly susceptible to Cl ingress. The curing conditions were also found to have a significant effect on the transport rate along these interface channels. The micro-XRF was deemed the ideal technique for this study because it has high resolution and may be used directly, i.e. is relatively non-destructive.

An interesting application of LIBS was presented by Chang et al. who used it for the classification of end of life concrete into recycled coarse aggregate and recycled fine aggregate.127 This is important because using concrete waste to produce aggregates is recognised as being a sustainable way of satisfying growing demand for concrete. Also determined were eight contaminants, the presence of which, even at relatively low concentration, can degrade the quality of the materials, the LIBS process and data analysis using cluster-based algorithms (Principal Component Analysis (PCA) and Chi square distribution) took place as the materials pass by on a conveyor belt. The paper discussed the mathematics and the methodology in detail. The Nd:YAG laser fired at a frequency of 100 Hz and with an energy of 170 mJ per pulse. The conveyor belt could move as fast as 50 cm s−1 but for the experiment was moved at 20 cm s−1. This means that the laser fired every 2 mm on the sample stream. The model developed proved to be very successful, with an F1 value of 0.98 being achieved (where an F1 value of 1 would indicate a perfect classification performance). This on-line method clearly offers a quick and accurate method of analysis.

Another paper to have classified concrete types was presented by Nedeljkovic et al. who used handheld XRF for the task.128 The method was tested on a range of different raw powders and on concretes containing different types of cements. Examples included CEM I 42.5 (Portland cement), CEM II/B-V 42.5 N (Portland fly ash cement) and CEM III/B 42.5 N (GGBFS cement). The results from the handheld instrument methodology were validated using conventional desktop instrumentation and energy dispersive spectrometry. Curing of the concrete affects the results significantly because a dried concrete surface is optimal since it limits the impact of surface moisture and efflorescence on characteristic oxides such as CaO. The CEM III/B 42.5 N concrete surface was easily distinguished from the others using Al2O3, MgO and Fe2O3 as characteristic oxides. The core contained enriched SiO2. It was concluded that, as long as the surface of the materials are dry and that characteristic oxides are defined well, the handheld XRF instrument can classify different concretes. The handheld instrument had the advantage of speed since measurement time was only 30 s and had the capability of undertaking measurements in situ.

The American Standard Testing Method (ASTM) was compared with the Japanese Standard Association equivalent as well as a portable XRF instrumental method for the determination of SO3 in slag aggregates.129 The measurement is important because high levels can result in significant loss of strength of the concrete produced from the raw materials. This study, by Aljobeh et al.,129 analysed 27 specimens of slag aggregates which were composed of 5 different types: blast furnace, basic oxygen furnace, electric arc furnace, old bank and open hearth. The data produced using the ASTM method and those from the portable XRF instrument were in pretty good agreement. However, the data obtained using the Japanese standard method were significantly higher. This was attributed to the complexity of performing it. Given that the ASTM and XRF data were in agreement, the authors recommended the Japanese standard method not be used.

Several other applications of the analysis of building materials have also been presented. Included in this number was one by Do et al. who developed a hydrochloric acid-free solid phases extraction with ICP-MS/MS detection for the monitoring of concrete rubble for the radionuclide 126Sn.130 Following the Fukushima accident, any building rubble has to be monitored so that it can be treated appropriately. An extraction process involving nitric acid, several evaporation steps, filtrations and residue uptakes led the analyte to be extracted. Once extracted, it was retained on TEVA® resin enabling the matrix components to be removed and a preconcentration to occur. The method was optimised and characterised and indicated that 95% of the Sn from the concrete was recovered. Interferences from the isobar 126Te and potential polyatomic interferences were removed using the ICP-MS/MS instrument's reaction cell. The decontamination factor for 126Te was 105 indicating almost complete removal. When the 126Sn was measured at m/z 160, the LOD was 12.1 pg g−1, which is equivalent to 6.1 mBq g−1. It was concluded that this relatively rapid approach yielded sensitivity comparable to that using conventional radiation-counting methods and is sufficient for verifying the presence of 126Sn in the concrete, i.e. it was fit for purpose.

An environmentally friendly method of determining Fe, Mn, Ni and Pb in construction materials was reported by Gomez-Nieto et al.131 Instead of embarking on a time-consuming acid dissolution/extraction procedure that uses large volumes of environmentally unfriendly and caustic materials, they used slurry sampling prior to analysis using continuum source (CS)-GFAAS. Stable slurries were prepared by suspending 10 mg of ground solid material into 10 mL of 1% (v/v) Triton X-100 and 1% (v/v) nitric acid solution and then sonicating for a minute. Analysis of this slurry was then undertaken in three stages, with the Fe and Ni being detected simultaneously and the other two analytes determined individually. Since the samples could potentially contain significant amounts of these analytes, the wavelengths used were often not the most sensitive. Instead, the Fe line at 232.036 nm was used (1.4% of the sensitivity of the main wavelength), Mn at 403.080 nm (6.7%), Ni at 232.003 nm (the primary line) and Pb at 283.06 nm (42%). The pyrolysis and atomise temperatures as well as the matrix modifier composition were optimised taking into account both aqueous standards and slurries. Under optimal conditions, the working ranges for the analytes were: Fe = 0.15–60 mg g−1, Mn = 2–600 μg g−1, Ni = 4–75 μg g−1 and Pb = 1.5–80 μg g−1. The methodology was validated in two ways: the analysis of a fly ash certified reference material prepared and analysed in the same way as described above and CS-GFAAS following an acid dissolution. The Student's t-test indicated that there was no significant difference between the methods, implying that the developed method was accurate and suitable for the task it was developed for.

4.3 Inorganic chemicals

Several nice applications have been presented covering different sample types and analysis techniques. Bunina et al. described a method for the determination of As in Gadobutrol, a gadolinium-based magnetic resonance imaging contrast agent, using cloud point extraction and ICP-OES detection.132 Arsenic was oxidised to its +5 state using either chloramine T (in 24 hours) or a 10-fold excess of potassium peroxo monosulfate (in 1 hour) and the AsV then complexed with ammonium molybdate in the presence of the surfactant Triton X-100. Optimal conditions in terms of the pH (2.5), the concentration of Triton X-100 (2 g L−1), etc. were all optimised. Phase separation was achieved through a salting out effect utilising the Triton X-100 and ammonium sulfate (0.2 g mL−1). The figures of merit for the procedure were: working range = 0.01–50 mg L−1, precision for 0.5 mg L−1 (n = 3) = 1.26% RSD, a five-fold preconcentration was obtained and LOD = 0.6 μg L−1. An interference study indicated that the worst problem was caused by a 200[thin space (1/6-em)]:[thin space (1/6-em)]1 ratio of Ca to As, which still yielded a recovery of 92.1%. The authors concluded that this was close to being interference-free.

Two papers have described the analysis of salt. In one by Park et al. LIBS analysis was followed by a two-stage partial least squares discriminant analysis (PLSDA) to classify six commercial edible sea salt samples from three different countries (Japan, South Korea and France).133 Although this is a fairly common approach during classification studies, this one had some novelty. In this study, the data were first sorted by the PLSDA into four classes and one extended class. This extended class represented the combination of the two unclassified groups of materials, i.e. the ones that could not be classified easily. The data from this extended group then underwent a second round of PLSDA analysis which managed to split the extended group into the other two classes. By maximising the differences between the two confusing classes, the authors improved the accuracy the PLSDA significantly.

The other paper to analyse salt was presented by Ding et al. who used solid phase extraction to obtain a matrix removal and preconcentration stage prior to GFAAS detection of Pb.134 Polystyrene microspheres were used as carriers for a chitosan coated polystyrene. The material was synthesised by sulfonating with sulfonyl chloride, coating with chitosan and then cross-linking with glutaraldehyde. Once prepared, the material was characterised using FT-IR, SEM, XRD and thermal analysis. A column of the material was prepared and the sample flowed though. The matrix was removed using 0.1 mol L−1 HCl and, when necessary, the Pb2+ was eluted using 2 mmol L−1 sodium EDTA. Aliquots of the eluted material were then analysed using GFAAS. Recovery values of spikes were between 101.7 and 106.45%. The material could remove the sodium matrix efficiently and the presence of other cations at relatively low concentrations did not cause problems through occupation of the active sites.

Brines are used for several industrial processes and their analysis can be challenging because of the high concentration of dissolved salts. The determination of K, Mg and Na in brine using liquid cathode glow discharge optical emission spectrometry was described by Ling et al.135 Under normal circumstances, a severe matrix effect can have a deleterious effect on the accuracy of this methodology. However, Ling et al. devised a correction method that was based on the mutual interference between the three analytes that improved the accuracy of the analysis. In particular, this study focussed on ways of alleviating the Mg interferences on the determination of K and Na. The mathematics of the correction were based on the Lomakin-Scherbe formula and was explained at length in the paper. A schematic diagram of the instrumentation was provided to aid the reader. Operating conditions were optimised, e.g. the voltage was varied between 500 and 550 V (below that the plasma did not discharge normally and above that the anode became hot and caused instability). An operating voltage of 540 V was chosen, with a flow rate of 1.4 mL min−1. The performance of the developed method was evaluated by analysing four brine samples and comparing the results with those obtained using ICP-OES. The relative errors for K and Na were between 1.79 and 8.65% which were a significant improvement on those where the correction was not used and were comparable to those from ICP-OES which were 3.88 to 6.49%. The Mg was also determined successfully but required the use of the internal standard Al.

An interesting modification to ETV-ICP-MS was described by Silva et al. who determined Hg in bioresorbable calcium phosphate.136 Up to 10 mg of material could be analysed at a time by placing it in an ETV device that comprised a halogen lamp enclosed in a glass chamber. The halogen lamp raised the temperature to 650 °C for a period of 10 s enabling the Hg vapour to be removed from the sample and carried to the plasma in a gas flow. A cooling down period of 120 s followed prior to the next replicate/sample being analysed. Calibration was achieved through the analysis of aqueous standards. For these, the temperature programme had to be modified slightly to ensure that the liquid could be dried. Therefore, an extra step of heating for 20 s at 80 °C was applied prior to the vaporisation step. The authors estimated that 95% of the solid sample remained in the lamp’s dimpled cavity after analysis thereby reducing the potential interferences and matrix effects to a minimum. Method validation was achieved through the successful analysis of the CRM MESS-3 (a sediment) and by dissolution of the material followed by conventional nebulisation into the ICP-MS instrument. The method had several advantages including that of speed (a sample could be analysed in less than three minutes), a much improved LOD compared with conventional nebulisation (0.2 ng g−1 compared with 60 ng g−1) and, as stated above, was interference-free. The one drawback was that it did not have a very good precision (better than 15% RSD). This is common to many solid sampling ETV applications and arises because of the difficulty in weighing small masses accurately and transferring them to the atomisation device.

A paper by Guselnikova et al. reported the use of a collision cell ICP-MS instrument for the analysis of high purity germanium and germanium dioxide.137 The paper discussed how polyatomic interferences can be so disruptive to analyses and then presented a study of the interference effects arising from the germanium matrix and how they could be minimised. The effects of parameters such as the helium flow rate and the cell voltage were optimised so that their effects on analyte signal to background ratios and LOD were maximised. The effects of several polyatomic ions could be alleviated including some that did not originate from the germanium matrix. These interferences included but were not limited to 35Cl16O+, 40Ar16OH+, 38Ar18OH+, 70GeH+, 40Ar35Cl+, 73Ge16O+ and 74Ge16O+ which would normally interfere with 51V+, 57Fe+. 57Fe+, 71Ga+, 75As+, 89Y+ and 90Zr+, respectively. Analyte LOD were in the range 0.05–1 μg g−1 and those for Ni, Se and Sr were improved by an order of magnitude compared with those obtained using the instrument's standard mode. In the absence of a suitable CRM, the authors validated their methodology using spiking experiments and by comparing the results obtained using their methodology with those obtained using direct current arc optical emission spectrometry.

Cui et al. had the objective of providing a less damaging method of analysis of yttrium-doped barium fluoride crystals and therefore developed an LA-ICP-MS protocol.138 The LA methodology tends to be minimally damaging in that it produces craters in the sample surface at the micron level and can therefore often not be seen by the naked eye. Using the methodology they had developed, the authors discovered that the Y content of the material increased further away from the seeding end. Using the increased sensitivity obtained through analysing a solid material directly rather than performing a dissolution procedure, the authors also undertook a study of trace contaminants. Calibration curves for each element were achieved by making pressed pellets of barium fluoride powder (4 g) with known volumes of 5 mg mL−1 Y (0.3, 0.5, 1, 2, 4 and 6 mL) and of different volumes of a 100 μg mL−1 mixed standard. Regression coefficients of between 0.9918 and 0.9995 were achieved. The LOD obtained were: 0.05, 0.01, 0.01 and 0.03 μg g−1 for Mn, Pb, Sr and Zn, respectively. The data obtained using LA-ICP-MS were in agreement with those obtained using a wet chemical dissolution followed by ICP-MS/ICP-OES analysis, indicating the accuracy of the developed method.

In common with many other sections of this review, the tandem use of LIBS and Raman spectroscopy in conjunction with chemometric analysis of the data has been used to good effect. Pinson et al. used the techniques in combination with data fusion and a machine learning algorithm to investigate the reactions that occur when lithium hydride or lithium hydroxide react with moisture or carbon dioxide.139 These compounds have several industrial uses (e.g. batteries or nuclear applications) but once they react they form other molecules that do not function so well. The study included “weathering” experiments undertaken in an environmental chamber. The analytical system employed a pulse laser and an echelle spectrometer in a novel setup that enabled the LIBS and Raman measurements to be performed at the same time. The data from the techniques were used to “train” and test the model so that predictions of what happens to the compounds may be made.

4.4 Ceramics and refractories

The large majority of applications for the analysis of ceramics have been associated with historical ceramic samples and are consequently covered in the heritage section of this review. The reader is directed there for those applications. In truth, there has been very little novel work published for the analysis of industrial ceramics. Given the difficulty of bringing the materials into a liquid form so that they may readily be analysed using nebulisation into standard atomic spectrometric instrumentation, the focus has been on the analysis of the solid materials directly.

The most common method of analysis during this review period has been LIBS. Although there can be problems in quantifying analytes using this technique because the signal is very dependent on sample matrix, it has still found application for the determination of the major and some trace analytes. A paper by Zhang et al. described the use of a fiber-laser-based desktop LIBS instrument to analyse the raw materials for ceramics quantitatively.140 The eight oxides Al2O3, CaO, Fe2O3, K2O, MgO, Na2O, SiO2 and TiO2 were determined in the geological samples as well as the loss on ignition. For data processing, the authors used a combination of continuous background deduction and full width at half maximum intensity integral and spectral sum normalisation. This approach improved the mean absolute errors from between 0.10% (for Ti) and 12.06% (for Al) when no data processing was used to 0.05% and 9.48%. This multivariate approach was extended further leading to an additional improvement with the value of the mean absolute error for many analytes including Al which reduced to 2.07%. These error levels then become sufficiently low for LIBS to be used for the analysis of the raw materials used to make building ceramics.

Gadolinium-doped ceria is used for the electrolytes and electrodes in solid oxide fuel cells. Park et al. investigated the use of a compact low-power diode-pumped laser combined with a hand-held spectrometer for the LIBS analysis of these materials.141 Over the spectral region monitored (397–450 nm), there were numerous very intense Ce and Gd emission lines. The low resolution hand-held spectrometer was not capable of resolving the individual lines. However, using a multivariate approach (partial least squares regression), useful information was extracted from the unresolved spectra, enabling mole fractions of Gd in the ceria to be determined. It was concluded that the very low cost instrumentation when used with the partial least squares regression calibration was a very adequate and cost effective method for this analysis.

Analyses using LIBS are not always as sensitive as hoped. One method used to increase the sensitivity is that of microwave enhancement. Ikeda et al. used microwave enhanced LIBS for the analysis of alumina.142 As the authors pointed out, if microwaves enhance the signal from a single laser shot, then there is less need for multiple shots and hence less material is ablated. This is of particular importance when radioactive substances or substances contaminated with radioactivity are being investigated. Ikeda and colleagues tested this hypothesis by analysing the interaction of the microwaves with the alumina surface with the aid of a time-series set of images of the plasma evolution. The crater characteristics (diameter and depth) and the temporal temperature of the plume were measured to evaluate the non-equilibrium plasma characteristics. No direct correlation between the ablation crater depth and volume was observed with the microwaves. In contrast, it was the number of laser shots that governed crater volume. Since absorption of the microwaves by the plasma took 100 μs after the laser shot, it was concluded that the microwaves had no influence at all on the ablation process but instead, further excited the already partially excited plasma plume of sample.

The final LIBS application of the analysis of ceramics worthy of note was presented by Wang and Lu.143 These authors used LIBS to analyse 40 ceramic samples used in the home in an attempt to distinguish between those that were high-priced goods and those that were low-priced but being used as counterfeit and being sold as high-priced. The materials were first cleaned, dried and then broken into fragments to ensure that any coating or glaze is not present on the part to be analysed. A curved or uneven surface can be problematic for LIBS analyses and so a fragment with a flat surface was also selected to ensure the highest accuracy. Some of the ceramic samples of known origin were used as a training set for the generalised regression neural network software used to conduct the analysis. A total of 28 wavelengths were chosen covering numerous analytes. Abnormal data were first removed using Mahalanobis distance to ensure that it did not confound the regression software. Then, principal component analysis (PCA) was used to reduce the dataset to a more manageable level, producing four principal components. These four components accounted for more than 95% of the data and were therefore considered representative of the entire set. After using Mahalanobis distance to remove errant data, the accuracy of the regression improved from 94.5% to 97.44%. This increased further to 100% once the PCA had been used to simplify the data. This optimised method offered significant advantages over using the “whole spectrum” approach. These advantages were that prediction accuracy was far better and a much reduced time of analysis (by a factor of 45) was achieved.

An alternative solid analysis method to LIBS is LA-ICP-MS. This has been used by Spanu et al. for the analysis of ultrapure (99.999% pure) silicon carbide.144 Wet chemical digestions for such a sample are extremely difficult and, given the high purity of the material, are prone to contamination arising from acids or from digestion vessels. The solid analysis approach circumvents these problems. Several approaches were tested with the best results being obtained when silicon carbide powder or grains were embedded in epoxy resin. No grinding, pressing or sintering was required, thereby decreasing the time required for analysis and minimising opportunities for contamination. This is especially true for the grinding because of the sample material being much harder than the grinding medium. The LA-ICP-MS operating conditions were optimised to ensure that the amount of sample ablated was sufficient for analysis but that it was not lost whilst being transported to the plasma or introduced to the plasma in large lumps. Increasing the fluence of the laser increased the noise of the signals, hence degrading the LOD. This was thought to be the result of larger particles entering the plasma. The wavelength at which the ArF laser operated was also optimised, with 266 nm producing sharper analyte peaks and improved sensitivity compared with 193 nm. Each sample spot was ablated three times, the first two to clean it and the third for quantitative measurement. The signal from 28Si+ was used to normalise the analyte signals. Overall, the precision was adequate with only a couple of elements having a precision of worse than 5% RSD. In general there was reasonable agreement with results obtained using GD-MS. Those analytes whose results were not in agreement when analysed using the two methods were often present at concentrations close to the LOD.

Arkhipenko et al. described an arc-based atomic emission method for the analysis of cerium oxide.145 The instrument used has two high resolution polychromators: one with a range of 195–350 nm and the other 350–700 nm. The high resolution offered by the polychromators enabled the large majority of the 15 rare earth elements and 19 other analytes to be determined interference-free, despite the exceptionally line-rich spectrum produced by the sample matrix. For those analytes that suffered from interferences, a manual correction could be made using the instrument's software. The work involved the optimisation of the operating conditions such as the shape and size of the electrodes (a lengthy description given), distance between electrodes (3 mm), the ratio of sample to graphite powder (5[thin space (1/6-em)]:[thin space (1/6-em)]1) and the carrier material. Complete evaporation of the sample (and hence measurement time) took only 120 seconds although this could be reduced to 100 s if the REE were not determined. Results obtained had a reasonable precision and were in agreement with those obtained using a standard method. It was concluded that the method was a big improvement on arc-based methods from the 1970s and that using this instrument was capable of meeting the requirements of this type of analysis.

4.5 Glass

It should be emphasised that many of the applications for the analysis of glasses are undertaken on historical samples and may consequently be found in the heritage section of this review. Alternatively, forensic applications of glass may be found in the forensic section of the review. The reader is therefore directed to those sections for those applications.

A paper by Carlson and Hervig discussed the analysis of some glass standard samples (both basaltic (i.e. geological in origin) and soda-lime silicate) using SIMS with varying internal standards.146 Although it is common to use one of the Si+ isotopes as an internal standard, these authors point out that some work has used Si2+ or even Si3+ for that task. Using B, Be and Li as test analytes, they therefore undertook an in-depth study to determine if all three internal standardisation methods are equally applicable to different glass types. It was noted that when using Si+ as an internal standard, the accuracy did not differ significantly when sample matrix changed. However, this was not the case when the multiple charged Si ions were used, with signals deviating by up to two-fold. Use of the multiple charged Al3+ as an internal standard yielded significant improvements in accuracy compared with the multiple charged Si ions. The mechanism of why the different Si ions have such an effect was then investigated by using 16O+ as an internal standard, with the O originating both from the sample and from the primary ion beam. Two very distinct calibrations were achieved for the two sample types which led the authors to conclude that the sputter yield (i.e. the ions ejected per primary ion impact) was the governing factor.

Given the difficulty of dissolving glass for a typical ICP-OES or ICP-MS analysis, it is common for a solid sampling/analysis approach to be adopted. This often involves the use of XRF spectrometry or LA-ICP-MS measurements. A problem with the LA-ICP-MS approach has been standardisation, with differing materials offering different responses because of different ablation rates arising from different sample melting points, etc. A paper by Michaliszyn et al. has attempted to address this by combining standard addition and isotopic dilution to improve SI traceability.147 From the trace elemental composition of the sample and from using SI traceable solutions, it is possible to calculate an uncertainty budget for an analyte. These authors used two concentration series – one for the analyte (B) and the other for the reference element, with these solutions having varied isotopic ratios. The different solutions were introduced in turn with ablated material from the sample (NIST 612) being introduced and mixed via a Y-piece connection. Compared with the certified value of 32 μg g−1 a result of 33.3 ± 3.7 μg g−1 was obtained, which the authors concluded as being an improvement on previous work. Since the calculations required are a ratio between the different isotopes, the authors pointed out that even if there is a partial blockage in the system, the accuracy would not be affected.

The use of NIST reference glasses as calibrants or as quality control samples for geological glasses or ores is commonplace but, as discussed above, can lead to problems when the sample and calibrant do not have a similar matrix. Singh et al. used these NIST glasses as calibrants during the analysis of fluid inclusions from the Malanjkhand copper deposit in Central India.148 Using LA-ICP-MS as the method of analysis and by optimising the experimental conditions, the authors successfully identified the presence of Cs, K and Rb. This indicated that the fluids in the inclusions were of magmatic origin. A laser operating at 193 nm had few matrix effects and caused negligible elemental fractionation. When operated at high fluence (8.5 J cm−2), better sensitivity and accuracy, lower LOD and even less time-dependent elemental fractionation resulted. The use of the glass standards provided better accuracy than calibration using liquids containing sodium chloride.

The speciation analysis of Sn2+ and Sn4+ in glass in the presence of Fe is problematic because Fe3+ can oxidise some of the Sn2+ to Sn4+. Saijo et al. developed a method by which the Sn was extracted using a mixture of hydrochloric and hydrofluoric acids in the presence of ascorbic acid.149 When performed under a flow of nitrogen, the ascorbic acid reduced the Fe3+ to Fe2+, hence preventing the oxidation of the Sn. The Sn2+ was then complexed with diethyldithiocarbamate enabling the remaining Sn4+ to be determined using ICP-OES. Analysis of a sub-sample prior to complexation enabled total Sn to be determined which, in turn, enabled the concentration of Sn2+ to be calculated by difference. The results from the analysis of soda lime glasses using this methodology were compared with those obtained using Mössbauer spectroscopy and were in good agreement.

The formation of Au nanoparticles in Au-doped sodium zinc borate glasses was studied by Kumar et al.150 Numerous analytical techniques, including SEM, XPS and TOF-SIMS were used to characterise the nanoparticles and to help postulate a mechanism of formation. The SEM data confirmed that the particles were spherical up to a formation temperature of 550 °C. The XPS and TOF-SIMS were used in an ultra-high vacuum to study the chemical state and the particles' thermal stability, respectively. The particles were formed near the surface of the glass and then chemical composition changes occurred. The anti-microbial and anti-fungal properties of the glasses against different strains was tested using the disk diffusion method. It was concluded that the glasses formed in this way could be explored for use in the pharmaceutical or biotechnology industries because of their anti-bacterial properties.

The analysis of glass surfaces or of diffusion within glass has been the subject of several studies. An example, by Mishra et al. used TOF-SIMS and energy dispersive microanalysis to investigate the corrosion behaviour of freshly broken soda lime silicate float glass.151 Over a duration of 0–30 days, the Na concentration at the surface decreased for two of the float glasses. However, for the float glass that was deficient in K and Mg, this was not the case. This was contrary to the findings for Si, which demonstrated an increasing concentration at the surface for the same two glasses over the same duration. The Ca concentration at the surface remained reasonably constant. Active nucleation sites for the corrosion were also observed on the surface. Analysis using TOF-SIMS indicated the presence of numerous Na degradation compounds, including Na2+, Na2O+, Na2OH+ as well as other Ca-based ones. Cycles of heating and cooling accelerated the process. The authors used the results to formulate a mechanism. Another paper reporting the analysis of the surface of a boroaluminosilicate glass was published by Agnello et al.152 These workers used different concentrations of hydrofluoric acid (0.5–15 M) as well as different concentrations and ratios of hydrochloric/hydrofluoric acid on the surface of the glass Corning Eagle XG®. They then used numerous techniques to analyse the surface. Included in this number was AFM for nanoscale topography measurements (the roughness) and TOF-SIMS for shallow depth-profile analysis. The TOF-SIMS analysis focussed on measuring the hydroxyl group concentration and, according to the authors, was an improvement over other similar methodologies in terms of the vacuum chamber gas used, the amount of physisorbed water present and the surface contamination. Unfortunately, this aspect of the study was not discussed in detail, with the authors stating that it would be in another publication. It was no surprise to discover that the surface of the glass after treatment varied with the concentration of the acid(s) used for etching. Possibly more surprisingly, the surface differed even when the same amount of glass had been dissolved. The mechanisms of dissolution were discussed in detail.

A paper entitled “insights on surface analysis techniques to study glass primary packaging” was presented by Pintori et al.153 During the manufacture of a vial from a glass tube, the glass is heated to temperatures of up to 1200 °C, with the viscosity being directly proportional to temperature. This high temperature can lead to loss of some of the most volatile species, e.g. alkali borates. It is therefore possible that the inner surface of the tube has higher concentrations of B and Na than the outer surface. The purpose of the study was therefore to identify possible new strategies for faster identification of factors that influence the chemical resistance of pharmaceutical glasses. To fulfil this goal, the authors used XPS to provide quantitative information on the chemical states of elements, and SEM to analyse the topography and to identify the most suitable locations for static TOF-SIMS analysis. Angle resolved XPS enabled greater sensitivity to be obtained by simply tilting the sample slightly. This enabled the relative concentration of volatile species, e.g. B and Na, to be determined and comparison with the expected concentrations made. This approach revealed ion exchange and enrichment of ions on the surface. The SEM identified nanoscale bulges on the surface. It was at these bulges that TOF-SIMS analyses were undertaken. These analyses identified that the B and Na were clearly associated with each other, especially in the form of the ion Na2BO2+.

Thermo-diffusion of trace and even major elements may lead to their enrichment or depletion at the hot or cold ends of silicate glasses. This phenomenon was studied by Jiang et al. who used 6 mm long capsules of dried aluminosilicate glass and a temperature gradient of 38 °C per mm with the cooler end being 1070 °C rising to 1300 °C at the hot end.154 Analysis of the material was achieved using electron probe microanalysis and LA-ICP-MS. Analysis across the capsule (i.e. from side to side) suggested that re-distribution was minimal, even after 168 hours. However, it was very greatly enhanced along the thermal gradient even after only 24 hours. The major matrix component Si redistributed itself more to the hot end of the capsule whereas Al, Ba, Ca, Ga, K, Mg, Mo, Na, Pb, Sr and U tended to segregate to the cooler end. Similarly, Nb, Ta and V very much redistributed themselves towards the cooler end. The study provided direct evidence of the thermo-diffusion effects and highlighted its potential role in chemical differentiation for aluminosilicate glass materials.

4.6 Nuclear materials

A range of analytical techniques were utilised for measurement of nuclear materials, with AMS, LIBS, ICP-MS, SIMS and X-ray methods commonly seen, with some studies using a combination of techniques. Development of materials and testing of impurities for fusion reactors remains a popular research area, as does measuring difficult-to-measure medium and long-lived radionuclides to expand measurement capabilities. Accurate and precise isotopic ratio measurement is a popular and critical area for source attribution in areas including nuclear safeguards and monitoring environmental radioactivity behaviour. Novel sample preparation techniques have been developed to aid measurement, with the availability of suitable reference standards and materials key for instrument calibration and method validation.
4.6.1 Fusion. Testing of materials and reactor conditions related to fusion is a popular topic, with development of LIBS a dominant technique.

Chen et al. used beta-induced X-rays (BIXS) to investigate tritium quantification for the deuterium–tritium (DT) fuel system of a fusion reactor.155 A method was proposed for a quantitative method to directly determine tritium concentration using X-ray spectrum counts and the theoretical relationship between tritium concentration and the BIXS spectra. Non-linear response coefficients (NRCs) were used to indicate the detector response at different gas pressures for different compositions, with an estimated error of 1.4%. This moves BIXS closer to an in-line method for tritium determination.

Hydrogen isotopes were the focus of measurements by Zhang et al., using AMS.156 A novel TiH2 sample preparation method was proposed that enabled simultaneous deuterium/hydrogen (D/H) and tritium/hydrogen (T/H) ratio measurement, with a linear relationship demonstrated between measured and nominal values.

A study by Jogi et al. investigated hydrogen detection in Mo in mixtures of Ar and N using LIBS.157 As a technique, LIBS is promising for assessing fuel retention in ITER plasma-facing components during maintenance, when the reactor is filled with N or air at atmospheric pressure. The LIBS signal has been shown to improve in the presence of Ar, with this study showing the H-alpha line intensity reducing faster with increased atmospheric N background. This was explained by the higher fraction of energy consumption in dissociation and excitation processes of molecular gas and increased excited H state quenching.

A paper by Kadi et al. acknowledged the need to develop new steel alloys for fusion that could tolerate the high temperature and radiation levels.158 Ding et al. used various techniques (including SEM, XRD, TEM, ICP-OES and compressive testing) to test the properties of WTaVCrTi refractory high-entropy alloy by vacuum levitation melting for fusion application.159 The results suggested good temperature and irradiation resistance, as well as strength and hardness that is promising for fusion applications. Boronized tungsten tiles were the focus of a study by Sun et al., where boronization is a wall-conditioning procedure at tokamaks, with measurement on the first wall and divertor an important parameter for determining impurities, transportation and deposition behaviour.160 The authors used LIBS successfully to determine impurity elements in boronized W tiles removed from a tokamak, with C, Si, Ca and Cu detected and depth profiles investigated using consecutive laser pulses. The difference in impurity distribution for the three different shaped boronized W tiles was valuable in understanding plasma wall interaction processes. Impurity measurements using LIBS was also the focus of a study by Imran et al., this time using calibration-free LIBS for measuring the deposition of the impurity layer on the lower hybrid wave gas puffing pipe in the Experimental Advanced Superconducting Tokamak (EAST).161 The impurity elements C, Cu, Li, Mo, Si and W were detected in the spectra with an uneven distribution pattern. The deposition was the result of eroded materials from plasma–surface interactions with various materials.

A study by Liu et al. investigated the ablation features of in situ LIBS for quantitative diagnosis of plasma–wall interaction processes on the first wall or divertor in tokamaks.162 Parameters investigated included spot size and crater volume. The spot size was found to affect the spectral intensity significantly, with an optimum value of 1.16 mm for the observed W ionic and atomic lines at a 200 mJ laser energy. This value also resulted in maxima for the electron temperature, electron density and ablation mass. Wust et al. employed Laser Ablation Molecular Isotopic Spectroscopy (LAMIS) to characterise C isotopes on graphite using 13C isotopically-marked methane.163 The carbon deposition pattern was studied ex situ using LAMIS on Plasma Facing Components on two graphite Test Divertor Units, with clear distinction between 12C and 13C. Ultra-short collinear double pulse configurations were used, the first to generate a laser-induced plasma and the second to direct the sample into the plasma. Results were compared with those obtained using Nuclear Reaction Analysis and a material transport and plasma–surface interaction code (ERO2.0).

4.6.2 Radionuclide separation. Separation of radionuclides of interest from the starting sample material using a range of methods is an important process to facilitate accurate quantification.

Molten salt leaching was a commonly used technique. Lee et al. investigated several Sr oxides and MgCl2 molten-salts for Sr separation.164 The reaction time, temperature and salt composition were investigated, with Sr removal efficiency, structure and morphology analysed via ICP-AES, XRD and SEM. The approach was determined to be viable for Sr separation from oxides, with recoveries of 97% or above across several oxide phases (SrZrO3, SrMoO4 and U2SrOy). Wang et al. investigated the electrochemical behaviour of Sm and Pb in LiCl–KCl molten salt systems for improved utilisation of spent fuel through fissile element removal from waste Sm.165 The redox mechanism of Sm(III) ions on the liquid Pb thin film electrode was analysed, with an average extraction of 94.23% determined using ICP-AES. A similar investigation was carried out by Wang et al. for Y recovery from spent fuel from molten LiCl–KCl, this time using a liquid Ga electrode.166 Similar analyses were performed as the previous study by the same authors, with a maximum extraction rate of 99.39% quantified by ICP-AES. Separation of U from Ce, La and Sm in LiCl–KCl systems using porous Al electrodes was investigated by Wang et al.167 Almost no lanthanides were detected in the electrolysis products using ICP-AES, with a rapid and efficient separation and improved kinetic performance using an Al honeycomb electrode with regular pores compared with Al foam electrodes with irregular pores.

Lee et al. used a high temperature combustion furnace for the extraction of 129I linked to nuclear waste characterisation.168 Conditions including pyrolysis temperature, mobile phase gas, catalyst and trapping solutions were optimised for recovery, with selective extraction enabling direct measurement using ICP-MS. The procedure was further tested for simultaneous analysis of other volatile radionuclides including 3H and 14C.

High pressure ion chromatography (HPIC) was utilised for rare earth separation in post-detonation debris for nuclear forensic analysis by Bradley and Brockman.169 Quantitative recovery of seven rare earth elements (REEs) was achieved using refractory geological materials as post-detonation debris surrogates, with HPIC coupled online with ICP-MS detection. When combined offline with gamma spectroscopy, detection limits 5000–20[thin space (1/6-em)]000 times lower than quadrupole ICP-MS was achieved for stable REEs.

Saha et al. opted for cloud point extraction for trace B impurity measurement of U-based nuclear fuels.170 High acidity of the digested sample solutions was highlighted as an issue that hampers coacervation of micelles, increasing cloud point temperature and resulting in loss of B as a volatile species. A 1 + 1 mixture of 2-ethyl hexane-1,3-diol and curcumin dispersed in Triton X-114 surfactant was used, with Br–water the most effective micelle surface modifier for lowering cloud point temperature. This resulted in quantitative (>95%) B recovery. The applicability of the method was tested using two uranium dioxide and two metallic uranium samples.

Solid phase extraction combined with ICP-MS/MS was used for the monitoring of 126Sn in concrete rubble related to decommissioning and environmental radioactivity monitoring.130 A combination of separation using TEVA resin and a tandem ICP-MS/MS setup achieved an isobaric 126Te decontamination factor of 105, with Sn recoveries of >95% and an estimated detection limit in concrete of 6.1 mBq g−1.

4.6.3 Enhancing radionuclide measurement capabilities. There is an ongoing drive to expand the number of radionuclides measurable, expanding capabilities for a range of nuclear applications.

The technique of ICP-MS remains popular for rapid measurement of long-lived radionuclides. Russell et al. demonstrated how ICP-MS/MS could be used for simultaneous measurement of multiple medium and long-lived radionuclides in a single run without prior sample preparation.171 The method was tested on a range of aqueous samples and solid digests using various exemption limit criteria, which showed that radionuclides with relatively short half-lives, e.g.63Ni and 90Sr, required pre-concentration and separation prior to measurement.

Carrier et al. also used ICP-MS/MS for measurement of gaseous 129I trapped in charcoal cartridges.172 Measurement was preceded by acid digestion of cartridges and chromatographic separation, with a detection limit of 2 Bq per sample in measurements made near the La Hague reprocessing site in France. Matsueda utilised ICP-MS equipped with a Dynamic Reaction Cell to achieve separation of Pu isotopes from Am, Cm and U-based interferences using CO2 as a reaction cell gas to almost eliminate background noise intensity.173

Accelerator mass spectrometry (AMS) was a very popular technique during this review period for measuring radionuclides that other mass spectrometric and decay counting techniques struggle to quantify. He et al. assessed the performance of a home-made 300 kV AMS system for 129I and 239Pu determination.174 For 129I, an uncertainty of 2.5% and 129I/127I sensitivity of 2 × 10−14 was achieved, with uncertainties of 5% and a 0.5 fg detection limit for 239Pu. A ‘small facility’ AMS instrument was used by Michel et al. to investigate measurement of 239Pu/240Pu, along with 14C and 36Cl.175 Tests were performed in a radiochemistry lab that had handled the radionuclides of interest for the study. Measured activities were well below the release values, suggesting that ‘classical’, laborious AMS facilities may not be necessary for some radionuclide applications. Stanciu et al. used AMS for Pu isotope measurement using a 1 MV Tandetron instrument.176 Ion transport parameters were determined using radiochemically separated U and Th isotopes, followed by measurement of a Pu standards, with isotopic ratios showing good agreement with consensus values. The molecular interferences associated with Pu measurement using AMS was also discussed.

Long-term performance of an AMS routinely used for determining 26Al, 10Be and primarily 14C was the focus of a study by Kumar et al.177 International Atomic Energy Authority standards were used to assess short and long-term AMS performance for 14C in terms of reproducibility, accuracy, precision and random machine error calculation. The importance of evaluating primary standards, secondary standards, blanks and replicates of experiments was highlighted. Carbon-14 AMS measurement was also the motivation for a paper by Gwozdz et al., who analysed gas diluted reactor graphite using a gas injection system as part of characterisation and disposal of graphite waste.178 The measurement process was automated for gaseous, diluted samples using an elemental analyser and gas injection system. The accuracy of the measurements were assessed by analysing samples with known 14C concentrations.

Several other radionuclides were measured using AMS. Pavetich et al. combined NAA with AMS to measure production cross sections for 93Zr from neutron capture of stable 92Zr for astrophysical and nuclear technology applications.179 Isobaric 93Nb and the availability of a suitable 93Zr-based reference material were highlighted as the main challenges.

Sasa et al. utilised AMS to quantify 36Cl in rainwater samples collected following the accident at the Fukushima Nuclear Power Plant to calculate depositions.180 The results were correlated with 129I (r = 0.7) and 137Cs (r = 0.8), with the 36Cl released from the Fukushima accident calculated to be 1.4 ± 0.2 × 108 Bq based on the March 2011 129I/36Cl deposition ratio. Froehlich et al. aimed to measure 210Pb in 1 kg of NaI using AMS as part of a dark matter experiment.181 The focus was on developing the optimal target material for high and stable negative ion output, based on Pb mixed with a Ag binder. Of the combinations tested, PbF2 performed best with a 1[thin space (1/6-em)]:[thin space (1/6-em)]1 ratio of fluorinating agents.

Veicht et al. showed the first experimentally determined cross-sections of 26Al and 41Ca from proton-induced spallation of V targets using AMS following radiochemical separation.182 Technetium-99 and 129I were the radionuclides of interest for Fu et al.183 Iodine-129 measurement was not affected by isobaric interferences as 129Xe was not present; the focus was on the +3 and +5 states. In the case of 99Tc, the isobaric interference from 99Ru cannot be resolved by the 3 MV AMS used in this study, but suppression and ionisation was achievable during extraction in the ion source. A detection limit of 1011 atoms for 99Tc was achieved. Isobaric suppression for 99Tc was also the focus of a study by Hain et al. using Ion-Laser InterAction Mass Spectrometry.184 Of the oxide and fluoride molecules investigated for Tc and isobaric Ru suppression, TcF5 was the most promising, with a RuF5 suppression factor of 105 with a 2.33 eV laser and a further suppression factor of 100 for Ru in the ion source. A remaining challenge was suppression of stable 97Mo that would enable the use of a 97Tc isotopic spike.

4.6.4 Nuclear forensics. Accurate and precise isotopic ratio measurements remains a valuable area for determining the source of nuclear contamination. Mass spectrometric techniques were dominant, but other techniques including LIBS185,186 and X-ray based techniques were also successfully applied.187

The measurement of U and transuranic isotopic ratios was the subject of a number of studies. Bradley et al. measured solid U particulates on cotton swipes with automated microextraction ICP-MS detection.188 The extraction probe extracted particulates from the sample surface and directed them into the ICP-MS instrument using a flowing solvent. An automated stage enabled higher sample throughput than previous work, with the saved location capability on the swipes reducing user sampling error. The measured U ratios were in good agreement with samples measured using MC-ICP-MS following sample dissolution. For example, the 235U/238U and 234U/238U ratios had a percentage relative difference of −0.041% and −1.7%.

Wimpenny et al. employed LA-ICP-MS for rapid isotopic analysis of U particles.189 Two U-particle materials were tested, with percent-level precision achieved for 234U/238U, 235U/238U and 236U/238U in samples with isotopic compositions spanning >2 orders of magnitude. Results were comparable to those obtained using Large Geometry SIMS.

Accelerator MS was also used in forensic applications. Wallner et al. determined U and Pu atom ratios in air filters from Austria, following measurement of 137Cs and 241Pu (via241Am) in the early 1960s.190 A range of ratios were determined, with 233U/236U atom ratios within 0.15–0.49 × 10−2 indicating that weapons testing in the early 1960s was not the source of 233U. A compact 300 kV AMS instrument was used by Xing et al. for simultaneous transuranium (Am, Cm, Np and Pu) radionuclide determination for nuclear forensics.191 Following chemical separation, all radionuclides were measured with detection limits below femtogram and even to attogram levels for Pu and Cm isotopes.

Nuclear forensic studies using SIMS remains a popular technique. Groopman et al. investigated improved U particle analysis by SIMS using O3 primary ions.192 Newer RF plasma sources make the use of O2 and O3 possible, with O3 showing an increased ionisation yield of up to 4.7%, a factor of two improvement over O, which improves precision and detection limits. As well as improved yield, smaller corrections and reduced substrate effects were also noted relative to other negative ion beams.

Lee et al. reported on small geometry-SIMS for particle analysis of environmental samples, including U isotopics.193 The method was validated using a set of field samples, with good agreement with previously reported values. In another paper, 240Pu/239Pu was measured in micrometric Pu and mixed oxide (MOX) particles using SIMS, linked to fragments emitted following nuclear accidents.194 Hydride-based interferences from UH and PuH generated during ionisation were highlighted as key considerations, with hydride rates in UO particles measures and then applied to correct for Pu signals. The estimated 239Pu/238U atomic ratio in MOX particles was similar to 1.6 × 10−3.

The techniques of AMS, TIMS and SIMS were compared for U isotope ratio measurement in individual particles related to nuclear safeguards by Gao et al.195 Background, accuracy and efficiency were the basis of the comparison using several uranium reference material samples. The relative error and external precision was reported for a range of U isotopic ratios. Errors ranged from within 5% using fission track (FT) TIMS for 234U/235U and 234U/236U up to 20% for 234U/236U using AMS. Both AMS and SEM-TIMS were determined to have shorter measurement times than FT-TIMS but longer than SIMS, with development prospects considered for AMS and SEM-TIMS. Multiple mass spectrometric techniques were tested by Park et al. for micrometer-sized U particles, combining measurement with a micro-manipulating technique.196 The most accurate and precise measurements were achieved using SIMS when compared with SEM-TIMS and SEM-MC-ICP-MS. Following U particle dissolution and isotope dilution measurement using MC-ICP-MS, the U mass per particle was close to the reference value.

4.6.5 Reference materials. The availability of reference sources and materials have been highlighted as a limiting factor in some studies in this Nuclear materials section. Some papers have looked into the development of such materials or reclassified existing materials, utilising state-of-the-art high sensitivity analytical techniques.

Beck et al. investigated extraction resins for separation of Th from U for the 230Th/234U radiochronometer as a ‘fingerprint’ of nuclear material to support particle reference material production.197 A double-stacked TEVA resin separation method achieved quantitative U recovery, with Th below ICP-MS detection limits, whilst 0.5% Th remained in the U fraction following UTEVA resin separation.

A certification campaign for three new uranium ore concentrate CRMs was discussed in a study by LeBlanc et al.198 Fifteen laboratories from 10 countries investigated sector field and ICP-MS/MS for the determination of 64 trace element impurities, with discussions around sample digestion techniques, high-U matrix analysis, as well as data handling and uncertainty evaluation.

Nishiizumi focused on AMS standard reference materials for 10Be and 26Al, based on these being two key radionuclides for primary normalisation standards in AMS labs.199 The previous stock materials had become depleted, and a range of 10Be/Be and 26Al/Al isotopic ratios were prepared ranging from 10−13 to 10−11. Samples were measured at five AMS laboratories, with good linearity and agreement with the previous 2001 standards, leading to reference materials becoming available for the AMS community.

Mathew et al. published several papers related to reference material development. In one study, several Pu CRMs (136, 137, 138 and 126-A) were evaluated for minor isotopic data using the total evaporation methodology.200 Uncertainties were a factor of 2–3 smaller using a per turret correction compared with an assumed correction factor using the specified abundance sensitivity. Minor isotopic abundances was the focus of a second study by Mathew et al., who looked into the impact of tailing corrections on major isotopic abundance ratios.201 The corrections made were assessed against ITV-2010 values used as pass/fail criteria in proficiency test programs, using Pu, U and a separated U sample to provide measured data. A final study by Mathew et al. looked into recertification of Pu CRMs 136, 137 and 138.202 Separations prior to analysis by TIMS typically use 20 mg of Pu. This was scaled up to 200 mg, with the same columns used for 20 mg of material. A successful separation of Am and U decay products from Pu isotopes, suggested that separations masses can be scaled up without impacting the quality of the separation.

4.7 Electronic materials

An enormous number of papers has reported the development of new batteries, components thereof, thin films, light emitting diodes, etc., and have usually reported their performance. However, the number of papers that use atomic spectrometry in anything but a routine way, are relatively few. Those that do will be discussed in the three sections below.

The technique of LIBS is still not as popular for the analysis of these sample types as it is for many other. Similarly, the traditional techniques of atomic absorption, ICP-OES and ICP-MS rarely seem to be used unless in conjunction with a laser. Instead, surface analysis techniques, e.g. SIMS, TOF-SIMS, and several X-ray-based techniques are the norm along with varying forms of microscopy. Techniques such as XANES are used routinely for oxidation state studies while XAFS relates to the local chemical environment surrounding the absorber atom, e.g. provides information on coordination number (number of atomic ligands) and distance. Meanwhile, XPS will identify the elements present, their concentration and which elements they are bound to. Since some of these applications are so routine, they will not be discussed at length. Instead, the more novel aspects of the spectroscopy will be discussed. Some papers have attempted to elucidate mechanistic aspects of how their device works or how it loses its functionality. Here again, it is often X-ray-based or microscopy techniques that provide the information.

4.7.1 Wafers, thin films and multi-layer materials. One of the main techniques for analysing surfaces and thin films is TOF-SIMS. A review by Priebe and Michler (with 195 references) assessed the recent advances in Gas-Assisted Focussed Ion Beam TOF-SIMS.203 This long review started by giving the theory of TOF-SIMS and presented a Table full of its advantages and disadvantages. A description of the instrumentation available was then given. The review then went on to discuss the use of gas assisted focussed ion beam TOF-SIMS, citing papers that have used it and provided useful reference tables showing signal enhancement factors. Another table summarising its advantages and disadvantages was also given. As usual, a future perspectives section was also provided. Although very long, the review does give a lot of very useful information.

Cho and Kim provided a short tutorial review with 14 references on thin film characterisation via synchrotron experiments including the techniques of XRR-TXRF, GIWAXS and three dimensional reciprocal space mapping.204 The review summarised the abilities of each of the techniques and explained that they have high spatial and time resolution. In addition, in situ real-time changes in films may be observed using SR X-rays under external stimuli, such as heat, pressure, mechanical stress, electrical bias and electromagnetic irradiation.

Revenko et al. reviewed, with the aid of 189 references, XRF analysis of solid state films, layers and coatings.205 As well as the composition of such materials, the review also discussed the ability of the technique to determine the thickness of the layers. The review was conveniently split into sections, including: the determination of surface film densities, the use of coupling equations and theoretical ways of accounting for matrix effects, determination of the chemical composition and coating thicknesses of samples with irregular surface shape. A section on the determination of the chemical composition and film thickness using TXRF was also provided. Also discussed were the results of a round robin test in which 11 laboratories participated and provided 15 pieces of data for the thickness of several films. Both energy and wavelength dispersive instruments were used with an assortment of X-ray tubes. The relative standard deviation of the measurements were between 4.3 and 6.6%.

Another paper to report the determination of layer thickness using XRF was presented by Scialla et al.206 A standard three-layered sample comprising layers of lead, silver and gold was used and several methods of calculating the gold layer's thickness compared. These methods were: two-line ratio of one or two elements, de Boer's analytical procedure, multivariate partial least squares analysis and percentage composition of elements in the layers. For this study, a portable XRF instrument was used. All methods were adequate for estimating the thickness of gold in multilayer samples, although the results were slightly worse for gold layers of thickness greater than 0.80 μm and when the silver cover layer is greater than 0.40 μm. The partial least squares method was faster than other methods and, since the results from the different methods all had similar accuracy, this was recommended for routine use.

A calibration-free LIBS application for the quantification of each element in an aluminium gallium arsenide wafer was reported by Alrebdi et al.207 A Nd:YAG laser operating at either 532 nm (at 200 mJ) or 1064 nm (at 400 mJ) was used for the task. The LIBS spectra were recorded between 200 and 720 nm using a commercial instrument and the spectral lines of Al, As and Ga were used to calculate the electron number density (Stark broadened line method giving 6.5 × 1017 cm−3) and the temperature (Boltzmann plot method giving 5744 ± 500 K). These values were then used to calculate the concentrations of Al, As and Ga in three samples. Results were mixed. There was no doubt that the method gave reasonable “ball park” figures, but absolute accuracy was lacking. For instance, one sample had certified values of 3%, 2% and 95% for Al, As and Ga, respectively whereas the experimental values were 4.77%, 1.23% and 94%. For the Al, this would represent an overestimate of 59% and for As, an underestimate of 35%.

The analysis of surfaces or thin films has, as always, been a popular area of research. An example, by Finsgar, reported the analysis of the surface of a film of 2-mercaptobenzothiazole formed on brass by immersing it in a 3% sodium chloride solution containing 100 mg L−1 of the compound.208 The two techniques of gas cluster ion beam (GCIB)-TOF-SIMS and GCIB-XPS were used for the analysis. The operating parameters were optimised, i.e. the acceleration energies and cluster sizes were changed to remove the surface layer but leave the monoatomic information intact. The optimisation protocol for the depth-profiling started with a low acceleration and large cluster type i.e. a lower acceleration energy per Ar atom in the cluster and then changed incrementally to a higher acceleration energy and smaller clusters, i.e., a higher acceleration energy per Ar atom in the cluster. The depth profile finished with sputtering using monoatomic Ar+. The TOF-SIMS was used for 2D and 3D imaging to show the molecular and elemental distribution of the surface species. Using the tandem TOF-SIMS capability, the spectra clearly confirmed the presence of mercaptobenzothiazole on the surface. In addition, the presence of organometallic complexes were indicated, which formed between the mercaptobenzothiazole and Cu ions released through the corrosion of the brass.

A new instrument type was developed by De Castro et al. that was a hybrid of magnetic sector SIMS and focussed ion beam SEM.209 The paper noted that there is an ever increasing demand for instrumentation that can analyse samples in both two and three dimensions with very high resolution (at the nm level). The development of this instrumentation was designed to fulfil this requirement. The paper discussed the development of the instrument, its setup and provided schematic diagrams. The capabilities of the instrument may be summarised as: having a secondary ion transmission of over 40%, a mass resolving power of over 400, being capable of detecting masses between 1 and 400, possessing a lateral resolution of 15 nm (for 2D mapping) and a depth resolution of 4 nm (for depth profiling or 3D mapping). In addition, it enabled a direct correlation between SIMS data with those obtained using SEM. The paper went on to demonstrate the instrument's capabilities for undertaking 2D and 3D imaging as well as depth-profiling for a wide range of sample types (solar cells, batteries, alloys and even those from life sciences).

The most widely used method for the formation of thin films is that of sputter deposition. It does, however, suffer the problem of being very time consuming when films of multiple elements are to be formed because as each layer is formed a chemical analysis is required to ensure that it has been formed correctly/it is of the correct composition. Imashuku et al. developed an optical emission spectrometry method for the on-line monitoring of copper–zinc thin films as they are formed.210 The methodology was described in full in the paper and was aided by a schematic diagram of the instrumentation. The selection of appropriate emission wavelengths was undertaken with testing at numerous input powers, chamber pressures and target compositions. By using the different target compositions, the linearity between the emission intensity ratio of Cu atom lines to Zn atom lines and between the atomic ratio of Cu/Zn was plotted. The lines that gave the highest intensity did not give the highest R2 values. Therefore, instead of using the primary Cu line at 324.8 nm, the alternative line at 296.1 nm was plotted against the emission from the Zn line at 307.6 nm. This resulted in the highest R2 value of 0.986 being obtained. Having identified the ideal wavelengths to use and set up a calibration, the on-line analysis of films being formed could be undertaken. It was noted that the chamber pressure had to be constant otherwise decreases in intensity result, thereby invalidating the calibration.

The 3D high-resolution chemical characterisation of sputtered lithium-rich NMC811 (Ni0.8Mn0.1Co0.1O2) thin films using TOF-SIMS was reported by Priebe et al.211 The surface passivation layer and buried structure were analysed using a primary ion beam of 69Ga+ with an energy of 20 keV and with the detector in positive ion mode. The techniques of XRD (for crystal structure), SEM and FTIR were also used for characterisation. The TOF-SIMS results showed the presence of over-lithiated grains of diameter 400 ± 100 nm and nanoparticles of diameter 100 ± 30 nm with an increased 7Li16O+ content in the buried part of the film. It was thought that these Li-rich areas could possibly be used as Li reservoirs to compensate for lithium losses during fabrication and cell operation. Unfortunately, these structures disappeared after 30 days at ambient conditions.

An assortment of X-ray-based techniques were assessed for the measurement of materials from ultra-thin layers to buried interfaces in a study reported by Bure et al.212 The study started by comparing WDXRF with liquid phase deposition ICP-MS analysis (a destructive method but regarded as being the reference technique). The values obtained for the WDXRF for the thickness of alumina films were in good agreement with those obtained using the reference method (R2 of 0.99 once a scaling factor of 0.94 had been accounted for). Parallel angle resolved X-ray photoelectron spectroscopy (pARXPS) was discussed and its theory and capabilities described. This was then compared with WDXRF for the determination of the dose and thickness quantification in ultra-thin alumina. The ARXPS was able to track the linear growth during deposition accurately. The ability of ARXPS to determine the thickness of an intermediate atomic layer deposition stack of silicon dioxide, hafnia and alumina layers was evaluated against the reference-free grazing incidence XRF (GIXRF) and an inelastic background analysis method. The GIXRF gave a value of 3.14 ± 0.24 ng mm−2 for the mass deposition of Al, while the ARXPS calculation yielded 2.90 ng mm−2. The Hf mass deposition was 7.4 ± 0.6 ng mm−2 according to GIXRF and 9.0 ng mm−2 using ARXPS. The similarity in results led the authors to conclude that the ARXPS was a suitable technique for dose measurements. The layer thickness data were less impressive with the alumina thickness of 2 nm being calculated to be 1.9 nm, 1.4 nm and 1.5 nm for XPS-IBA, pARXPS and GIXRF, respectively. The buried hafnia layer of 1.5 nm thickness was calculated as being 1.5 nm, 1.1 nm and 0.9 nm for the same techniques. A thicker multi-layer system (up to 28 nm) was then analysed using hard XPS (HAXPES) combined with the ion beam analysis. Interfaces were clearly identified and the layer thicknesses correlated well with nominal values.

4.7.2 Solar cells. A review entitled “what can glow discharge optical emission spectrometry (GD-OES) technique tell us about perovskite solar cells?” was presented by Zheng et al.213 The review cited 68 references and covered topics such as its advantages and limitations, how it provides information on structural properties, how it can be used at each stage of production, etc. Another of the interesting sections described how it may be used to investigate the aging of the cells. In the summary and future section, the authors acknowledge that although a powerful technique, it has not yet been widely used. However, since it offers reasonable detection limit, fast depth profiling, a fast analysis time and can analyse a large surface area, it was envisaged that its use would increase.

Thermal evaporation is a potential way of forming perovskite solar cells on a larger scale than spin-coating but the methodology has not yet been perfected. The lack of understanding of the mechanisms of the formation of high quality methylammonium triiodide films has led to sub-standard material being formed and consequently to devices with lower efficiency. Castro-Mendez et al. have studied these problems by analysing the crystalline properties of the material deposited by co-evaporation of lead iodide and methylammonium iodide.214 The methylammonium iodide source was temperature controlled to enable different evaporation rates, meanwhile the lead iodide deposition rate was controlled using a quartz crystal microbalance. The deposits formed were analysed using grazing incidence wide angle X-ray scattering (GIWAXS) spectrometry, XRD and XRF microscopy. The methylammonium iodide temperature plays a critical role in the formation of the perovskite film. Below 140 °C the film is lead iodide rich. Below 155 °C, GIWAXS identified that the film formed has a secondary orthorhombic phase. The XRF microscopy indicated that a stoichiometric composition occurs if heated to between 140 and 155 °C, whereas above 165 °C, the films become enriched in methylammonium iodide. Those films enriched with methylammonium iodide have poor crystallinity so when they are incorporated into solar cells, they have the greatest power conversion efficiency but at the cost of a lower open circuit voltage.

Another paper to have employed GIWAXS spectrometry for the analysis of thin films was presented by Vegso et al.215 Transition metal dichalcogenides are often prepared using the chemical vapour deposition method and it is obviously beneficial to be able to have an in situ inspection technique to observe the growth as it happens. This was accomplished in the study with full experimental details discussed. The growth of the vertically aligned layer MoS2 was observed in a one-zone chemical vapour deposition chamber using a laboratory table-top GIWAXS instrument. The authors discussed the relative merits of using a micro-focus X-ray source with Montel optics and a single photon counting, two-dimensional X-ray detector. The position-sensitive two-dimensional detector enabled the orientation of the molybdenum sulfide layers to easily be distinguished. The impressive performance of the GIWAXS instrument was improved further by suppressing the background scattering arising by using a guarding slit, a beamstop and helium gas in the chemical vapour deposition reactor. Layer growth could be monitored by following the width of the molybdenum sulfide diffraction peak. The monitoring of the layer growth in real-time is a useful asset for industry.

4.7.3 Analysis of electronic components. Numerous reviews pertinent to this area of analysis have been presented. Some of these are of particular techniques whereas others concentrate more on the sample type. A good example of a techniques-based review was presented by Kallquist et al., whose review of the advances in studying interfacial reactions in rechargeable batteries by photoelectron spectroscopy cited 227 references.216 The review gave an introduction to the materials and the problems associated with their analysis before presenting a lengthy section on the phenomena that occur at the interfaces. Further sections followed on XPS analysis of the bulk materials, operando and in situ XPS applications and other related X-ray-based techniques, e.g. Ambient-Pressure X-Ray Photoelectron Spectroscopy (APXPS). It finished with a summary and some future perspectives. The review is a useful contribution given the rapidly increasing use of XPS over the last 10 years.

Another review of instrumental methods for the analysis of electronic materials was presented by Kimura.217 This review, citing 90 references, discussed X-ray fluorescence holography (XFH) as a probe of hyper-ordered structures, e.g. some semiconductors and superconductors. A brief history of XFH was given followed by the principles and apparatus required. Examples of hyper-ordered structures in various materials observed by XFH were described. Here, three types of hyper-ordered structures in crystalline materials were introduced. These were: dopant clusters, local ordering of matrix atoms around dopants and metal complexes in organic materials. Another section was dedicated to the future perspectives of hyper-ordered structures investigated by XFH. Finally, conclusions were given.

A review by Lin et al. (with 159 references) of the in situ characterisation of advanced electrode materials for sodium ion batteries toward high electrochemical performances was made.218 After a brief introduction of sodium ion batteries, a discussion of the performance and capabilities of numerous analytical techniques that have been used was presented. Sections covering in situ applications of Raman spectroscopy, TEM, XRD and X-ray absorption near-edge structure (XANES) were presented. The bulk of the review concentrated on the analysis of components of the batteries and the use of the analytical techniques. Again, a useful summary and future perspectives section was provided.

The history and recent advances in elemental analysis of germanium-based functional materials (germanium and germanium dioxide) that contained 61 references, was presented by Guselnikova.219 Several analytical techniques were discussed including applications that used AAS, OES (including both ICP and two jet arc plasma (TJP)) and mass spectrometry (ICP-MS, GD-MS and spark source-MS). Further sections followed that discussed methods of analysis that used matrix removal, either by extracting the analytes, volatilisation of the matrix following an acid dissolution, reactive distillation of the germanium as a volatile compound, etc. The review was well presented, with tables of information provided to aid the reader.

The advanced characterisation techniques used for sulfide-based solid state lithium batteries was presented by Nomura and Yamamoto.220 This review cited 177 references and discussed the problems associated with lithium ion batteries including their electrochemical decomposition, mechanical degradation at interfaces, dendrite growth of lithium, slow Li diffusion, etc. The techniques used to investigate these phenomena included XPS, TOF-SIMS, TEM and X-ray computed tomography were all discussed.

A paper, by Wang et al., reviewed the use of synchrotron-based XRD and XAS studies of layered lithium nickel manganese cobalt (NMC) cathode materials that have the formula LiNixMnyCozO2.221 This extremely long review cited 153 references and discussed some of the problems and challenges associated with the NMC batteries use. It went on to discuss the in situ and operando studies of NMC materials using XRD and XAS. Several other sections then followed that focussed on various aspects that aid in the development of improved materials and discussed how the techniques have helped elucidate mechanisms of the processes.

Several interesting applications have also been presented during this review period. Included in this number is a paper by Guselnikova et al. who determined a suite of analytes (43) in high purity germanium using ICP-MS following a matrix removal through volatilization processing procedure.222 The volatilisation procedure was discussed in detail, but basically involved the electrolytic production of chlorine gas which was then dried by passing it through sulfuric acid and then used to volatilise the sample (1 g) at a temperature of 230 °C. The volatilised germanium tetrachloride was carried by the excess chlorine flow to a 20% sodium hydroxide solution, where it precipitated as germanium dioxide. When only about 3 mg of residual material was left in the volatilisation chamber, it was dissolved in a mixture of hydrochloric (0.3 mL) and nitric (0.15 mL) acids. After dilution to 1 mL, the internal standards (Bi, In, Re and Se) were added and the sample was ready for ICP-MS analysis. Detection limits were extremely impressive, ranging from 0.001 to 4 ng g−1, representing improvements of between 2.5 and 1700 times that of conventional ICP-MS analysis. The accuracy of the methodology was assessed using spike–recovery experiments. Most analytes had full recovery ± 20%. Some though, e.g., Ir, had random losses and had an average recovery of only 74%. The material analysed was found to have very high purity with the sum of all the analytes determined totalling 0.0000014%.

Another study to have used matrix volatilisation as a means of sample processing was presented by Medvedev et al.223 The sample was high purity tellurium, and the sample matrix (2 g) was first melted (600 °C) and then volatilised away under vacuum distillation. The 2 g of sample produced approximately 0.01 g of residue which was taken up in a mixture of nitric and hydrochloric acids (3[thin space (1/6-em)]:[thin space (1/6-em)]2). After dilution, the sample was ready for analysis using ICP-OES and ICP-MS. Method validation was accomplished through spike–recovery experiments and the results for 49 analytes were 70–120%, with the majority (42) being between 84 and 115%. Of those outside that range, Be, Fe, Ni, Pb and Zn all showed low recovery values. This was presumed to be because of volatilisation with the tellurium matrix. Detection limits for ICP-MS determinations were between 0.05 ng g−1 (for Ho and Lu) up to 90 ng g−1 (for Ca). The large majority were <10 ng g−1. The LOD obtained using ICP-OES were obviously higher, but were still impressive, being on the range 0.5 ng g−1 (for Ba and Li) up to 80 ng g−1 for Ir.

Although the use of LIBS is less common for the analysis of these materials, there have been some applications. Numerous types of batteries exist and so a classification methodology to identify them rapidly would be welcome. Paradis et al. used data obtained from the LIBS analysis of several types to develop, train and test algorithms designed to aid classification.224 The paper took the reader through the process of obtaining the data on a commercial spectrometer, normalising it, training the machine learning algorithms (support vector machine, spectral angle mapper and random forest) and then testing several deep neural network algorithms. The proportion of samples used for training and testing were varied. Using only 1% of samples to train led to poor classification, irrespective of which algorithms were used. Using 5% or more of the samples for training led to significantly improved classification success for most algorithms. The worst performing algorithm was spectral angle mapping. Random forest proved very successful and the four deep neural network algorithms all produced very high classification success rates (>97%) when 5% of samples were used for training, rising to >99% success when 20% were used.

A further paper to investigate the classification of materials was presented by Song et al.225 These workers used LIBS followed by a recursive feature elimination algorithm to select the features to be analysed. These were then input to linear discriminant analysis to decrease the dimensions of the data. Then, a back-propagation neural network was used to classify silicone rubber insulators from seven different manufacturers. The classification success rate was approximately 95%, so although not fool-proof, the method was regarded as being a success as well as being rapid and useful.

Two papers by the same research group have reported the use of LIBS for homogeneity measurements of lithium ion batteries226 and for depth-resolved elemental analysis on moving foils.227 In the former paper, a LIBS system operating at 1064 nm, a pulse duration of 6 ns and a varied pulse energy of between 0.25 and 50 mJ was used. Detection was achieved using three Czerny–Turner style spectrometers with ranges of 193–262 nm, 268–536 nm and 500–736 nm. The system enabled Al, C. Co, Li, Mn and Ni to be determined giving a quasi-depth-resolved relative concentration profile for each. Despite LIBS being renowned for giving poor precision, this work managed to separate the noise caused by the technique from genuine differences in signal from different regions of the sample. The latter paper used the same setup, but concentrated on measuring depth-profiles on moving samples.227 The focussing lens for the pulsed laser and a deflection mirror were mounted on a moving stage which was precisely aligned in height and orientation to the movement of a conveyor belt on which the sample was transported. The speed of the stage and the conveyor were synchronised to ensure the same spot on the sample was analysed each time. The setup was discussed in detail in the paper and was applied to the analysis of electrode foils for lithium ion batteries. For a 100 μm thick coating, a total of 10 laser pulses were required to obtain a full depth profile. Compared with previous approaches, such as applying different pulse energies at consecutive positions, the authors thought that this technique had great advantages for depth-resolved LIBS.

An interesting XRF application was described by Xu et al. who used it to calculate the thickness of lithium film anodes.228 Since Li is not usually measurable using XRF instrumentation because the signals are too weak, the authors used the signal from the copper substrate it was placed on instead. The Cu Lα line was used for the measurement, with the emission from this base material decreasing with coating thickness until it fades into the background. The method proved to be successful, with lithium films of thickness between 5 and 125 μm being measured reasonably accurately, the methodology was also capable of detecting defects and assessing coating uniformity. Since the method uses standard instrumentation, the authors envisaged no problems with it being utilised in industry as it represents a rapid and non-destructive way of analysis.

The analytical capabilities of TOF-SIMS and Orbitrap-SIMS were compared in a paper by Franquet et al.229 Although developed for organic molecules, the Orbitrap system can also be coupled with atomic spectrometry instrumentation and offers higher resolution and higher mass accuracy than standard spectrometers. In this study it was applied to the quantitative determination of P doping profiles in Si0.25Ge0.75 fins (20 nm wide) and the quantification of C and Si levels in InGaAs/GaAs fins (16 nm wide). The TOF-SIMS instrument used a Bi+ primary beam at an energy of 30 keV and a mass resolution of 10[thin space (1/6-em)]000 for the silicon germanide samples. For the indium gallium arsenide samples, the primary beam was Bi3+ at an energy of 30 keV and at a resolution of 8000. The Orbitrap system used a primary beam of Cs+ at 2 keV and at a resolution of 250[thin space (1/6-em)]000 for both sample types. Both TOF-SIMS and the Orbitrap-SIMS systems were operated under ultra-high vacuum to prevent any effects from residual gas on the mass spectra produced. The superior resolving power of the Orbitrap system managed to separate polyatomic interferences that even the “high” resolution TOF-MS systems could not, e.g. the 71GaO could easily be separated from the 75AsC peaks. This higher resolving power can potentially be put to good use for detecting impurities or implants in these nano-sized objects.

Monoatomic layer-by-layer analysis of MXenes (two-dimensional inorganic compounds consisting of atomically thin layers of transition metal carbides, nitrides, or carbonitrides) and their parent MAX phases using ultra-low energy SIMS was reported by Michalowski et al.230 Single particles of MXenes could have their depth profiles taken with atomic level resolution. The study detected O in the carbon sub-lattice. This was attributed to the presence of oxycarbide MXenes. The study went on to determine the composition of the adjacent surface termination layers, showing that they interact with each other. Analysis of the metal sub-lattice showed that the MAX parent compound Mo2TiAlC2 exhibited perfect out of plane ordering. However, the Cr2TiAlC2 showed some inter-mixing between the Cr and the Ti in the inner transition metal layer. The accurate measurement of remaining capacity of batteries is important for battery management and possible second use. A paper by Hou et al. attempted to estimate the remaining capacity of lithium ion batteries by using X-ray computed tomography.231 The paper gave a thorough grounding of the mathematics and theory behind the method and also provided a schematic diagram of the instrumentation required. Simply put, an algorithm was developed that correlated the state of charge with the Li content. Once that had been achieved, Faraday's law and the Peukert equation were used to derive the capacity remaining. Other parameters that required consideration were the density, thickness and area of the material as well as the operating conditions. Three lithium iron phosphate batteries were then tested using the method developed and the results compared with those obtained using the traditional method. The maximum prediction error of the developed method was 5.2% which was a significant improvement on that of the traditional method (24.1%).

Luo et al. investigated the charge–discharge mechanism of high entropy cobalt-free spinel oxide ((CrMnFeNiCu)3O4), which is an attractive class of anode materials for lithium ion batteries.232Operando quick-scanning XAS was used to study the lithiation/de-lithiation process. A monochromator oscillation frequency of 2 Hz was used and 240 spectra were integrated to achieve a 2 min time resolution. High-photon-flux synchrotron radiation was employed to increase the XAS sensitivity. The results indicated that the Cu2+ and Ni2+ cations were reduced to their metallic states during lithiation. However, their oxidation reactions were less favourable compared with the other elements upon delithiation. The Mn2+/3+ and Fe2+/3+ cations undergo two-step conversion reactions to form metallic phases, with MnO and FeO as the intermediate species, respectively. During delithiation, the oxidation of Mn occurs before that of Fe. The Cr3+ ion is reduced to CrO and then Cr during lithiation. A relatively large overpotential is required to activate the Cr re-oxidation reaction. Armed with this knowledge, the authors said that it should help with material design to ensure improved performance.

Several applications of the analysis of electronic materials are interesting and these are summarised in Table 1 below.

Table 1 Applications of the analysis of electronic materials
Analyte Matrix Technique Comments Reference
F and N Amorphous carbon film TOF-SIMS; XPS; XRR; radial distribution function Effect of N doping on the microstructure and dry etch properties of amorphous carbon films. As the amount of N increased, the density of the film decreased. In addition, the number of pyridinic and pyrrolic nitrogen bonds with low formation energy increased. The TOF-SIMS data showed that the penetration depth of F from the etchant decreased as the amount of N increased. It was possible to improve the etch properties of amorphous carbon films by increasing the density of the film and by placing N in graphitic positions rather than in pyrrolic and pyridinic positions while increasing the amount of nitrogen in the film 233
Several (61) Grade T 000 tellurium ICP-OES Tellurium formed by hydrometallurgical processing of electric filter dust concentrate. Analytical wavelengths and viewing height were optimised for compromise conditions. Detection limits in the range 10−7 to 10−4 wt%. Intra-laboratory precision was better than 25% 234
Co, Cu, Ni and Zn High entropy oxides used as lithium ion anodes (Mg0.2Co0.2Ni0.2Cu0.2)O EXAFS The individual contribution of each metal in this composition to the electrochemical reaction mechanism on charge and discharge was assessed. The number of near neighbours and bond distances extracted from the EXAFS data elucidated the electrochemical activity of each atomic species. In the first cycles, Co, Cu, Ni and Zn were converted to their metallic state during the first lithiation step. Conversion of Co, Ni, and Zn to oxides was partially reversible during delithiation but Cu was not 235
Several (10) Printed circuit boards pXRF; ICP-MS Comparison of data from portable XRF and ICP-MS after an aqua regia digestion. Three different particle size ranges tested: 200 μm, 750 μm and 2 mm. Three sample preparation methods tested for pXRF analyses. The ICP-MS results from the 200 μm particle size material were obtained, with a precision of 5% RSD. Data for 200 μm particles from pXRF had a precision of better than 10% RSD and agreed to within 20% with ICP-MS values. The loose powder method was sufficient with regard to sample preparation. Larger particle sizes had trouble due to inhomogeneity 236
Li, O, S Pyrene-4,5,9,10-tetraone and argyrodite composite cathodes TOF-SIMS An air-free analysis methodology was developed to prevent oxidation of the Li7PS5Cl. Sample was prepared in a glove box then transported to a TOF-SIMS instrument using an air-free transfer vessel. The TOF-SIMS then performed an analysis with better than 150 nm resolution. Analysis using SEM and hardness measurements also undertaken in an air-free environment enabled a chemo-mechanical mapping of the composite cathode 237
Cd and Te Cadmium telluride thin film LIBS Orthogonal double pulse LIBS used. First pulse was of low energy (0.5 mJ) at 266 nm to minimise damage to the sample whereas second pulse was at a much higher energy (30 mJ) and at 1064 nm to re-excite the sample plume. Time acquisition delay and inter-pulse delay were both optimised. Double pulse LIBS increased emission intensity while decreasing self-absorption when compared with single pulse LIBS and also improved accuracy of the analysis 238
Al and Cu Aluminium-doped copper oxide films LIBS Films prepared using RF magnetron sputtering under different pressures and powers. A ns-LIBS system was used to quantitatively and qualitatively analyse films. The calibrations from the intensity ratios and the atomic content ratios were virtually identical, indicating the data were reliable. A 2D map of the surface was drawn to show the distribution of Al. A picosecond-LIBS system was then used to determine film thickness 239
Cu, Ga, In and Se Copper indium gallium selenide (CIGS) films TOF-SIMS; SEM Effect of four different sputter sources (argon ion gun, oxygen gun, caesium metal gun and gas cluster ion beam with oxygen) and at three temperatures (−120 °C, −50 °C and 20 °C) on depth profiles and the morphology of the craters was studied. The argon gun produced the worst matrix effects, the oxygen gun also produced significant effects, but could be used to profile Ga and In only. The caesium gun provided the fewest matrix effects and the least crater facets. However, it also provided the lowest sensitivity. Analysis at −120 °C also caused fewest problems 240


4.8 Nanostructures

Atomic spectrometry, through techniques such as XRD, XPS, XRF, single nanoparticle (sNP) ICP-MS and ICP-OES has a key role in the characterisation and detection of nanoparticles (NPs) with over 170 papers published in the period covered by this ASU. This section focuses on the analysis of the NPs. Work involving detection of NPs in other sample matrices features in the other ASUs in this series.2–5 In addition, many published articles only mention the technique(s) used without any further analytical detail and as such are not reported on here.
4.8.1 Topical reviews. A number of reviews covering NP measurements have been published this year. The first of these (115 references) covers the analytical chemistry of engineered nanomaterials (NMs) with regard to legislation and metrology.241 The paper gives examples of the techniques required for determining different properties of NMs, such as composition, structure, shape, size distribution, concentration and surface functionality, and also the different analytical scenarios that may be encountered, e.g. pristine NMs or those encountered in more complex matrices. A large part of the paper is devoted to the regulatory frameworks covering NM production and use and the associated metrology required to ensure valid analytical measurements. Of particular benefit is a table itemising all of the currently available reference and test materials available from accredited producers along with tables covering all of the available standards and general guidance documents for NM characterisation. The authors concluded that there is a lack of standards and RMs for instrument calibration and method validation for some methods, particularly those for determining number-based concentration or identification/quantification of surface functional groups or coatings and that there are currently very few RMs and methods that can deal with complex engineered NMs and with detection in complex matrices. They also highlight other challenges including the need for methods that are capable of analysing the nanoscale chemical composition of particles with a complex core/shell structure and surface coatings. These topics are addressed in two recent papers. Sanmartin-Matalobos et al. reviewed (214 cited references) the analysis of semiconductor quantum dots (QDs), with a focus on their properties, surface chemistry and detection.242 The section on analytical techniques used to characterise QDs in these terms gives examples of the use of XAS, XPS, XRD, XRF, whilst the section on detection and quantification cites reports using ICP-MS, ICP-OES and GFAAS. Each section also gave a brief overview of which facet of a QD is being probed and included coverage of molecular and UV-vis spectrometry as well as the use of voltammetry. The authors also came to a similar conclusion to the first review covered here, that current analytical techniques have limitations when applied to determining QDs in complex media and new methodologies need to be developed.

In biological systems NPs interact with proteins forming a biological corona complex which drive the interaction of nanomaterials with and within the cells, giving rise to potential medical applications and also toxicological concerns. Fuentes-Cervantes et al. reviewed the use of ICP-MS for NP–corona characterisation.243 The review gave a brief overview of all of the possible techniques used for NP–corona analysis before again pointing out the lack of standardisation and validation in this type of analysis, which is a common theme from the analytical chemistry community working in this field. The paper also discussed the reasons for a lack of reproducibility in LC-MS protein analysis and then discussed how coupling separation techniques, such as size exclusion chromatography (SEC) and field flow fractionation (FFF) with ICP-MS can improve this situation by, for example, determining the S and/or P and metal content of the NP–corona complex to allow stoichiometries and information of surface functionalisation and ligand density to be estimated. The uses of single particle (SP) ICP-MS in these studies are also discussed along with protein tagging with ICP-MS detection. The authors concluded that the combination of the atomic spectrometry techniques discussed can be a useful complementary tool in the development of methods to understand the NP–protein corona complexes arising in biological systems.

Two review papers published this year have focussed on SP-ICP-MS. The review by Laborda et al., with 137 references cited, is structured around the fundamentals of the technique, discussing both nebulisation and LA for sample introduction, the processes that occur in the plasma, the effects of collision/reaction cells, mass analysers, detectors, data acquisition and signal processing.244 The benefits and limitations of the technique were also discussed, with the latter aspect including the need to assume the NP shape, the restriction of LA to low fluences (generally lower than 1 J cm−2, to avoid NP degradation), the challenges with the upper size LOD due to incomplete volatilisation, the potential limitations on particle transport efficiency measurements due to a lack of CRMs and RMs and the need for harmonisation of criteria for data processing. This latter point arises because of the different criteria adopted by different workers for differentiating between signal baselines and those signals originating from particle events. These different criteria adopted by different workers can affect the LOD estimation. With the major instrument manufacturers all now incorporating SP calculation tools in the instrument software this could mean that comparing results between laboratories becomes more challenging, by adding to the natural variation that will occur due to the analytical process. It is hoped that instrument manufacturers are aware of this and are updating their software as data processing theory develops due to ongoing research by groups such as that of Laborda, and as SP-ICP-MS matures. The authors of the review also considered the future perspectives for SP-ICP-MS, highlighting the recent move towards detecting plastic particles, directly through 12C and 13C or via polyatomic species, where the LODsize increases to 1 μm or greater due to the relatively high C background. They also discussed developments in sample introduction systems, and how those for SP and single cell analyses are converging for larger particle sizes, and the recent development of different instrument configurations for SP and single cell work. The paper is recommended reading for new and existing practitioners in the field.

Most SP-ICP-MS work detects only a single isotope due to the need to capture all of the information in a single analytical scan. The introduction of ICP-MS instruments with a TOF mass analyser has allowed the simultaneous detection of multiple signals and this aspect has been reviewed (107 references) by Tian et al.245 As the authors pointed out, NPs in environmental and biological samples are unlikely to be in the pristine form, highlighting the benefit of this type of instrument. The review included a section on the principles of SP-ICP-MS, including the fundamental equations behind the technique as well as the use of quadrupole, sector field (SF) mass analysers and multi-collector (MC) detection with the latter. A detailed section on the operation of the TOF mass analyser was also provided. The authors noted that it opens up the possibilities of particle mass and particle number quantification using isotope dilution analysis. Some 40 plus references concerned with multi-isotope/element SP-ICP-MS were summarised in a table whilst the review also discussed online microdroplet calibration, differentiating between engineered and naturally occurring NPs along with source tracing and the possibilities of multi-elemental single cell analysis.

Finally, a review has been published, with 100 references cited, into the use of SR-XRF for understanding the behaviour of nanomaterials in biological systems.246 The review gave a brief introduction of a few widely used SR-based analytical techniques, including XAS, μXRF, scanning transmission X-ray microscopy (STXM) and circular dichroism spectroscopy. Recent advances of their applications in the analysis of nanomaterial behaviours were then discussed, including different nanomaterial transformations such as biodistribution, biomolecule interaction, decomposition, redox reaction, and recrystallization/agglomeration. Finally, some of the challenges faced in this field were proposed.

4.8.2 Fundamental studies. A number of papers published this year report research into the fundamentals of SP-ICP-MS, such as data processing parameters, the effect of dwell time, calibration strategies and the upper and lower LODsize. Gimenez-Ingalaturre et al. developed an approach to critically assess SP-ICP-MS data and how it should be reported to maintain its validity.247 The main focus of the work was the effect of the baseline signal on the capability to detect particles and described an overall quality control strategy based on successive dilutions in combination with the estimation of size critical values. The theory behind the work and the definition of the size critical value is given in detail in the paper. Data were acquired at two different dwell times, 0.1 and 5 ms with total acquisition times of 60 and 300 s. Nebulisation efficiency was determined using Au NPs, with similar results obtained for both the frequency and the size methods, and gravimetrically measured sample uptake. Data processing was by both the instrument manufacturer's software (Syngistix Nano-Application module version 2.5), with a 5-sigma threshold applied, calculated as five times the square root of the mean baseline intensity of the time scan, or by using SPCal.248 The work showed that, as the baseline signal increased due to increasing amounts of dissolved ions of the target Ag NP, the particle number concentration recovery rapidly decreased from 100% for a suspension of NPs only, with recoveries of 2% achieved with an ionic Ag concentration of 0.3 and 2.4 ng mL−1 for dwell times of 5 and 0.1 ms, respectively. Thus, the authors recommended that samples are diluted as much as possible whilst still giving sufficient particle-like events for the detection of the NPs in a sample. For samples which have a high dissolved ionic analyte content, such as fish livers or natural waters then this may not always be possible and separation techniques, such as FFF or CPE, should be considered. Liu et al. investigated the effect of different dwell times (0.05, 0.1, 1 and 5 ms) on the data acquired by SP-ICP-MS analysis.249 The paper gave the theory behind and discussed in detail the data processing needed for dwell times above and below 1 ms, including the measurement of transport efficiency, the distinction of signal and background, the evaluation of LODdiameter and the quantification of mass, size and particle number concentration of NPs by SP-ICP-MS. The authors pointed out that the use of longer dwell times can lead to overlapping signal events and also recommended that samples are adequately diluted to minimise the probability of this occurring.

In SP-ICP-MS, instrumental calibration is usually determined through the analysis of a series of ionic calibration standards prepared in high purity water or, for example with tissue extracts, a mimic of the diluted sample matrix. Aramendia et al. assessed an online standard additions approach in which blank, NP or ionic standards were added to the sample uptake line via a T-piece.250 This results in mixed histograms, which included the sample and standard signals combined. A signal deconvolution routine was then used to extract the instrument responses due to the sample and added standard followed by calculation of the particle size distribution and number concentration. The approach was compared to results obtained by conventional external calibration for 50 nm Au NPs, dispersed in three matrices, water 5% EtOH and 2.5% TMAH, and were found to be in good agreement for particle size and particle number concentration with recoveries of 97% or greater. For 60 nm Ag NPs dispersed in water, the particle size was in agreement with that obtained using TEM. It would be interesting to see how the approaches compared for a more complex matrix such as river or wastewater.

In order to extend the upper LODsize limit in SP-ICP-MS, Ojeda et al. used a linear pass spray chamber designed for single cell analysis.251 Initial studies with SiO2 particles gave an upper LODsize of 1200 nm, compared with an estimate of <1000 nm for a cyclonic spray chamber, with 29Si being the monitored isotope, and measured without the use of a collision cell as this has been reported to induce band broadening which impacts the lower and upper LODsize limits. When applied to the 27Al signal arising from clay particles it was observed that particle above 420 nm, which were shown to be present by TEM analysis, were not detected. To test the assumption that the larger particles were not detected due to detector saturation, the instrument was significantly detuned to reduce ion transmission. This allowed detection of particles up to 1000 nm in diameter with a consequent increase in the lower LODsize from 35 to 110 nm. Thus, with two different tune settings the size range for clay particles, based on the Al content, was 35 to 1000 nm.

In TOF mass analysers discrete packets of ions are detected quasi-simultaneously, which means that multi-elemental information can be gained when NMs are analysed. This gives a lot of data, as ion packets of, for example, ∼2 ms duration are sampled every 25 to 30 ms. Three to 10 of these spectra are then summed and stored for interpretation by some form of a machine learning algorithm. Tharaud et al. applied cluster analysis to SP-ICP-TOF-MS data to identify engineered and natural NPs from model systems of various mono-, di- and tri-metallic NPs and two clay minerals of known composition.252 The algorithm classified the NPs within each sample using the Ward minimum variance method calculated with the Euclidean distance as the recursive merging strategy. For single element, Ag and Au, and bimetallic AuAg core–shell, the measured particle sizes were in statistical agreement with the manufacturers data, as were the mole fractions for the AuAg NPs. From this data the depth of the Ag shell was estimated to be 17 or 90 nm depending on the overall particle size. For Ni or Zn CoFe oxide NPs the calculated mole fraction of each element was in agreement with the theoretical values, but the size was overestimated in each case by a factor of approximately two. The authors hypothesised that this is due to either the instrumental LOD being above that of the signal arising from the minor elements, i.e. Ni, Zn and then Co, in smaller particles or that there were pure and bi-metallic NPs also present in the products, or a combination of both. The measured and theoretical Al[thin space (1/6-em)]:[thin space (1/6-em)]Si molar ratios in the kaolinite sample studied were in agreement but for the montmorillonite the measured value was computed to be greater than the theoretical value, possibly due to the presence of Al NPs as well as the clay. The authors concluded that this work provides a starting point to differentiate between anthropogenic and natural NPs but caution that a thorough evaluation of the data must be performed in order to avoid misclassifications of NPs. Harycki and Gundlach-Graham reported the use of an ICP-TOF-MS instrument for the analysis of microdroplets of NPs and microplastics.253 The instrument, an icpTOF-S2, has an improved sensitivity compared with the manufacturer's other models due to a shorter flight distance and allows data to be acquired at a rate as short as 12 μs per spectrum in an m/z range of 6 to 254. In this work the target analytes were Ag and Au NPs of varying sizes, 30 to 80 and 10 to 50 nm, respectively, and multi-element doped polystyrene microspheres. After optimisation of the various microdroplet sample introduction and instrumental parameters, the estimated sizes of the Ag and Au NPs were in statistical agreement with the supplier's data except for the 10 nm Au NPs for which the size was overestimated. The authors attributed this latter deviation to the truncation of the particle size measurements due to the LOD being reached, with the calculated LOD value for Au size being 15 nm. Reducing the sampling depth to improve sensitivity gave a lower cut of point but from the images presented they would still appear to be truncated at the lower end of the NP size distribution. For the polystyrene microspheres measurements, with the instrument tuning biased for increased sensitivity for low m/z values, the mean size was 3.1 μm, with a critical mass for 12C detection of 3.2 pg and a critical diameter of 1.8 μm. Both of these papers contained a wealth of detail on the fundamentals of the instruments used and the data acquisition and processing parameters.

Tian et al. compared the performance of three quadrupole, two TOF and one MC-ICP-MS instruments for measuring the 109Ag[thin space (1/6-em)]:[thin space (1/6-em)]107Ag isotope ratio in silver NPs and Ag exposed cyanobacteria cells.254 After optimisation, peak hopping mode, with sub ms dwell time and settling times was used for analysis using ICP-Q-MS. The TOF instrument monitored full mass spectra with a dwell time of 5 ms and for SP/single cell-MC-ICP-MS analysis, 30[thin space (1/6-em)]000 cycles with a dwell time of 8 ms was used. The results obtained led the authors to conclude that ICP-Q-MS performed poorly in accurately determining the isotope ratios within single NPs/cells due to missing information as a result of peak hopping. They also concluded that with MC-ICP-MS it is hard to identify the SP events from NPs or cells with minimal metal mass due to the extremely low signal-to-noise ratios. Therefore, ICP-TOF-MS would be the instrument of choice for time resolved isotope ratio measurements due to the rapid acquisition of full m/z spectra. Again, the paper provided a wealth of detail on the methods employed for optimisation and all three papers covered here would be good reading for those new to TOF instruments.

Continuing the theme of investigations into the fundamentals of SP-ICP-MS, Chun et al. investigated the possibility of making these measurements with a triple quadrupole (TQ) ICP-MS instrument.255 The instrument was tuned using a process and a model (which was fully described), whereby the first quadrupole would allow transmission of more than one m/z but not potential polyatomic interferences. This is known as bandpass mode. To demonstrate the approach the 109Ag[thin space (1/6-em)]:[thin space (1/6-em)]107Ag isotope ratio of Ag NPs was measured in suspension in high purity water and increasing amounts of Zr, which could form 91Zr16O, using O2 as the collision gas. For all of these measurements the relative error from the true value was stated to be less than 3%. The system was then used to determine the Ag fraction % in AgSn NPs and this was found to be 3.60%. Unfortunately, no statement of accuracy was given and neither was the particle composition determined using another method. The system was then tested for the 176Yb[thin space (1/6-em)]:[thin space (1/6-em)]174Yb ratio in the presence of the potential interferent Gd and results, shown by scatterplot, suggested that in bandpass mode the GdO interferences were not present. The authors concluded that the technique shows promise for measuring elemental or isotopic ratios with small mass differences, and that it should also work in mass shift mode. However, although the paper contains a mass of detail and theory it is hampered by the lack of readily available validation data with regard to the accuracy of the ratios presented.

To determine particle sizes in mixtures of monodisperse NPs, a separation step, such as FFF or hydrodynamic chromatography, is normally used. Gao et al. have developed an alternative approach using a Bayesian discriminant analysis based on a finite mixture model consisting of kernel density estimation.256 The mixtures under study were made from individual 30, 40, 50 and 60 nm Au NP suspensions. The method gave a lower LODsize of 5 nm and the measured particle number concentrations in the mixtures were in agreement (±3%) with the theoretical values based on the manufacturer's data.

When analysed, suspensions of NMs often contain traces of organic components arising from either the original matrix for pristine NMs or from the sample type or extraction reagents used to prepare experimental or real samples. To this end, the influence of organic carbon (6% w/w glycerol and 10% w/w ethanol), on the operating conditions (sampling depth and nebuliser gas flow) have been evaluated when measuring NP composition and size (Au, Pt and Se NPs) using SP-ICP-MS.257 It was observed that 10% w/w ethanol caused positive matrix effects on the number concentration due to changes in aerosol generation and transport when compared with data for the NPs in water. Positive or negative bias on the size distributions were obtained for both organic matrices and these effects depended on plasma operating conditions and NP composition. In addition to changes in transport efficiency, matrix effects on size distribution also depend on plasma characteristics and carbon-based charge transfer reactions (Au and Se). The authors suggested the use of an internal standard mitigated the effects of these interferences however, this might impact the accuracy of the results obtained depending on the instrument type used, e.g. a single quadrupole ICP-MS. The work does however highlight the need to matrix match calibration standards if the organic component is high, or to dilute the samples sufficiently if this is possible within the lower number of particles that can be detected to give reliable data.

In processing the data from SP-ICP-MS, an assumption is made that the NPs under study are spherical which may not always be the case. Koolen et al. have described a modelling tool that extends the measurement of particle morphologies beyond spherical to include cubes, truncated octahedra, and tetrahedra.258 The model introduces a further function into the equations used in data processing, a dimensional descriptor such as edge-lengths, once the particle mass has been calculated. The model was successfully used with four NP shapes (Au spheres and Cu cubes, truncated octahedra, and tetrahedra) and three Cu-based surface-alloy NPs (CuAg, CuPd, and CuPdAg).

4.8.3 Sample introduction and separation techniques. A number of alternative sample introduction systems for NPs have been reported in the literature this year. The synthesis of NPs often uses hydrophobic organic ligands which are only soluble in non-polar organic solvents which can prove a challenge for analysis by ICP-MS. One approach to achieve this though is the use of a microdroplet generator, which significantly reduces the solvent load on the plasma, and this has been undertaken for the direct introduction of toluene and mesitylene into an ICP-MS instrument.259 Soot formation, which can cause instrumental drift, was minimised and the transport efficiency was maintained at 100%. The effect of different vacuum interface configurations and the addition of oxygen or nitrogen on the detection efficiency and instrumental background signals was also investigated for a range of elements. The highest detection efficiency was obtained for a “Jet” interface with the addition of nitrogen at a flow rate of 10 mL min−1 resulting in an increase by a factor of 2–8 depending on the element. The lowest detectable mass, based on counting statistics, was 1.4 ag for Pb, which corresponds to a diameter of 6.1 nm of a spherical, metallic NP. The developed method was applied to quantify TiO2 NPs in sunscreen which found that the NPs rapidly settled when diluted into toluene but were more stable in mesitylene which the authors attributed to the increase in viscosity of this solvent.

A study into the use of microfluidic chips for on-line sample pretreatment for SP-ICP-MS was reported by Kajner et al.260 The polydimethylsiloxane microfluidic chips, capable of high-range dilution and sample injection, were cast in 3D-printed moulds. The devices allowed the determination of NPs using only a few tens of microliters of sample with elimination of solute interferences by dilution, solution-based size calibration and the characterisation of binary nanoparticles. The design of the chips was such that they could be linked together to extend the dilution range of the system by more than a magnitude per chip. The application of the system was demonstrated by the analysis of monodisperse suspensions of Ag and Au NPs and an AuAg core–shell NP. In all cases the calculated particle metrics, particle size, number and concentration, were in agreement with the manufacturer's values.

Beauchemin's research group have continued their work with heated spray chambers.261 In this study, a total consumption, so near 100% transport efficiency, IR heated sample introduction system was evaluated for SP-ICP-MS while preserving plasma robustness by not removing water. A 50 mL cyclonic spray chamber was modified so that a pen IR heater could be introduced within the baffle in its centre. The optimisation parameters included the sample uptake rate (25–75 μL min−1) and IR heating temperature (20–300 °C), with Au NPs as the target analyte. Optimal conditions were found to be a 50 μL min−1 sample uptake rate and an IR heating temperature of 110 °C. A transport efficiency of 99.2% was achieved versus the 17% when using the standard pneumatic nebulisation system with a double-pass spray chamber. As a result, the IR-heated system decreased the solution detection limit five-fold and improved the method size detection limit from 26 to 16 nm compared with the standard system, without degrading the accuracy for the measurement of 60 nm Au NPs. Stiborek et al. reported a new technique, IR-LA-SP ICP MS, for the digital mapping of biomarkers in tissues based on desorption and counting intact Au NP tags.262 In contrast to conventional UV laser ablation, Au NPs are not disintegrated during the desorption process due to their low absorption at 2940 nm and up to 83% of Au NPs in the tissues were detected. The technique was demonstrated by mapping a nuclear protein, Ki-67, in aggregates of colorectal carcinoma cells with the results compared with those obtained using confocal fluorescence microscopy and UV-LA-ICP-MS. Precise counting of 20 nm Au NPs was achieved with this method, with a single-particle detection limit in each pixel, which generated sharp distribution maps of the biomarker protein.

Electrical asymmetric flow field-flow fractionation (EAF4) is an emerging separation technique and its use has been investigated for the characterisation of spherical metallic Ag, Au and Pt NPs with both citrate and phosphate coatings in different carriers and with different detectors including ICP-MS.263 After optimisation of the various EAF4 parameters, which is discussed in detail, the NPs were successfully detected by either dynamic light scattering, ICP-MS, multi-angle light scattering or UV-vis spectroscopy, although the parameters under study, such as electrophoretic mobility and zeta-potential were found to vary with detector. The authors discussed the causes of these variations and concluded that, from a practical point of view, the technique is still in its infancy and further studies are necessary for robust characterisation of NPs.

Three groups have reported on the coupling of a differential mobility analyser to an ICP-MS instrument for the detection of NPs. The first of these reports, by workers at NIST, focussed on the development of an accurate calibration method for quantifying NPs. The target analyte was AuTiO2 NPs (4 nm gold NPs adsorbed on 100 to 300 nm TiO2 NPs). When ionic standards were used and compared with the AuTiO2 control, a repeatable difference in the slope was observed which led to an over-estimation of both the Au and Ti content by a factor of almost four. This discrepancy was found to be derived from the metal oxide NPs (independent of the presence of Au) and was caused by the electrospray-differential mobility analyser and not the ICP-MS instrument, due to multiple charged species. After correcting for this, the agreement between the ionic standard and AuTiO2 metal quantification improved significantly. The paper included a discussion on material properties for improved accuracy across different shapes, sizes, compositions and surface chemistries to demonstrate the general utility of the calibration across a broad number of fields. Bierwirth et al. coupled a spark discharge NP generator to a differential mobility analyser which was in turn coupled to an ICP-MS instrument and a condensation particle counter (CPC) in parallel.264 After optimisation of the coupling of these components the system was used, with Au NPs as the model, to gain insights into the processes influencing particle structure and the limitations of the technique. For instance interactions of the carrier gas molecules with the particle surface were investigated. The use of the setup to study the influence of the particle morphology on the charging characteristics was also outlined.

Hsieh et al. reported the coupling of an in-house fabricated atomiser coupled with differential mobility analyser – SP-ICP-MS.265 The atomiser comprised a nebuliser followed by two drying stages to remove solvent and was described as relatively easy to operate. The system was able to resolve the size fractions of mixed 30 nm and 50 nm Au NPs and analyse suspensions of 50 nm Au NPs with lower and upper particle number concentrations of 4.1 × 105 to 107 ng mL−1. The method was then applied to environmental samples by characterising Au and Ag NPs spiked into wastewater. It was found that both NP types formed heteroaggregates with colloids in the wastewater.

Table 2 shows other applications of nanomaterial characterisation and/or detection presented in the literature during the time period covered by this ASU.

Table 2 Applications of nanomaterial characterisation and/or detection
Analyte Technique Comments Reference
Au NP doped nanoplastics SC-ICP-MS Study into the uptake of nanoplastics by RAW 264.7 mouse macrophage cells 266
Metal-tagged human B cells SC-ICP-MS The absolute quantification of nanoparticle interactions with individual human b cells 267
Au NPs functionalised with alkylthiols Nanoprojectile SIMS The nanometrology of NPs and their interfaces 268
Ag, Au and AgSiO2 NPs GD-OES High-throughput NP characterisation elemental mapping 269
Carbon black NPs SP-ICP-MS The use of metal impurities for the detection/quantification of carbon black particles in water 270
CuO NPs SP-ICP-MS, DLS Guidance and applications for nanomaterial characterisation in complex media 271
Ag NPs SP-ICP-MS, FFF-ICP-MS Study monitoring Ag NPs during synthesis 272
Ag NPs SP-ICP-MS Estimation of the contribution of signal noise to the diameter uncertainty of Ag NPs 273
Pd NPs SP-ICP-MS Study on the influence of H2 partial pressure on biogenic Pd NP production 274
Ag, TiO2 NPs LIBS The use of LIBS to assess the penetration of biocides in limestones 275
Ag, Au NPs LIBS Study into the use of NPs for enhancement of the LIBS signal 276
Au NPs AF4-ICP-MS Characterisation of NP species present in oligonucleotide–gold nanoparticle conjugates 277
None, theory paper SP-ICP-MS Theoretical study into the statistical properties of spikes in SP-ICP-MS time scans 278
Iron oxide magnetic NPs μXRF, XAFS Study into the effect of magnetic particle hyperthermia application on the composition and spatial distribution of iron oxide NPs 279
Cu, CuFe2O4, graphite NPs LIBS Investigation of the diffusion dynamics and characterisation of optically trapped NPs 280
TiO2 nanostructures GEXRF Study on GEXRF for the characterisation of advanced nanostructured surfaces 281
AuAg NPs Spark ablation OES On-line compositional measurements of AuAg aerosol nanoparticles generated by spark ablation using OES 282
Si3N4 lamellar grating GEXRF Proof-of-concept scan-free GEXRF measurements for the investigation of lateral ordered 2D nanostructures in the soft X-ray range 283
TiO2 NPs SP-ICP-MS Method for background signal correction in SP-ICP-MS 284
Ge nano-object Nano-XRF Study into the quantitative analysis of individual nano-objects 285


5 Forensic analysis

Numerous sample types relevant to this review may be analysed for forensic purposes and these will be discussed in the sections below. As always, a forensic analysis should cause minimal damage to the sample so that it may be stored and re-analysed if improved tests are developed. Therefore, XRF-based techniques that are non-damaging and laser-based techniques, e.g. LA-ICP-MS or LIBS, that are minimally damaging are commonly used.

The development of a TOF-SIMS method for detecting fingerprints has long been a goal for forensic scientists because it would be more sensitive than the standard tests. A contribution towards this was presented by Charlton et al. who analysed fingerprints on three common exhibit-type surfaces: paper, polyethylene and stainless-steel and then compared the results with the standard processes.286 Measurements of the fingerprints occurred after aging for 12 days for the paper, 14 days for the polyethylene and 16 days for the stainless steel. Analysis of the fingerprints was undertaken using a 25 keV Bi3+ primary ion beam. Raw data sets of total ion images were acquired at 600 × 600 pixels resolution. From these, mass selected images were constructed. These parameters provided good image resolution within an acceptable time frame for 6 × 6 mm image size (36 min 48 s per image acquisition). Better resolution could have been obtained, but at the expense of an unacceptable acquisition time. All of the TOF-SIMS images were normalised to the Total Ion Current to correct for fluctuations in ion beam current and differences in geometry. It had the added bonus of improving the quality of ridge detail. The TOF-SIMS data therefore compared very favourably with those from the standard method and was capable of detecting more fingerprints and improving the quality of others significantly.

A recently developed method is SIMS using very high energy primary beams (MeV) to emit secondary ions from surfaces. This recently developed method was employed by Siketic et al. to analyse organic materials including a section of mouse brain, crossing ink lines on paper and a fingerprint.287 The instrumentation was described in detail with the assistance of a photograph. The primary ion beam used was 14 MeV of Cu4+ with a primary ion dose of approximately 3.5 × 1010 ions per cm2 for ink imaging, 1.5 × 1010 ions per cm2 for brain imaging and 5 × 109 ions per cm2 for fingerprint imaging. The mass of the analysed molecules was determined with a reflectron-type TOF analyser, where the start signal for the TOF measurement was generated by the secondary electrons emitted from a 5 nm thick carbon foil placed over the beam collimator. This was a new instrument configuration that had a lateral resolution of approximately 20 μm. It also had a very high secondary ion yield, i.e. it had greater sensitivity than most other instruments and could be used for samples of any thickness. The potential of the technique in the fields of biology and forensic science was discussed and the conclusion was that it had huge promise.

5.1 Forensic applications of glass analysis

Glass is a very common sample type in forensic analysis, so it is fitting that it has been the most popular area of research during this review period. A review of the elemental spectrochemical analysis of glass for forensic applications was presented by Trejos.288 The review contained 100 references and was split into sections covering μ-XRF and ICP-based methods including laser ablation (LA) as sample introduction. A useful table was presented detailing the advantages, disadvantages and abilities of numerous techniques and whether or not it is currently used for forensic analysis. Included in the table were techniques such as LIBS, which is not yet used in forensic laboratories routinely. However, given that it is virtually sample non-destructive, is extremely rapid and may be used on particles of <0.4 mm, then if the calibration difficulties can be overcome, surely it will be adopted in the future. The review emphasises this in its emerging methods section. Other techniques such as PIGE, INAA, SEM-EDS, PIXE and TXRF were also discussed in this table. Another large section was devoted to the chemometric treatment and analysis of the analytical data. This is an important aspect of the analysis because it enables classification of samples and possible matching with a degree of certainty of samples found at crime scenes and samples taken from a suspect. Multivariate analysis, such as PCA, random forest, Soft Independent Modelling of Class Analogy (SIMCA), and k-nearest neighbour (KNN) methods for glass discrimination and comparisons have been used. However, the author emphasises that most workers use assessment of comparison intervals and likelihood ratios.

An inter-laboratory comparison, conducted as part of an IAEA study, for the forensic analysis of glasses was reported by Kaspi et al.289 Samples (48) were collected by the Israeli police from cars of different model (13) and manufacturer (10) and distributed to laboratories in Israel, Croatia and Finland. Ion beam methods of analysis, particularly PIXE, was used by the laboratories. The reference samples NIST 610 and 620 were used by all three laboratories to calibrate the instruments. Once results in accordance with certified values had been obtained (to within 10%), the laboratories analysed the samples. Data from these were then input to principal component analysis to reduce the dimensions of the data prior to being input to the random forest machine learning algorithm. As always, much of the data was used as a training set while some samples were the “test” set. The resulting model provided results with a classification accuracy of >80% which was equal to or better than any lab-specific model. The model will likely increase in accuracy as more specimens are measured and recorded. It was thought that the overall approach would be applicable to a wide variety of forensic applications, e.g. gunshot residue, illicit drugs, etc. Another paper by the same research group was very similar in approach to the previous one but instead of only using data from PIXE analyses, LA-ICP-MS, PIGE, SEM-EDS and prompt gamma-ray neutron activation analysis (PGAA) was also used.290 As well as random forest, support vector machine was also tested for classification. However, the random forest performed better. In general, the PGAA results led to poor classification success. Similarly, SEM-EDS also produced poor classification unless it was supplemented with Ca and Fe data obtained using PIXE. However, the overall conclusion was that data obtained using different techniques could effectively be combined into one database that could be used for classification.

Two papers by the same research group have also analysed glasses. In an inter-laboratory comparison by Corzo et al. the relative merits of using μXRF equipped with modern silicon drift detectors rather than the traditional lithium doped silicon detectors were evaluated.291 The standard test method for the forensic analysis of glass using μXRF (ASTM E2926) gives recommendations for the number of replicate measurements to characterise a known source. It also provides the criteria for the comparison between the known and questioned samples. However, these recommendations were based on the use of μXRF instrumentation equipped with traditional lithium-doped silicon (SiLi) detectors. Four laboratories took part, with three using silicon drift detectors and one using a lithium doped silicon detector. Two other labs provided reference data using LA-ICP-MS and LIBS. Those labs using LA-ICP-MS operated according to ASTM E2927. As recommended by ASTM E2926, only elements with a signal to noise ratio >10 were included in pairwise comparisons. The μXRF instruments equipped with silicon drift detectors provided improved precision and detection limits (between 1.4 μg g−1 and 1386 μg g−1) and excellent discrimination (>99%) of different-source samples. However, the false exclusion rates for same-source samples were relatively high (>20%). This disappointingly high rate was addressed in two ways. Increasing the number of fragments of glass from the same sample to at least five decreased the false exclusion rate to less than 5%. An alternative was to modify the recommended comparison criterion. This reduced the false exclusion rate from 23% to 2%, while maintaining low false inclusions (<1%).

The other paper by this group analysed 30 different portable electronic devices screens, 15 screen protectors and 3 brands of liquid glass using a μ-XRF instrument equipped with two silicon drift detectors.292 Once obtained, the spectra/data were treated in several different ways including spectral overlay (comparing the spectra to see if they match) and spectral contrast angle ratios. The latter was explained in the text. Basically, it measures the similarity between two spectra using a comparison of vectors. The length of a vector and its orientation are determined by the energy and intensities for each spectrum. The angle between the vectors from the two spectra is termed the spectral contrast angle. A smaller spectral angle indicates that there are fewer differences between the two spectra. The initial spectral overlay analysis classified the samples into five groups for the portable devices and 4 groups for the screen protectors. The precision when measuring glass from the same source was better than 8%. Discrimination could be improved further for within-group samples when considering reproducible differences in signal intensities (discrimination 98.4% for the portable devices and 98.1% for the screen protectors). The lowest false exclusion rates among same source samples were 3.3% for the portable device screens and 0.8% for the screen protectors. The multivariate techniques of linear discriminant analysis (LDA) and PCA were also used to discriminate between materials and produced clear separations between the five groups for the portable devices and four groups for the screen protectors. Although only a preliminary study using a limited number of samples, the potential for μXRF to be used for forensic classification was clear.

Non-destructive analysis of the black ceramic writing on 37 glasses from known sources was achieved using a portable XRF instrument by Ishimi et al.293 The reference material NIST 612 was used for quality control and for calculating the LODs of the analytes, which ranged from 0.8 to 3.7 mg kg−1. These were dependent on the acceleration voltage used. Three acceleration voltage ranges were tested, these were: 15 keV (suitable for Si Kα (1.740 keV) to Ni Kα (7.473 keV)), 30 keV (suitable for Cu Kα (8.042 keV) to Pb Lα (10.552 keV)) and 50 keV which was suitable for those analytes above Bi Kα (10.839 keV). Results indicated two main types – the Bi type and the Pb type. A line pair ratio method was used for further discrimination, with the Bi type being best discriminated using the Zr Kα/Bi Lα and Cu Kα/Cr Kα. Discrimination for the Pb types was achieved using Zr Kα/Pb Lα and Cu Kα/Cr Kα line ratios. Overall, the discrimination rate was 98.4%, which was deemed successful and consequently, it was concluded that portable XRF could be used for discriminating these ceramic prints.

5.2 Forensic analysis of organic materials

5.2.1 Drugs. A review entitled “State-of-the-Art Analytical Approaches for Illicit Drug Profiling in Forensic Investigations” was presented by Ahmed et al.294 Although the review contained 54 references, it should be emphasised that the majority of the text was focussed on chromatographic methods of analysis, e.g. GC-MS, GC-IRMS, HPLC, UHPLC and LC-MS. There were also sections on physical methods and IRMS. Metal traces might exist in drugs through plants used in producing the drug, the production method, the containers and possibly airborne particles. Therefore there was also a section on ICP-MS applications. As well as text, a table presented the data for easy reference.

A study of the analysis of Cannabis seized in Ghana followed by chemometric methods to try and classify different production farms was reported by Douvris et al.295 Twelve analytes (As, Ca, Cd, Cu, Fe, Hg, K, Mg, Mn, Na, Pb and Zn) were determined using ICP-MS following a microwave-assisted acid digestion (in the presence of hydrogen peroxide). Also sampled and digested were soil samples from three known cannabis farms. A total of 46 samples were analysed, 34 of which were cannabis drugs and 12 soils. Once the data had been obtained, they were input to discriminant analysis. No Hg was detected and so data from 11 analytes were used. Canonical plots for the soil samples showed a very clear distinction between the three farms. Similar plots for the drugs showed that many examples could be allocated to a farm, but others, especially those drugs seized in 2020 appeared not to originate from any of the three areas studied.

5.2.2 Documents. A pan-European inter-laboratory comparison that involved 17 laboratories in 16 countries was reported by Fischer et al.296 A three-page rental agreement was sent to the laboratories with instructions to attempt six things. These were: identify the printing technique, determine whether all three pages were printed using the same printer, were all three pages the same paper, were the pages originally stapled, were headings and signatures of the same ink and were they of the same age? A huge selection of different techniques was used for the study including UV-luminescence, LIBS; infrared spectroscopy, Raman and FTIR (micro-)spectroscopy; X-ray spectroscopy (including SEM-EDX, PIXE and XPS); mass spectrometry (including ICP-MS, SIMS and matrix-assisted laser desorption/ionisation (MALDI)) and electrostatic imaging. Other, non-imaging methods, such as non-multimodal visual inspection, micro-spectroscopy, physical testing and thin layer chromatography were also conducted. Unsurprisingly, no single technique could fulfil the requirements of all six tasks. Atomic spectrometry was used for only some of the tasks, with XPS, LA-ICP-TOF-MS, LIBS and SIMS being used successfully to identify the printer used. Task 3, the paper discrimination, was undertaken using LIBS, XPS, SIMS and LA-ICP-TOF-MS with varying degrees of success. The LIBS data identified that page 2 was made from different paper to that of pages 1 and 3. The SIMS imaging resulted in a false positive, while LA-ICP-TOF-MS and XPS claimed “undecidability”. Task 5, the discrimination of inks, was undertaken by the same four techniques. Success was achieved using LIBS and XPS, whereas SIMS and LA-ICP-TOF-MS failed to identify that the signatures on page 2 were from different inks. It was concluded that a multi-method mode of analysis was therefore required to succeed with most of the tasks, although no technique successfully discriminated the age of the inks.

The age of documents was also studied by Pigorsch et al.297 These authors used the accelerator mass spectrometry (AMS) measurements of 14C present in the starch of the paper to estimate the age (±3 years) of paper manufactured between 1950 and 2018. The 14C originated from the atmospheric bomb tests after the second world war. The starch used in paper originates from annually grown crops and so it was thought that it could be a useful indicator of age. Starch was extracted from the paper using water and manual disaggregation. Sufficient starch for analysis could be obtained from 1 or 2 g of paper. Meanwhile, paper fibres were obtained through a series of Soxhlet-style extraction steps using several solvents to remove fillers, colourants etc. The solid that remained was extracted using 1% hydrochloric acid at 85 °C, washed in deionized water, and finally dried. The curve for 14C increases, reaches a maximum and then decreases again. It is therefore possible for two dates to give a similar result. However, other tests on the papers' composition enabled differentiation between something made in the 1960s from the 1990s. The measured 14C concentration values in the starch extracts for most of the paper samples were highly correlated with the data of the 14C bomb peak calibration curve. The results from the fibres were less impressive, but still showed good correlation.

A further paper that assessed the veracity of documents was presented by Al-Ameri et al. who used LIBS to analyse the inks followed by a chemometric approach to analyse the data.298 A correlation matrix of the spectra was calculated for both the original and questioned documents together. This was then transformed into an adjacency matrix with the aim of converting it into a weighted network under the concept of graph theory. Clustering algorithms were then applied to the network to determine the number of clusters. A number of different scenarios was used, including different laser printer inks, different photocopier toners, different papers, etc. The method developed was easy to implement and, because it gave a visual output, was also easy to interpret, even by a non-expert. It also had a high degree of success in detecting forgeries.

5.2.3 Paints. The analysis of automotive paints is a common task for forensic scientists. However, the analysis of multi-layered paint can be troublesome. This problem has been addressed by Merk et al. who combined the use of LIBS and Raman spectroscopy to analyse the same spot of the layered paints sequentially.299 Eight paint samples of assorted colours were collected from car crashes. Minimal sample preparation was required prior to the analysis with only one edge being smoothed using a diamond coated polymer film to ensure that no irregular lumps or other blemishes existed, this is an important point to note because uneven or rough surfaces can lead to severe sensitivity and stability problems during LIBS analyses. The paint chips' sizes ranged from 5–20 mm and depths from 110–250 μm. The LIBS operating conditions were described and the analyses took the form of four parallel lines of laser shots with each laser target area and each line being 50 μm apart to ensure that the craters did not overlap. Each target area was analysed in triplicate (3 laser shots). The values for each of the elements at each spot were then averaged over the four parallel lines. Results from the 19 analytes were plotted and showed clear differences – even for samples of the same colour. The Raman spectra were taken over the range 180–1750 cm−1and the spectra baseline corrected, normalised and then input to the k-means ++ algorithm. The combination of the LIBS data for metallic analytes and the Raman spectra for organics led to clear distinction between samples. Analysis of the same sample again led to the same results being obtained, indicating good reproducibility.

A multi-technique characterisation was used for the analysis and identification of forgeries of five paintings attributed to the artist Pippo Oriani.300 The non-invasive techniques, such as UV imaging, IR and IR-false colour photography, XRF, μ-Raman and FTIR spectroscopies were used during the study. The XRF was used to analyse the pigments and the cardboard base. The XRF results showed the same elements for all five paintings. In particular, Ca, Cu, Fe, Ti and Zn, and traces of Cr and Mn were detected in each investigated point, regardless of the colour. This suggested that their signals originated from the cardboard rather than the pigments. However, different colours did show enhanced signals for some analytes, e.g. the dark blue had large amounts of Cu whereas the light blue had Cu and Ti. Some colours, e.g. red and purple, gave no extra XRF signal. This study prepared the foundation to create a scientific database to support the future authentication of Oriani's paintings.

5.3 Forensic applications of inorganic materials

5.3.1 Gunshot residue. Several papers have been published in this area during this review period. One of the more interesting ones was presented by Brunjes et al. who used a single particle analysis approach employing an ICP-TOF-MS instrument.301 Residue from four different weapons (two revolvers and two semi-automatic handguns) firing ammunition from three different manufacturers was collected and then analysed. Minimal sample preparation was required which is an obvious advantage. The shooter wore a new pair of powder-free nitrile gloves for each of five shots. The gloves from the third shot were then removed and turned inside out to ensure particle losses were minimised and placed in polyethylene bags. The gloves from the fourth shot were analysed using SEM-EDS for comparative purposes. For the sample preparation, the gloves were filled with a detergent solution and agitated to release the particles. The solution was then diluted to 500 mL after standing for 30 minutes to allow large particles to sink, 2 mL aliquots were removed for analysis. Gold nanoparticles of 100 nm were used as a particle standard. With the exception of 44Ca, only the most abundant isotopes of each element were used for data processing. Approximately 2000 particles were observed per injection of 300 μL and for a measurement duration of 1 min. Of those 2000, between 123 and 414 could be classed as gunshot residue. Particles characteristic of gunshot residue, i.e. ones that contain Ba–Pb–Sb, were found in four examples from Geco ammunition, 13 from Magtech and 34 from Remington. However, because full scans were obtained, information on other elements was also collected. This enabled multi-element fingerprinting. This would be useful when analysing ammunition that does not contain Pb or Sb.

A similar approach was adopted by Szakas et al. who also used single particle ICP-TOF-MS to characterise gunshot residue.302 These authors fired shots (having removed the bullet and propellant but leaving the cartridge and primer) and then collected the residue in a flask. Two primers were tested: one containing the classical mixture of Ba, Pb and Sb and the other being Pb-free. As well as the typical particles containing the Ba, Pb and Sb, the technique also detected appreciable amounts of other particles, including Pb–Cu in the leaded primer and Ti–Zn. Interestingly, the mean particle size for the leaded primer were significantly smaller than those from the lead-free (180 and 320 nm, respectively). These authors also stated that the single particle technology offered a rapid and sensitive method capable of determining many more particles per unit time than the traditional SEM-EDS method. Elemental characterisation of individual particles as well as the determination of particle size distribution and particle number concentration was all possible. It was also capable of measuring smaller particles, i.e. those significantly smaller than 1 μm.

Menking-Hoggatt et al. analysed numerous gunshot residue samples taken from hand swabs using LIBS and square wave anodic stripping voltammetry.303 Samples originated from six sub-groups: known shooters that used leaded ammunition, shooters that used lead-free, shooters that used mixed ammunition and those that had been a shooter but had since done activity that may or may not have caused contamination and/or alteration of the elements present on their hands. The other two sets were non-shooters split into sub-groups of those that were at low risk of exposure to sources that could mimic gunshot residue and those that were of high risk. The combined use of LIBS and electrochemistry enabled rapid screening of samples based on the presence or absence of inorganic or organic gunshot residue profiles. The LIBS analysis took only 2 min for the collection of spectra collected from 25 different micro-areas while consuming less than 0.2% of the sample. The combined methods proved to be fit-for-purpose by detecting residue from leaded and lead-free ammunition fired from various firearms. Electrochemical data were obtained in 5–10 min and could be confined to a very similar area as the LIBS. The same samples were successfully analysed using SEM-EDS and the elemental profiles observed were very similar to those obtained using LIBS. Analytical data were treated using likelihood ratios. Two likelihood ratios were calculated: one considered each shooter population (leaded, lead-free and mixed ammunition) separately and the other considered all the shooter populations combined. The equations used for the calculations were presented in the text. The combination of LIBS and electrochemistry enabled a successful prediction of which of the six sub-groups an individual belonged to with a success rate of 89.6% being obtained. When the data were input to a neural network program (with 60% used for training, 20% for validation and 20% for testing), the success rate increased to over 93%. Importantly, discrimination between shooter and non-shooter was excellent.

Unfortunately, the recipe for primers used by all cartridge makers in China is identical meaning that methods such as those described above are of very limited use for discriminating between them. Guo et al. therefore reported a method using Pb isotope ratios to discriminate between different types.304 This is because the abundance of Pb isotopes in primer components of lead styphnate varies significantly. Forty-four samples were characterised from their isotope ratios of 206Pb/204Pb, 207Pb/204Pb, and 208Pb/204Pb using LA-multicollector-ICP-MS. Material was collected by discharging the firearm and then taking the cartridge and tapping it onto clean sheets of paper. The powder that emerged was then transferred to glass slides using double sided sticky tape. A small volume of Tl solution was then added and allowed to dry. This was to act as a correction for mass bias. The reference material NIST SRM 981 was used to correct for instrumental mass discrimination. Hotelling’s T2 test was used to evaluate the significance of changes in the Pb isotopes from before and after the gun discharge. The authors provided the mathematical basis of this test in the text. No significant difference was found in the pre- and post-discharge Pb ratios meaning that any differences in Pb ratios was due to the ammunition rather than the heat and pressure of the discharge. The 206Pb/204Pb ranged from 17.36 to 20.31, 207Pb/204Pb ranged from 15.53 to 15.78 and 208Pb/204Pb ratios ranged from 37.24 to 39.19. Multivariate likelihood ratios were then calculated and led to a low rate of misleading evidence (<0.53%).

5.3.2 Explosives and propellants. A paper entitled “Recent developments of image processing to improve explosive detection methodologies and spectroscopic imaging techniques for explosive and drug detection” was presented by Sharma et al.305 Ideally, explosives should be detected from a safe distance and the paper describes LIBS applications for this and compares LIBS with other methods. It also described a drone capable of detecting explosives and drugs from a range of 2.8 km.

Ippoliti et al. measured the isotopic abundance of analytes and the trace metal signature pre- and post-detonation in a number of explosives, notably ammonium nitrate-aluminium (plus RDX and TNT).306 A field study was conducted to recover samples of post-blast explosives from controlled detonations of ammonium nitrate-aluminium which were then analysed using isotope ratio mass spectrometry (IRMS) (for N and O isotope ratios) and ICP-MS for the 26 trace metals. The experimental detail of how the collection plates were arranged was explained in the text. The N and O ratios pre- and post-detonation were in relatively good agreement (within one standard deviation). Many of the trace metals also showed considerable overlap, including B, Cd, Cr, Ni, Sn, V and Zn. The conclusion was that forensic analysis of post-detonation residues can be directly related back to the pre-detonation materials.

6 Cultural heritage samples

In common with the forensic applications, there is a requirement that the analysis should be non-damaging (or at least minimally damaging) to the samples. Most of the applications therefore employ XRF-based methods or those that cause so little damage, it is not visible, e.g. LA-ICP-MS and LIBS.

There have been several useful review articles published in this research area. One by Harikrishnan et al., that contained 111 references, was entitled “Archaeophotonics: applications of laser spectroscopic techniques for the analysis of archaeological samples”.307 The techniques covered were LIBS, LIF and Raman spectroscopy. An introduction to the theory of each was provided followed by lengthy sections in which their applications were discussed. Tables were provided for easy reference. Sample types included in the review were coins, ceramics and pigments. Multi-modal spectroscopy, i.e. a combination of techniques, was also discussed.

Another review, this time by Ghervase and Cortea, cited 154 references while discussing the past and future of LIF in the field of cultural goods.308 A large table covered many of the applications giving the sample type as well as the excitation and emission wavelengths. A section on different types of laser that have been used was also presented. Also included were sections on hybrid techniques, e.g. LIBS-LIF and the post-acquisition data-processing using techniques such as PCA, HCA, Spectral Angle Mapper and multi-curve resolution.

A review entitled “The assets of laser induced breakdown spectroscopy (LIBS) for the future of heritage science” was presented by Detalle and Bai.309 The review had 171 references and covered the period 2015–2020. It started by presenting data on the geographical areas from which the LIBS papers originate, the materials it has been used to characterise (with textiles, ceramics, metals and paintings being the most common) before giving some theory. After discussing some instrumentation (including portable instruments), it went on to discuss data treatments and included classification programs such as PCA, random forest and ANN. Importantly, it also discussed the advances in the instrumentation and optimisation of the operating parameters to ensure minimal damage is inflicted on the samples. Similarly, important advances in calibration were also highlighted. Clearly fans of the technique, the authors stated that their aim was to encourage more scientists working in the cultural heritage field to use LIBS.

A paper (in Chinese) containing 61 references focussed on the progress of non-destructive spectroscopy on the conservation of cultural relics.310 The methods discussed included XRF, XPS, LIBS, Raman spectroscopy, IR, diffuse reflectance spectroscopy (DRS) and hyper-spectral imaging. The abilities and applications of each of the techniques was highlighted.

For many years the National Institute for Nuclear Physics, Italy has been using PIXE and XRF to analyse cultural heritage samples. A review of the work containing 62 references was presented by Sottili et al.311 The work has included the development of many instrumental advances, including a portable macro-XRF scanner and the MACHINA (moveable accelerator for cultural heritage in situ non-destructive analysis) for on-site ion beam analysis.

Other papers have been presented that would be of interest to archaeologists and cultural heritage scientists. One such paper was presented by McMillan et al. who discussed how the instrumental advances in XRF (both benchtop and portable) have enabled even non-specialists to obtain reliable and accurate data.312 However, interpretation of the data can be problematic and so non-specialists can quickly run into trouble. They therefore developed an open source interface designed to facilitate reproducible and robust outcomes during lithic sourcing studies. The program called SourceXplorer, was described as being intuitive and easy to use. It bundled together a series of chemometric multivariate analysis tools including the unsupervised PCA, supervised LDA and, because neither of these have the ability to classify samples of “unknown origin”, partial least squares discriminant analysis. The authors demonstrated the use of SourceXplorer by analysing materials from British Columbia, Canada.

It is known that many techniques e.g. XRF and LIBS work best when the sample surface is flat and smooth. The extent to which the signal differs when a curved surface is presented was studied by Trojek and Trojkova.313 A number of sample types were studied including a coin, steel ball bearings, ball bearing-shaped cavities, flat objects that were deliberately tilted, etc. Monte Carlo simulations were undertaken as well as XRF measurements. The uncertainties of measurement of minor components was typically 5–10% (sometimes much higher) which is significantly worse than that obtained on a flat surface. Sharp edges posed a particularly bad problem. For curved surfaces, more accurate and precise data were obtained when convex areas were analysed compared with concave ones. It was noted that for analytes that had similar energy wavelengths, the effects of a curved surface were very similar. This was not necessarily the case for analytes with very different emission energies. This could have a significant effect on accuracy and precision and decrease the efficiency of the classification of cultural heritage materials.

Palleschi et al. developed a method called Interesting Features Finder which is apparently a simple algorithm for detecting “interesting” spectral features independently of the variances they represent in a series of spectra.314 The algorithm was applied to the analysis of a painting from the Etruscan hypogeal tomb of the Volumni. The results were compared with those obtained using the established techniques of Blind Sources Separation and Self-Organised Maps. It was concluded that the algorithm developed could rapidly and simultaneously obtain several sets of images that contain the “interesting” features and information equivalent to that obtained from self-organised maps and blind sources separation combined.

A further paper to use an artificial neural network approach for high throughput spectral processing of data (in this case, LIBS spectra) was presented by Herreyre et al.315 The artificial neural network was trained by obtaining 1353 LIBS reference spectra of materials that may potentially be present in mortars. The authors used eight categories (binder, carbonate, alumino-silicate, tile, quartz, resin, charcoal and a blank they termed “Hole”) and the number of LIBS spectra per category varied between 102 and 230. Once obtained, the LIBS spectra were input to PCA to check that each one was clearly representative of its specific category and not a contribution from a different component of mortar because of a misidentification. The ANN parameters of pre-treatment, number of neurons and iterations were also optimised to ensure best possible classification while avoiding over-training. The result was a fast and accurate method of identifying the components and classifying the mortars. Since the processing could easily be automated, the authors emphasised that it could be used for numerous applications for cultural heritage and not just confined to mortars.

A recently commercialised instrument capable of making simultaneous measurements of XRF, visible & near-infrared (380–1100 nm) and short wave infrared reflectance spectroscopy (1100–2500 nm) was applied to the analysis of a multi-layered painting.316 It was therefore capable of obtaining atomic and molecular information simultaneously from samples. Data obtained were imported to an innovative multivariate and multi-block, high throughput data processing algorithm. This was a combination of PCA, brushing, correlation diagrams and maps (within and between spectral blocks) and had been developed specifically for the task of analysing multi-layered samples. When analysed, the results from the painting showed clear evidence of protein and lipids from the egg binder as well as gypsum, malachite and several metals. The combination of the instrument and the software enabled the characterisation of multi-layered objects without the need for performing micro-sampling.

Other, more specialised reviews and applications are given in the individual sections below.

6.1 Metallic artefacts of cultural heritage

Metallic cultural heritage artefacts are frequently discovered in a badly corroded state. Analysis techniques that do not damage the sample further are, as always, almost a pre-requisite. However, the techniques are usually confronted by a complex multi-layered material and so ideally should be able to differentiate different layers. A review into the analysis of metal artefacts using XRF was presented by Silveira and Falcade.317 The review cited 100 references and started by giving the reader a decent introduction to the theory before moving on to the analysis of layered structures and the determination of low z elements. The use of synchrotron radiation was also discussed. The use of portable instruments for larger samples or samples that could/should not be sub-sampled was also described.

Numerous applications have also been published in the area of metallic cultural heritage samples. These are summarised in Table 3.

Table 3 Applications for metallic cultural heritage samples
Analytes Sample Techniques Comments Reference
Au and several trace analytes Fifteen gilded objects from the Royal Palace in Caserta, Italy XRF XRF data input to numerous chemometric techniques (PCA), hierarchical cluster analysis (HCA), cluster analysis and graph analysis. Three clusters obtained: those made of wood, lead and those objects that had been restored with porporina. Graph analysis was the most successful of the chemometric techniques for discrimination purposes. The thickness of the gilding was also studied using ratios of the intensities of the fluorescence lines 318
Au and various Artworks from the Royal Palace, Caserta, Italy Portable XRF Spectra from XRF were analysed using PCA, dendrogram, k-means, graphical clustering and partial least squares regression. Several different gilding techniques were identified. The thickness of the gold leaf or foil was also determined using the partial least squares methodology 319
Various Bronze swords excavated from China pXRF; MC-ICP-MS Trace metals and lead isotope ratios determined in swords from Ba and Shu states in China. The swords look similar but have very different compositions. The Sn content is very different and the Pb is clearly from a different source 320
Various Working tools from Sardinia XRF Working tools from the Nuragic civilization were analysed to determine the composition of the bulk and the corrosion layer. Specialist Monte Carlo simulation software package for XRF (XRMC) was used to model the data. Very complex corrosion layers were reproduced. Corrosion layers were enriched in Pb and Sn. This normally leads to overestimates of their content for surface analysis techniques such as XRF. Results were in good agreement with those from the more destructive techniques of AAS and NAA 321
Various (34) and Os isotopes Corroded iron billets from a shipwreck off Cyprus LA-ICP-MS; negative TIMS Iron from a 2400 year old shipwreck was analysed to determine if it was going to or coming from Cyprus. The LA-ICP-MS was used for trace elements and the negative TIMS for Os isotope ratios. For LA-ICP-MS, the reference material NIST 1262b was used for method validation. For those analytes not certified, the NIST 610 glass was used to calculate concentrations by including empirical correction factors. Materials NBS 65d and NBS 152a were also used for quality purposes. Results indicated that the billets did not match local Cypriot ores and so it was concluded that the ship was going to Cyprus 322
Au and various Gilded samples ARXRF A very theoretical paper in which ARXRF was used to determine the thickness of gilded layers. The gilding was made for the experiments, but the work could be applied to cultural heritage samples 323
Various Corroded coins from burial soils in Egypt LIBS; SEM-EDS The progress of the corrosion removal process on four types of corroded coins was monitored using LIBS. The SEM-EDS was used to analyse the coins before and after cleaning. The Cu content decreased significantly after cleaning whereas the C, O and Pb increased. The conclusion was that Cu was one of the major corrosion products 324


6.2 Cultural heritage samples of organic origin

A paper by Calligaro et al. discussed nuclear methods for historical painting authentication.325 Techniques discussed were 14C dating using AMS (for the binder, canvas and support age dating), MeV SIMS (for binder and pigment molecular composition), ion beam analysis (IBA; for paint layer composition and stratigraphy), differential PIXE and full-field PIXE mapping and optical photothermal IR spectroscopy. A combination of these techniques was used by the authors to analyse a test painting. The data yielded enabled the history of the painting to be reconstructed.

A new software package for the processing of macro (MA)-XRF datasets using machine learning called XRFast was described by Vermeulen et al.326 It is an open-source, open access unsupervised dictionary learning algorithm that reduces the complexity of large datasets containing tens of thousands of spectra and identifies patterns. The methodology through which it achieves this was given in the text. This approach quickly reduces the number of variables and creates correlated elemental maps, characteristic for pigments containing various elements or for pigment mixtures. It creates an over-complete dictionary which is learned from the input data itself reducing the requirement of a priori user knowledge of the sample. The feasibility of this method was demonstrated by applying it to a board containing several known pigment mixtures and then to an 18th century Mexican painting. The pigment Smalt was identified which contains As, Bi, Co, K and Ni. In addition, mixtures of vermilion and lead white, and two complex conservation materials/interventions were also identified. When the process was compared with traditional methods, very similar results were obtained but the new algorithm achieved it much more rapidly.

The same research group proposed an end-to-end pigment identification framework, including pigment library creation, XRF spectra simulation, mock-up preparation, a pigment identification deep learning model and a 2D pigment map generation.327 As a case study, the framework was applied to late 19th/early 20th-century paintings by Paul Gauguin and Paul Cezanne. The former painting had been analysed previously using a combination of XRF, reflectance imaging spectroscopy and cross-section analysis and so a reliable dataset already existed for comparison. The framework was fully automated and offered high sensitivity for the underlying pigments as well as for pigments present at low concentration. Results obtained were in good agreement with those obtained using elemental mapping in the earlier study.

Another open access web-based app called Artificial Intelligence for digital REStoration of Cultural Heritage (AIRES-CH) was developed by Bombini et al.328 It aims to be a digital restoration of pictorial artworks through Computer Vision technologies applied to physical imaging raw data from techniques such as PIXE, PIGE, XRF and FTIR. Since these techniques have a very wide wavelength range they help identify the components of the pigments. The app is a multi-dimensional neural network that is specifically designed to identify and help restore damaged areas and find hidden pictures. The app will be available on the National Institute for Nuclear Physics (INFN) website.

Other applications for organic materials of cultural heritage are given in Table 4.

Table 4 Applications of the analysis of organic materials of cultural heritage
Analytes Sample Techniques Comments Reference
Various Postage stamps from Portugal and colonies XRF; ATR-FTIR; UV-vis reflectance spectroscopy 23 specimens from Portugal and five from ex-colonies (four of which were forgeries) were analysed. Each sample was analysed in three positions representing dye/pigment, the paper and the cancellation. Data from XRF analyses were input to HCA and PCA. Some elements' (Ca, K, S and Si) data had to be removed from the dataset because they caused confusion. The HCA identified five clusters and these related the metallic (and non-metallic) elements to the colours of the stamps. Interestingly, the known forgeries aligned with the different colours and could not be distinguished using this methodology 329
Various 18th century Greek religious panel painting Macro-XRF Two data investigation modes used: elemental distribution maps constructed for the different analytes and a statistical analysis approach. The latter used PCA to reduce the dimensions of the data and then a non-linear t-distributed stochastic neighbour embedding (t-SNE) method. The statistical data analysis approach led to the grouping of all spectra into distinct clusters with common features. The two analytical approaches allowed detailed information about the pigments used and paint layer stratigraphy (i.e., painting technique) as well as restoration interventions/state of preservation to be extracted 330
Various Mural pigments LIBS A database of LIBS spectra (220–420 nm) was established from synthesised mineral pigments. A spectral matching algorithm comprising four methods was developed to identify the pigments. The four methods used in the algorithm were: Euclidean distances, spectral angle mapper, spectral correlation mapper and spectral information divergence (each was described briefly). The database was used successfully to identify pigments in the murals 331
Various Paintings Macro-XRF; photoluminescence; reflectance imaging spectroscopy An instrument was developed that could simultaneously analyse the same spot on a painting using the three techniques. Five datasets were obtained: XRF, reflectance imaging spectroscopy between 400 and 1000 nm, photoluminescence with irradiation at 250 nm, 365 nm and 655 nm 332
Various Murals LIBS The effect of buffer gas on crater size and signal intensity on analytes was studied. The gases air, argon, nitrogen, oxygen and helium were supplied by gas cylinder. Both crater size and discolouration area around the crater decreased in the order: air, argon, nitrogen, oxygen and helium. The signal was greatest in the argon atmosphere but SNR was best in helium. It was concluded that helium was probably the best atmosphere because it caused least damage and provided greatest stability 333
Various Mural pigments LIBS A comparison of machine learning methods for the identification of mural pigments. Algorithms compared were: k nearest neighbour, support vector machine, back propagation artificial neural network, convolutional neural network and random forest. A total of 900 spectra were taken from two mock murals containing relevant pigments. 600 spectra used for training of the algorithms and 300 to test them. For the mock murals, all except the k nearest neighbour algorithm had greater than 99% prediction accuracy. However, when analysing a real sample, the only algorithm that had a success rate of >90% (94%) was 2D-CNN 334
Various 26 Easel paintings by Giovanni Santi XRF; reflectance spectrometry The IR reflectography analysed the underdrawings whereas the XRF was used to analyse the pigments and dyes. Data were input to PCA and HCA. Numerous pigments and painting techniques were identified 335


6.3 Ceramic materials of cultural heritage

An “end to end” toolkit based on deep learning designed to aid in the classification of ceramic materials was described by Qi et al.336 The toolkit used the Fully Convolutional Network algorithm to analyse the relationships between the chemical compositions of the samples. In this case, the algorithm was applied to the analysis of 84 black glazed wares from three adjacent kilns (with a minimum distance apart of 1 km) that were operational during the Song dynasty. Ten XRF measurements were made of each glaze and a further 10 of each ceramic body. The paper discussed the theory of the Fully Convolutional Network at length. Similarly, the use of and theory behind the Matthews Correlation Coefficient was also explained. The results of the classification were compared with those obtained using the established models of Random Forest, k Nearest Neighbour, Decision Tree, XGBoost and Logistic Regression. The Fully Convolutional Network out-performed all of them in terms of both accuracy of classification (92.76%) and the Matthews Correlation Coefficient (89.14%). Given that the kilns were so close to each other, obtaining such a high success rate for classification was impressive.

Other applications of the analysis of ceramic cultural heritage samples are given in Table 5.

Table 5 Applications of the analysis of ceramic objects of cultural heritage
Analytes Sample Techniques Comments Reference
Various Porcelain from 19th century France PIXE; PIGE; SEM-EDS; XRD Glazes and ceramic bodies of 28 sherds representing three different production periods of the same company were analysed. Data input to PCA which clearly separated the periods. Bi-plots also used to discriminate between pieces 337
Various (17) Pottery from the Lucayan Islands LA-ICP-MS Major and trace elements were determined using LA-ICP-MS and the data input to PCA in an attempt to classify the pottery sherds and to identify from which of the islands it originated. Although the appearance of the pottery looks the same, the chemical components differed. The PCA identified nine main compositional groups which the authors attributed to originating in different islands. Three of the groups came from Hispaniola 338
Various Ceramics from Brazilian slave quarters XRF As well as the ceramics, clays from close to Rio de Janeiro were also analysed. Method validated using the reference materials SARM 69 and the plastic clay material PT32. Data from analyses were input to PCA and HCA. A clear-cut separation between clay sources and ceramic sherds was obtained, indicating that the ceramic vessels were not produced in the plantations' space 339
Various (33) Prehistoric ceramics from Korea Portable XRF; INAA; LA-ICP-MS; WDXRF A study of 92 ceramic sherds from several different periods. Most analyses undertaken using pXRF with the other techniques providing comparative data. Several (12) CRMs used for calibration. Seventeen analytes had a calibration correlation coefficient of 0.93 or higher. For these 17 analytes, the pXRF result was compared with those from LA-ICP-MS, WDXRF and INAA (for 24 samples). Results were mixed, with some analytes having good agreement and others not 340
Various (18) Pottery fragments from Spain XRF Several chemometric tools tested for their ability to classify the provenance of pottery sherds. The non-supervised technique PCA failed but the supervised machine learning techniques of K nearest neighbour, linear discriminant analysis, artificial neural network and random forest were more successful. A database was established that helped identify the provenance. This database is available to other scientists and is modifiable to suit their needs. Apparently, it needs no specialist knowledge in the chemometric tools 341
Various 149 pieces of Chalcolithic pottery from Portugal XRF; XRD Both PCA and hierarchical ascending classification (HAC) were used to analyse data. The PCA clearly identified pottery that was not of local manufacture but local material was less easily discriminated, with two groups being identified (one with higher silica, K and Rb and the other with higher Al, Fe and Ti). Both chemometric techniques indicated that there were multiple sources of the raw materials. The XRD data indicated a firing temperature of 700–800 °C throughout the period 342
Various 70 Neolithic pottery sherds from the Scheldt estuary and 10 local sediments LA-ICP-MS Calibration achieved using five US geological survey reference materials as well as NIST 610 and 612 glasses. Method validation was achieved using a pressed pellet of NIST SRM 679 Brick Clay which was run every two or three samples. The LA-ICP-MS results were input to HCA and PCA. Results were inconclusive unless they were complemented by petrographic analysis 343
Various (10) Archaeological ceramics TXRF Data from TXRF analysis of 81 ceramic sherds input to several chemometric and machine learning algorithms (PCA, k-means, cluster analysis and support vector machine (SVM)) in an attempt to elucidate provenance. Results indicated that PCA could not classify all samples, but gave useful indicators. Once trained using a database, the SVM and k-means classified most samples 344
Various 12 Chinese archaeological ceramics LIBS; ICP-OES The LIBS operating parameters (laser energy, delay time, etc.) were optimised prior to analysis. Data obtained were validated by comparison with those obtained using ICP-OES following a digestion protocol. The LIBS data were then input to a machine learning program comprising Sequence Backward Selection (a data reduction method described in the text) combined with random forest regression. The combination of algorithms out-performed partial least squares regression, support vector machine and random forest alone 345
Various Ancient pottery from the Brazilian Amazon coast XRF; PIXE; XRD; Mossbauer spectroscopy; Computed radiography Three archaeological sites and five clay sources analysed. For XRF measurements, IAEA–Soil 7 and plastic clay–IPT 32 were used to assess accuracy and reproducibility. Both XRF and PIXE data input to PCA and cluster analysis. Two clear groups were identified: Bacanga and Panaquatira sites were in one group and Rabo de Porco and clay from source D were in the other. The obvious conclusion was that clay from source D was used to make the pottery at Rabo de Porco. The pottery from the other two sites were made from none of the clays analysed. The other techniques identified that the firing temperature of the pottery was 750–900 °C. The computed radiography indicated different production technologies 346


6.4 Glass materials of cultural heritage

A summary of the applications for the analysis of glass cultural heritage samples is given in Table 6.
Table 6 Applications of the analysis of glass cultural heritage samples
Analytes Sample Techniques Comments Reference
Various (27 trace analytes and four major elements) Blue Roman glass artefacts LA-ICP-MS; EPMA; SEM-EDS Blue glass samples dating from between 1st century BC to 5th century AD were analysed. The ICP-MS determined the trace analytes whereas EPMA determined four major elements. NIST 610 was used to monitor and prevent elemental fractionation and as a calibrant. NIST 612 was used as a “test sample”. All glasses were identified as being soda-lime. The samples were clearly segregated into four groups by PCA using the first two principal components (which accounted for 93% of the variance). Information was gleaned regarding the flux, colourants and decolourants as well as the extent of recycling 347
Various (10) Medieval glasses from Spain LA-ICP-MS Fragments of natron glasses (268) were analysed using LA-ICP-MS. Calibration was achieved using the reference materials NIST 610, Corning B, C and D and APL1. Silicon was used as an internal standard. Data from 10 analytes (Al, Ca, K, Li, Mg, Na, Rb, Th, Ti and Zr) were input to PCA which identified 7 sub-classes. There was significant evidence of glass recycling as indicated by an enrichment of Li 348
Various (15) Glass paintings from the Lipari collection Portable XRF Data were treated before being input to PCA and HCA. The pre-treatments were either normalisation to the Si signal or the net area percentages of the analytes were estimated from each spectrum. Net peak area values were obtained after peak deconvolution and background subtraction procedures. Finally, each spectrum was normalized with respect to its total area. The net peak area values yielded the more reliable classification 349
Various (34) Glass ingots from Uluburun shipwreck LA-ICP-MS 192 samples of greenish-blue or purplish-blue glass ingots, five Mycaenean relief beads and 373 late bronze age glass samples from Amarna (Egypt) were analysed using LA-ICP-MS. Reference glasses NIST 612 and Corning A were used to assess accuracy and precision. Data were input to PCA and cluster analysis. These identified that the large majority of glass ingots were not of the same type as the glass from Amarna. The cobalt-rich Mycaenean beads were of closer match to the Uluburun glass ingots 350
Various Three reference glasses of known composition from the Corning museum of glass and five archaeological glasses Portable XRF Four portable XRF instruments from different manufacturers were compared. In general, the results compared fairly well as long as the data were treated in the same way. The data could be corrected by linear regression correction when four (or more) reference data points were available. When four reference data points were not available, ratio coefficients correction could be used on the fundamental parameters data 351
Various Corroded Roman glass LA-ICP-MS 2D and 3D analysis of corroded Roman glass sherds. Eight reference materials used for calibration (Corning A–D, Society of Glass Technology Glass Standards 7, 10, and 11 and NIST SRM 612). A transfer system with a very rapid washout was used to give a better signal. The washout time required was 14.5 ± 0.5 ms for NIST 612 but 20 ± 11 ms for a sample. This was attributed to heterogeneity in the real samples. Lateral resolution of 20 μm was obtained. Corrosion layers were enriched in Al, Na and Si 352


7 Abbreviations

2Dtwo dimensional
3Dthree dimensional
AASatomic absorption spectrometry
ABSacrylonitrile butadiene
AESAuger electron spectrometry
AFFFasymmetric field flow fractionation
AF4asymmetric flow–field flow fractionation
AFSatomic fluorescence spectrometry
AFMatomic force microscopy
AMSaccelerator mass spectrometry
ANOVAanalysis of variants
ARECVaverage relative error of cross-validation
ASTMAmerican Society for Testing of Materials
ATRattenuated total reflection
BCRCommunity Bureau of Reference
CCDcharge coupled device
CEcapillary electrophoresis
CFFFcentrifugal field flow fractionation
CIGScopper indium gallium selenide
CRMcertified reference material
CScontinuum source
CTcomputerised tomography
CVcold vapour
DAdiscriminant analysis
DLSdynamic light scattering
DLTVdiode laser thermal vaporisation
DP-RLIBSdouble pulse resonance laser induced breakdown spectrometry
DMSOdimethyl sulfoxide
DRCdynamic reaction cell
EASTexperimental advanced super conducting tokamak
EDIXSenergy dispersive inelastic X-ray scattering
EDSenergy dispersive spectrometry
EDXRDenergy dispersive X-ray diffraction
EDXRFenergy dispersive X-ray fluorescence
ELMextreme learning machine
EPMAelectron probe microanalysis
ESI-MSelectrospray ionisation mass spectrometry
ETAASelectrothermal atomic absorption spectrometry
ETVelectrothermal vaporisation
EXAFSextended X-ray absorption fine structure
FAASflame atomic absorption spectrometry
FESEMfield emission scanning electron microscopy
FFFFflow field flow fractionation
FIflow injection
FIBfocused ion beam
FI-CVGflow injection chemical vapour generation
FTIRFourier transform infrared
FWHMfull width at half maximum
GCgas chromatography
GD-MSglow discharge mass spectrometry
GD-OESglow discharge optical emission spectrometry
GF-AASgraphite furnace atomic absorption spectrometry
GI-SAXSgrazing incidence small angle X-ray scattering
GIXRDgrazing incidence X-ray diffraction
GIXRFgrazing incidence X-ray fluorescence
HAXPEShard X-ray photoelectron spectroscopy
HDChydrodynamic chromatography
HGhydride generation
HPLChigh performance liquid chromatography
HR-CSAAShigh resolution continuum source atomic absorption spectrometry
IAEAInternational Atomic Energy Agency
IBAion beam analysis
ICAindependent component analysis
ICPinductively coupled plasma
ICP-MSinductively coupled plasma mass spectrometry
ICP-OESinductively coupled plasma optical emission spectrometry
ICP-QMSinductively coupled plasma quadrupole mass spectrometry
ICP-TOFMSinductively coupled plasma time-of-flight mass spectrometry
IDisotope dilution
IL-DLLMEionic liquid-dispersive liquid–liquid microextraction
IPInstitute of Petroleum
IRMSisotope ratio mass spectrometry
ISOInternational Organisation for Standardisation
ITERInternational thermonuclear experimental reactor
JETJoint European Torus
K-SVM-RFE k-fold support vector machine recursive feature elimination
LAlaser ablation
LASILlaser ablation of sample in liquid
LCliquid chromatography
LDAlinear discriminant analysis
LIASlaser induced ablation spectrometry
LIBSlaser induced breakdown spectrometry
LIBS-LAMSlaser induced breakdown spectrometry-laser ablation mass spectrometry
LIFlaser induced fluorescence
LIPSlaser induced plasma spectroscopy
LODlimit of detection
LOQlimit of quantification
LTElocal thermal equilibrium
MALDI-TOFmatrix-assisted laser desorption ionisation time-of-flight
MALSmulti-angle light scattering
MASmagic angle spinning
MCmulticollector
MEISmedium energy ion scattering spectroscopy
MIBKmethyl isobutyl ketone
MIPmicrowave induced plasma
MIP-AESmicrowave plasma atomic emission spectrometry
MSmass spectrometry
MSFAmono-segmented flow analysis
MWCNTmulti-wall carbon nanotubes
NAAneutron activation analysis
NAARneutron activation autoradiography
NDneutron diffraction
NDI-MSnear-field desorption ionization mass spectrometry
Nd:YAGneodymium doped-yttrium aluminium garnet
NEXAFSnear edge X-ray fine structure
NISTNational Institute of Standards and Technology
NMRnuclear magnetic resonance
NRAnuclear reaction analysis
OESoptical emission spectrometry
PARAFACparallel factor analysis
PBSphosphate buffered saline
PCAprincipal component analysis
PCRprincipal component regression
PETpolyethylene terephthalate
PFCplasma facing components
PGAAprompt gamma neutron activation analysis
PGMplatinum group metals
PIGEparticle induced gamma ray emission
PIXEparticle-induced X-ray emission
PLALpulsed laser ablation in liquids
PLSpartial least squares
PLS-DApartial least squares discriminant analysis
PLSRpartial least squares regression
ppbparts per billion
ppmparts per million
RAFMreduced activation ferritic/martensitic
RBSRutherford backscattering spectrometry
RDAregularised discriminant analysis
REErare earth elements
rfradiofrequency
RIMSresonance ionisation mass spectrometry
RMSECVroot mean square error of cross validation
RMSEProot mean squared error of prediction
RoHSrestriction of hazardous substances
RSDrelative standard deviation
RSFrelative sensitivity factor
SDS-PAGEsodium dodecyl sulfate–polyacrylamide gel electrophoresis
SECsize exclusion chromatography
SEMscanning electron microscopy
SEM-EDSscanning electron microscopy-energy dispersive spectrometry
SFsector field
SIAsequential injection analysis
SIBSspark induced breakdown spectrometry
SIMCAsoft independent modelling of class analogy
SIMSsecondary ion mass spectrometry
SNRsignal to noise ratio
SPsingle particle
SPIONsuperparamagnetic iron oxide nanoparticles
SRsynchrotron radiation
SRMstandard reference material
SXRFsynchrotron X-ray fluorescence
SVRsupport vector regression
STXMscanning transmission X-ray microscopy
TEMtransmission electron microscopy
TDSthermal desorption mass spectrometry
TGAthermogravimetric analysis
TIMSthermal ionisation mass spectrometry
TOFtime of flight
T-PGAAtime-resolved prompt gamma activation analysis
TPRtemperature programmed reduction
TXRFtotal reflection X-ray fluorescence
UAEultra-sound assisted extraction
UV-visultraviolet-visible
WDXRFwavelength dispersive X-ray fluorescence
WEEEwaste electrical and electronic equipment
XAFSX-ray absorption fine structure spectrometry
XANESX-ray absorption near edge structure
XASX-ray absorption spectroscopy
XPSX-ray photoelectron spectroscopy
XRDX-ray diffraction
XRFX-ray fluorescence
XRPDX-ray powder diffraction
XRRX-ray reflectometry

Conflicts of interest

There are no conflicts to declare.

References

  1. S. Carter, R. Clough, A. Fisher, B. Gibson and B. Russell, J. Anal. At. Spectrom., 2022, 37(11), 2207–2281 RSC.
  2. E. H. Evans, J. Pisonero, C. M. M. Smith and R. N. Taylor, J. Anal. At. Spectrom., 2022, 37(5), 942–965 RSC.
  3. C. Vanhoof, J. R. Bacon, U. E. A. Fittschen and L. Vincze, J. Anal. At. Spectrom., 2022, 37(9), 1761–1775 RSC.
  4. J. R. Bacon, O. T. Butler, W. R. L. Cairns, O. Cavoura, J. M. Cook, C. M. Davidson and R. Mertz-Kraus, J. Anal. At. Spectrom., 2023, 38(1), 10–56 RSC.
  5. R. Clough, C. F. Harrington, S. J. Hill, Y. Madrid and J. F. Tyson, J. Anal. At. Spectrom., 2022, 37(7), 1387–1430 RSC.
  6. Z. H. Khan, M. H. Ullah, B. Rahman, A. I. Talukder, M. Wahadoszamen, K. M. Abedin and A. F. M. Y. Haider, J. Spectrosc., 2022, 2022, 3887038,  DOI:10.1155/2022/3887038.
  7. V. Gardette, V. Motto-Ros, C. Alvarez-Llamas, L. Sancey, L. Duponchel and B. Busser, Anal. Chem., 2023, 95(1), 49–69 CrossRef CAS PubMed.
  8. L. Brunnbauer, Z. Gajarska, H. Lohninger and A. Limbeck, TrAC, Trends Anal. Chem., 2023, 159, 116859 CrossRef CAS.
  9. N. Zhang, T. X. Ou, M. Wang, Z. J. Lin, C. Lv, Y. Z. Qin, J. M. Li, H. Yang, N. Zhao and Q. M. Zhang, Front. Phys., 2022, 10, 887171 CrossRef.
  10. N. H. Cheung, Spectrochim. Acta, Part B, 2022, 194, 106473 CrossRef CAS.
  11. J. Willner, L. Brunnbauer, S. Larisegger, M. Nelhiebel, M. Marchetti-Deschmann and A. Limbeck, Talanta, 2023, 256, 124305 CrossRef CAS PubMed.
  12. J. Peterson, M. Gilbert-Gatty, K. Ekstrom, L. Hagesjo and A. Bengtson, Appl. Spectrosc., 2023, 77(8), 907–914 CrossRef CAS PubMed.
  13. M. V. Belkov, D. A. Borisevich, K. Y. Katsalap and M. A. Khodasevich, Opt. Spectrosc., 2022, 130(8), 488–494 CrossRef CAS.
  14. M. V. Belkov, V. V. Kiris and K. Y. Catsalap, J. Appl. Spectrosc., 2023, 89(6), 1040–1046 CrossRef CAS.
  15. S. J. Shen, P. Chen, H. Luo, Z. Z. Wang, J. J. Yan and Y. Deguchi, J. Phys. D: Appl. Phys., 2023, 56(7), 075201 CrossRef.
  16. A. Fayyaz, U. Liaqat, K. Yaqoob, R. Ahmed, Z. A. Umar and M. A. Baig, Spectrochim. Acta, Part B, 2022, 198, 106562 CrossRef CAS.
  17. H. Jun, Y. Liang, T. Z. Lin, H. Y. Shi, Q. Peng, H. Yong and L. Yande, Opt. Express, 2022, 30(26), 47970–47982 CrossRef PubMed.
  18. L. Sheng, L. J. Yuan, Y. H. Jia, L. Zhao, X. F. Zhang, L. Yu, Q. C. Zhang and H. Z. Wang, J. Anal. At. Spectrom., 2022, 37(7), 1522–1534 RSC.
  19. A. G. Ouyang, T. Z. Lin, J. Hu, B. Yu and Y. D. Liu, Spectrosc. Spectral Anal., 2022, 42(10), 3109–3115 CAS.
  20. J. Hu, L. Yang, T. Z. Lin, H. Y. Shi, P. Qiao, Y. He and Y. D. Liu, J. Anal. At. Spectrom., 2022, 37(11), 2309–2319 RSC.
  21. Z. Li, Q. Y. Zhang, L. H. Kong, G. F. Lian and P. Li, Chin. J. Lasers, 2022, 49(9), 0911002 CrossRef.
  22. Y. Tang, J. F. Li, S. X. Ma, Z. L. Hu, X. X. Peng, W. P. Zhou, X. Yuan and M. Shen, Plasma Sci. Technol., 2023, 25(6), 065505 CrossRef.
  23. W. Y. Bai, W. F. Chen, C. J. Yang, Y. Hong and R. X. Wang, Laser Optoelectron. Prog., 2022, 59(23), 2330001 CrossRef.
  24. I. Traparic and M. Ivkovic, Eur. Phys. J. D, 2023, 77(2), 30 CrossRef CAS.
  25. Q. D. Zeng, G. H. Chen, W. X. Li, Z. T. Li, J. H. Tong, M. T. Yuan, B. Y. Wang, H. H. Ma, Y. Liu, L. B. Guo and H. Q. Yu, Plasma Sci. Technol., 2022, 24(8), 084009 CrossRef.
  26. L. Y. Yang, Y. Ding, Y. Zhan, S. N. Zhu, Y. J. Chen, F. Deng and X. Q. Zhao, Spectrosc. Spectral Anal., 2022, 42(6), 1804–1808 CAS.
  27. Z. L. Hu, J. F. Nie, F. Chen, D. Zhang, Z. Y. Ouyang, Y. C. Liu, Y. W. Chu and L. B. Guo, J. Anal. At. Spectrom., 2022, 37(12), 2573–2580 RSC.
  28. J. J. Valdiande, M. Martinez-Minchero, A. Cobo, J. M. Lopez-Higuera and J. Mirapeix, Spectrochim. Acta, Part B, 2022, 194, 106474 CrossRef CAS.
  29. L. Quackatz, A. Griesche and T. Kannengiesser, Spectrochim. Acta, Part B, 2022, 193, 106439 CrossRef CAS.
  30. M. C. Cui, H. R. Guo, Y. D. Chi, L. Tan, C. F. Yao, D. H. Zhang and Y. Deguchi, Spectrochim. Acta, Part B, 2022, 191, 106398 CrossRef CAS.
  31. S. Grunberger, V. Ehrentraut, S. Eschlbock-Fuchs, J. Hofstadler, A. Pissenberger and J. D. Pedarnig, Anal. Chim. Acta, 2023, 1251, 341005 CrossRef PubMed.
  32. X. Y. He, Q. Yang, D. X. Ling, D. S. Wei and H. C. Wang, Chemosensors, 2022, 10(9), 370 CrossRef CAS.
  33. K. Sanyal and S. Dhara, Anal. Sci., 2022, 38(4), 665–673 CrossRef CAS PubMed.
  34. J. Long, W. R. Song, Z. Y. Hou and Z. Wang, Plasma Sci. Technol., 2023, 25(7), 075501 CrossRef.
  35. W. L. Gu, Z. Y. Hou, W. R. Song, L. Li, X. Yu, J. C. Liu, Y. Z. Song, M. S. Afgan, Z. Li, Z. J. Liu and Z. Wang, Anal. Chim. Acta, 2022, 1205, 339752 CrossRef CAS PubMed.
  36. J. C. Liu, W. R. Song, W. L. Gu, Z. Y. Hou, K. K. Kou and Z. Wang, Anal. Chim. Acta, 2023, 1251, 341004 CrossRef CAS PubMed.
  37. S. J. Li, R. E. Zheng, Y. Deguchi, W. Q. Ye, Y. Tian, J. J. Guo, Y. Li and Y. Lu, Plasma Sci. Technol., 2023, 25(4), 045510 CrossRef.
  38. M. P. Mateo and G. Nicolas, Anal. Chim. Acta, 2022, 1227, 340260 CrossRef CAS PubMed.
  39. J. Iqbal, T. A. Alrabdi, A. Fayyaz, H. Asghar, S. K. H. Shah and M. Naeem, Laser Phys., 2023, 33(3), 036001 CrossRef.
  40. Y. F. Li, Y. Q. Chen, S. S. Li and X. Q. Huang, Opt. Express, 2022, 30(19), 34545–34556 CrossRef CAS PubMed.
  41. S. X. Ma, Y. C. Liu, H. W. Tian, L. B. Guo and D. M. Dong, J. Anal. At. Spectrom., 2023, 38(2), 342–346 RSC.
  42. B. H. Xu, Z. J. Lin, Z. Y. Xu, X. T. Yan, X. Y. Peng, C. K. Li, Y. Z. Qin, J. M. Li, N. Zhao and Q. M. Zhang, J. Anal. At. Spectrom., 2023, 38(5), 1032–1042 RSC.
  43. Z. J. Lin, N. Zhang, Z. Y. Xu, J. P. Liao, H. Yuan, E. Chenshen, J. M. Liu, J. M. Li, N. Zhao and Q. M. Zhang, J. Anal. At. Spectrom., 2022, 37(10), 2082–2088 RSC.
  44. C. Lv, N. Zhang, Z. J. Lin, T. X. Ou, J. M. Li, N. Zhao, X. Y. Yang, Q. X. Ma, L. Guo and Q. M. Zhang, J. Laser Appl., 2023, 35(1), 012021 CrossRef CAS.
  45. J. Q. Yao, Q. Yang, X. Y. He, J. L. Li, D. X. Ling, D. S. Wei and Y. P. Liao, RSC Adv., 2022, 12(50), 32230–32236 RSC.
  46. J. L. Li, Q. Yang, J. Q. Yao, X. Y. He and F. J. Wang, Anal. Chim. Acta, 2023, 1238, 340613 CrossRef CAS PubMed.
  47. Q. Yang, X. Y. He, D. X. Ling, Z. C. Wei, D. S. Wei and Q. N. Zhang, Spectrochim. Acta, Part B, 2023, 199, 106599 CrossRef CAS.
  48. R. A. Awan, K. Siraj, S. U. Haq, Q. Abbas, M. S. A. Rahim, Q. Younas, S. Fareed, R. Ahsen, Z. Ahmad, M. Irshad and A. Latif, Opt. Quantum Electron., 2023, 55(1), 73 CrossRef CAS.
  49. N. Ahmed, J. A. Awan, K. Fatima, S. M. Z. Iqbal, M. Rafique, S. A. Abbasi and M. A. Baig, Eur. Phys. J. Plus, 2022, 137(6), 671 CrossRef CAS.
  50. D. J. Diaz-Romero, S. Van den Eynde, W. Sterkens, A. Eckert, I. Zaplana, T. Goedeme and J. Peeters, Spectrochim. Acta, Part B, 2022, 196, 106519 CrossRef CAS.
  51. Y. Ding, L. Y. Yang, J. Chen, X. Y. Wang, X. R. Guo, X. C. Xu, X. Q. Zhao, Y. Luo and W. J. Chen, Laser Optoelectron. Prog., 2022, 59(13), 1314006 CrossRef.
  52. Z. L. Hu, J. F. Nie, Z. Y. Ouyang, D. Zhang, Y. C. Liu, Y. W. Chu and L. B. Guo, Opt. Lett., 2023, 48(1), 1–4 CrossRef CAS PubMed.
  53. Z. Y. Hou, W. L. Gu, T. Q. Li, Z. Wang, L. Li, X. Yu, Y. C. Zhang and Z. J. Liu, Front. Phys., 2022, 17(6), 62503 CrossRef.
  54. L. M. John, R. C. Issac, S. Sankararaman and K. K. Anoop, J. Anal. At. Spectrom., 2022, 37(11), 2451–2460 RSC.
  55. Q. Zou, C. Li, Z. J. Ding, Z. Y. Bao, T. Zhang, Q. B. Zhang, L. L. Zhang and Y. F. Bi, J. Anal. At. Spectrom., 2023, 38(2), 429–436 RSC.
  56. L. Fu, G. S. Huang, Y. B. Hu, X. H. Chen, J. F. Wang and F. S. Pan, Anal. Chem., 2023, 95(11), 4950–4956 CrossRef CAS PubMed.
  57. S. Wagner, C. Hummel, J. Santner, M. Puschenreiter, J. Irrgeher, W. W. Wenzel, S. M. Borisov and T. Prohaska, Anal. Chim. Acta, 2022, 1212, 339910 CrossRef CAS PubMed.
  58. C. T. Cakir, T. Piotrowiak, U. Reinholz, A. Ludwig, F. Emmerling, C. Streli, A. G. Buzanich and M. Radtke, Anal. Chem., 2023, 95(10), 4810–4818 CrossRef CAS PubMed.
  59. H. Choi, H. Kim, S. H. Han, S. Kim, S. Jung, S. H. Nam and Y. Lee, Appl. Spectrosc., 2023, 77(4), 371–381 CrossRef CAS PubMed.
  60. A. V. Kuptsov, N. S. Medvedev, E. V. Polyakova, A. I. Saprykin and V. A. Labusov, Spectrochim. Acta, Part B, 2022, 194, 106475 CrossRef CAS.
  61. W. F. Zhang, X. P. Xia, S. Yuto, T. Eiichi, Z. X. Cui, Y. Q. Zhang, Q. Yang, Y. N. Yang, L. Li and Y. G. Xu, J. Anal. At. Spectrom., 2023, 38(3), 609–618 RSC.
  62. L. M. Narlagiri, M. S. S. Bharati, R. Beeram, D. Banerjee and V. R. Soma, TrAC, Trends Anal. Chem., 2022, 153, 116645 CrossRef CAS.
  63. A. Li, X. Y. Zhang, Y. S. Yin, X. S. Wang, Y. G. He, Y. H. Shan, Y. Zhang, X. D. Liu, L. X. Zhong and R. B. Liu, J. Anal. At. Spectrom., 2023, 38(4), 810–817 RSC.
  64. S. Choi, J. Choi and C. Park, Appl. Phys. B: Lasers Opt., 2023, 129(1), 8 CrossRef CAS.
  65. T. N. da Silva, D. B. Batista, B. F. Braz, A. S. Luna, R. E. Santelli, M. A. D. Fernandez and J. S. de Gois, Talanta, 2022, 250, 123718 CrossRef CAS PubMed.
  66. Q. Guo, H. Jiang, J. J. Yang, K. N. Wu and J. Man, Laser Optoelectron. Prog., 2022, 59(4), 0434001 CrossRef.
  67. H. Haekkaenen, K. Backfolk, P. Sirvio and J. Korppi-Tommola, Coatings, 2023, 13(2), 346 CrossRef.
  68. Z. Sun, X. Zhang, Y. Chen, E. Wan, Z. Chen and Y. Liu, At. Spectrosc., 2022, 43(5), 403–411 CAS.
  69. K. Wei, Q. Q. Wang, G. E. Teng, X. J. Xu, Z. F. Zhao and G. Y. Chen, Appl. Sci., 2022, 12(10), 4981 CrossRef CAS.
  70. V. Dwivedi, P. Chauhan, G. S. Maurya, A. M. Roldan, P. Veis and A. K. Pathak, Laser Phys., 2022, 32(7), 075604 CrossRef.
  71. A. H. Farhadian, S. S. Mousavifard and M. Mollaei, Laser Phys., 2022, 32(8), 085601 CrossRef.
  72. L. Whitty-Leveille, Z. L. VanAernum, J. A. Pavon, C. Murphy, K. Neal, W. Forest, X. L. Gao, W. Y. Zhong, D. D. Richardson and H. A. Schuessler, mAbs, 2023, 15(1), 2199466,  DOI:10.1080/19420862.2023.2199466.
  73. J. Yamazaki, I. Inoue, A. Arakawa, S. Karakawa, K. Takahashi and A. Nakayama, Anal. Methods, 2022, 14(22), 2219–2226 RSC.
  74. Z. Z. Yao, B. R. Li, C. Li, B. H. Wang, M. Zhao and Z. H. Ma, J. Anal. At. Spectrom., 2022, 37(8), 1652–1657 RSC.
  75. F. A. Redeker, J. E. Lesniewski, G. Hahm, W. P. McMahon and K. Jorabchi, Anal. Chem., 2022, 94(34), 11865–11872 CrossRef CAS PubMed.
  76. L. Zhu, Y. Teng, W. D. Han, L. H. Yin and X. Teng, Analyst, 2023, 148(5), 1116–1122 RSC.
  77. E. V. Chuparina, A. S. Maltsev, E. V. Stolpovskaya and N. A. Neverova, Spectrochim. Acta, Part B, 2022, 197, 106542 CrossRef CAS.
  78. J. F. S. Joca, R. Papai, L. C. Nunes, M. S. Luz, F. J. Krug and I. Gaubeur, Spectrochim. Acta, Part B, 2022, 198, 106566 CrossRef CAS.
  79. L. Rubio, P. Barrulas, M. Costa, C. Garcia-Jares, M. Lores and C. B. Dias, RSC Adv., 2022, 12(53), 34414–34424 RSC.
  80. M. da Costa, A. P. Benevides, D. V. Cesar, A. S. Luna and J. S. de Gois, Anal. Lett., 2023, 56(1), 132–147 CrossRef CAS.
  81. A. Pawlaczyk, M. Gajek, M. Balcerek and M. I. Szynkowska-Jozwik, Molecules, 2021, 26(21), 6753 CrossRef CAS PubMed.
  82. K. S. Coelho, E. S. Chaves, D. L. G. Borges, S. R. Krzyzaniak, R. C. C. Rocha, T. D. S. Pierre and V. Bascunan, Int. J. Environ. Anal. Chem., 2022 DOI:10.1080/03067319.2022.2098481.
  83. W. K. Zhu, X. H. Li, R. Sun, Y. H. Yan, M. F. Yuan, X. H. Ren, X. X. Meng and X. Yu, Combust. Flame, 2022, 244, 112237 CrossRef CAS.
  84. H. Rajavelu, N. J. Vasa and S. Seshadri, Appl. Phys. A: Mater. Sci. Process., 2022, 128(10), 868 CrossRef CAS.
  85. P. Lu, Z. Zhuo, W. H. Zhang, T. F. Sun, J. Tang and J. Q. Lu, J. Anal. At. Spectrom., 2022, 37(11), 2320–2329 RSC.
  86. Y. S. Ling, Y. Wang, W. B. Jia, Q. Shan, D. Q. Hei, J. D. Zhang, Z. C. Zhang and Z. Wang, Anal. Methods, 2023, 15(14), 1812–1818 RSC.
  87. Z. H. Tian, X. L. Li, G. Wang, L. Zhang, J. X. Li, S. Q. Wang, Y. Bai, W. F. Zhang, H. Yue, X. F. Ma, W. B. Yin and S. T. Jia, Plasma Sci. Technol., 2022, 24(8), 084007 CrossRef.
  88. Z. P. Wen, H. T. Liu, M. Q. Zhou, C. Liu and C. C. Zhou, Fuel, 2023, 332, 125991 CrossRef CAS.
  89. J. Z. Huang, Z. Q. Li, B. A. Chen, S. Cui, Z. L. Lu, W. Dai, Y. M. Zhao, C. L. Duan and L. Dong, J. Min. Inst., 2022, 256, 663–676 CrossRef.
  90. C. Liu, J. Y. Jiang, J. G. Jiang, Z. Z. Zhou and S. Ye, Spectroscopy, 2023, 38(2), 23–30 CAS.
  91. J. Chen, Q. Z. Li, K. Liu, X. Y. Li, B. Lu and G. Q. Li, J. Anal. At. Spectrom., 2022, 37(8), 1658–1664 RSC.
  92. F. A. Silva, B. R. Rigui, C. R. Andriolli, E. M. M. Flores, P. A. Mello and R. S. Picoloto, Talanta, 2023, 257, 124297 CrossRef CAS PubMed.
  93. A. Shishov, S. Savinov, N. Volodina, I. Gurev and A. Bulatov, Microchem. J., 2022, 179, 107456 CrossRef CAS.
  94. H. N. A. Altuwaijari, M. A. Farajzadeh, M. R. A. Mogaddam and S. M. Sorouraddin, Talanta, 2023, 257, 124378 CrossRef PubMed.
  95. B. P. Xu, S. M. Liu, B. Y. Lei, Y. H. Liu, W. F. Zhang, J. Tang, Y. S. Wang, W. Zhao and Y. X. Duan, J. Anal. At. Spectrom., 2022, 37(6), 1350–1359 RSC.
  96. Y. S. Liu, Z. J. Shi, C. X. Chen, Y. Hardalupas and A. Taylor, Exp. Therm. Fluid Sci., 2022, 136, 110652 CrossRef CAS.
  97. T. Varga, I. Hajdas, L. Calcagnile, G. Quarta, I. Major, A. J. T. Jull, A. Molnar and M. Molnar, Radiocarbon, 2023, 65(2), 539–548 CrossRef CAS.
  98. A. Mere, M. Enrico, H. G. Zhou, E. Tessier and B. Bouyssiere, ACS Omega, 2022, 7(43), 38150–38157 CrossRef CAS PubMed.
  99. T. H. Gao, T. X. Ren, Y. J. Zhou, P. S. Song and S. Wang, J. Anal. At. Spectrom., 2023, 38(4), 893–901 RSC.
  100. A. Rogoll, K. Heimler, D. Friebel, E. Brendler and C. Vogt, J. Anal. At. Spectrom., 2022, 37(12), 2647–2661 RSC.
  101. U. K. Adarsh, V. B. Kartha, C. Santhosh and V. K. Unnikrishnan, TrAC, Trends Anal. Chem., 2022, 149, 116534 CrossRef CAS.
  102. M. Alghamdi, M. A. E. Abdallah and S. Harrad, Emerging Contam., 2022, 8, 9–20 CrossRef CAS.
  103. S. Baidurah, Polymers, 2022, 14(22), 4928 CrossRef CAS PubMed.
  104. A. K. Tarai and M. K. Gundawar, presented at IEEE Workshop on Recent Advances in Photonics, WRAP, Mumbai, India, 2022,  DOI:10.1109/WRAP54064.2022.9758392.
  105. J. F. Nie, X. L. Wen, X. C. Niu, Y. W. Chu, F. Chen, W. L. Wang, D. Zhang, Z. L. Hu, J. L. Xiao and L. B. Guo, Polym. Test., 2022, 112, 107624 CrossRef CAS.
  106. U. K. Adarsh, E. B. Gowd, A. Bankapur, V. B. Kartha, S. Chidangil and V. K. Unnikrishnan, Waste Manage., 2022, 150, 339–351 CrossRef CAS PubMed.
  107. H. K. Angeyo and S. Gari, Appl. Radiat. Isot., 2022, 186, 110274 CrossRef CAS PubMed.
  108. P. Porizka, L. Brunnbauer, M. Porkert, U. Rozman, G. Marolt, D. Holub, M. Kizovsky, M. Benesova, O. Samek, A. Limbeck, J. Kaiser and G. Kalcikova, Chemosphere, 2023, 313, 137373 CrossRef CAS PubMed.
  109. A. Aynard, C. Courreges, J. Jimenez-Lamana, A. Raad, C. Miqueu, B. Grassl and S. Reynaud, Environ. Pollut., 2023, 323, 121229 CrossRef CAS PubMed.
  110. J. D. Chen, X. M. Jiang, Y. F. Tian and X. D. Hou, J. Anal. At. Spectrom., 2023, 38(2), 276–280 RSC.
  111. F. Gelman, M. Muszynska, J. Karasinski, O. Lev and L. Halicz, J. Anal. At. Spectrom., 2022, 37(11), 2282–2285 RSC.
  112. R. Giugliano, B. Cocciaro, F. Poggialini, S. Legnaioli, V. Palleschi, M. Locritani and S. Merlino, Sensors, 2022, 22(18), 6910 CrossRef CAS PubMed.
  113. C. Sommer, J. Nguyen, T. Menzel, J. A. Prume, H. Ruckdaschel and M. Koch, Polym. Test., 2022, 112, 107623 CrossRef CAS.
  114. L. M. Magana-Maldonado, K. Wrobel, T. L. E. Cruz, E. Y. Barrientos, A. R. C. Escobosa and K. Wrobel, Chemosphere, 2023, 313, 137316 CrossRef CAS PubMed.
  115. N. Gilon, M. Soyer, M. Redon and P. Fauvet, Sensors, 2023, 23(5), 2648 CrossRef CAS PubMed.
  116. K. Sanjana, M. S. Babu, R. Sarathi and N. Chillu, IEEE Access, 2023, 11, 1752–1760 Search PubMed.
  117. R. A. Krivina, M. Zlatar, T. N. Stovall, G. A. Lindquist, D. Escalera-Lopez, A. K. Cook, J. E. Hutchison, S. Cherevko and S. W. Boettcher, ACS Catal., 2023, 13, 902–915 CrossRef CAS.
  118. Y. J. Ko, M. H. Han, C. Lim, S. H. Yu, C. H. Choi, B. K. Min, J. Y. Choi, W. H. Lee and H. S. Oh, J. Energy Chem., 2023, 77, 54–61 CrossRef CAS.
  119. M. Zlatar, D. Nater, D. Escalera-Lopez, R. M. Joy, P. Pobedinskas, K. Haenen, C. Coperet and S. Cherevko, Electrochim. Acta, 2023, 444, 141982 CrossRef CAS.
  120. M. E. Kreider, G. A. Kamat, J. A. Z. Zeledon, L. Z. Wei, D. Sokaras, A. Gallo, M. B. Stevens and T. F. Jaramillo, J. Am. Chem. Soc., 2022, 144(49), 22549–22561 CrossRef CAS PubMed.
  121. E. Hornberger, M. Klingenhof, S. Polani, P. Paciok, A. Kormanyos, R. Chattot, K. E. MacArthur, X. L. Wang, L. J. Pan, J. Drnec, S. Cherevko, M. Heggen, R. E. Dunin-Borkowski and P. Strasser, Chem. Sci., 2022, 13(32), 9295–9304 RSC.
  122. M. H. Han, Y. J. Ko, B. K. Min, S. H. Yu, C. H. Choi, J. Y. Choi, W. H. Lee and H. S. Oh, Electrochem. Commun., 2022, 139, 107301 CrossRef.
  123. H. Wang, Y. W. Yang, Z. Y. Li, X. H. Kong, P. Martin, G. Q. Cui and R. X. Wang, Int. J. Hydrogen Energy, 2023, 48(24), 8921–8931 CrossRef CAS.
  124. L. J. Fernandez-Menendez, C. Mendez-Lopez, C. Gonzalez-Gago, J. Pisonero and N. Bordel, J. Anal. At. Spectrom., 2023, 38(2), 325–332 RSC.
  125. Z. Zhang, J. Wu, Y. H. Hang, Y. Zhou, Z. J. Tang, M. X. Shi, Y. Qiu, K. X. Liao, T. Liu and X. W. Li, Spectrochim. Acta, Part B, 2022, 191, 106392 CrossRef CAS.
  126. P. Bran-Anleu, T. Wangler, V. N. Nerella, V. Mechtcherine, P. Trtik and R. J. Flatt, Mater. Struct., 2023, 56(3), 51 CrossRef CAS PubMed.
  127. C. Chang, F. Di Maio, P. Rem, A. T. Gebremariam, F. Mehari and H. Xia, Resour., Conserv. Recycl., 2022, 185, 106507 CrossRef CAS.
  128. M. Nedeljkovic, N. Tosic, P. Holthuizen, F. F. de Mendonca, O. Copuroglu, E. Schlangen and S. Fennis, Mater. Struct., 2023, 56(3), 54 CrossRef CAS.
  129. Z. Aljobeh, R. Gillman and T. Kolba, J. Test. Eval., 2022, 50(5), 2501–2506 CAS.
  130. V. K. Do, T. Furuse, Y. Ohta, H. Iwahashi, T. Hirosawa, M. Watanabe and S. Sato, J. Radioanal. Nucl. Chem., 2022, 331(12), 5631–5640 CrossRef CAS.
  131. B. Gomez-Nieto, C. Isabel-Cabrera, M. J. Gismera, M. T. Sevilla, J. R. Procopio and M. I. S. de Rojas, Anal. Methods, 2023, 15(9), 1105–1115 RSC.
  132. Z. Bunina, V. Varchenko, K. Bryleva, V. Chornyi and K. Belikov, ChemistrySelect, 2022, 7(29), e202201457 CrossRef CAS.
  133. J. Park, S. Kumar, S. H. Han, V. K. Singh, S. H. Nam and Y. Lee, Appl. Spectrosc., 2022, 76(9), 1042–1050 CrossRef CAS PubMed.
  134. X. Y. Ding, X. Teng, Z. X. She, Y. Li, Y. Y. Liu, Y. Zhuang and C. C. Wang, RSC Adv., 2022, 12(50), 32526–32533 RSC.
  135. Y. S. Ling, Z. Wang, W. B. Jia, Q. Shan, Z. C. Zhang, D. Q. Hei and Y. Wang, Anal. Lett., 2023, 56(11), 1803–1813 CrossRef CAS.
  136. J. S. Silva, G. M. Heidrich, B. O. Poletto, J. N. G. Paniz, V. L. Dressler and E. M. M. Flores, J. Anal. At. Spectrom., 2023, 38(5), 1000–1006 RSC.
  137. T. Y. Guselnikova, A. R. Tsygankova, N. S. Medvedev and A. I. Saprykin, At. Spectrosc., 2022, 43(6), 459–464 CAS.
  138. W. X. Cui, Z. Q. Cai, Q. Li, H. Y. Qu, J. Q. Zheng, D. G. Yu, J. F. Chen and Z. Wang, Talanta, 2023, 255, 124248 CrossRef CAS PubMed.
  139. R. E. Pinson, A. V. Giminaro, C. L. Dugan, P. R. Jenkins and A. K. Patnaik, Appl. Opt., 2023, 62(6), 118–126 CrossRef PubMed.
  140. P. Zhang, L. X. Sun, H. B. Yu and P. Zeng, Plasma Sci. Technol., 2022, 24(8), 084006 CrossRef.
  141. J. Park, S. Kumar, S. H. Han, V. K. Singh, S. H. Nam, C. Y. Yoo and Y. Lee, Optik, 2021, 247, 167919 CrossRef CAS.
  142. Y. Ikeda, J. K. Soriano and I. Wakaida, Opt. Laser Technol., 2023, 159, 108982 CrossRef CAS.
  143. Y. Wang and J. Q. Lu, Chin. J. Lasers, 2022, 49(21), 2111001 CrossRef.
  144. D. Spanu, A. Palestra, V. Prina, D. Monticelli, S. Bonanomi, S. U. Nanot, G. Binda, L. Rampazzi, G. Sessa, D. C. Munoz and S. Recchia, Molecules, 2023, 28(6), 2845 CrossRef CAS PubMed.
  145. A. A. Arkhipenko, E. S. Koshel and V. B. Baranovskaya, Inorg. Mater., 2022, 58(14), 1472–1478 CrossRef CAS.
  146. E. N. Carlson and R. L. Hervig, Am. Mineral., 2023, 108(3), 447–454 CrossRef.
  147. L. Michaliszyn, A. Pramann, A. Rothke and O. Rienitz, J. Anal. At. Spectrom., 2022, 37(11), 2442–2450 RSC.
  148. T. Singh, D. Upadhyay and B. Mishra, Int. J. Mass Spectrom., 2023, 485, 117010 CrossRef CAS.
  149. Y. Saijo, M. Murata, T. Kajihara, H. Hijiya, Y. Suzuki, R. Akiyama, M. Shimizu, Y. Shimotsuma and K. Miura, Anal. Sci., 2022, 38(6), 881–888 CrossRef CAS PubMed.
  150. P. Kumar, M. C. Mathpal, S. Ghosh, G. K. Inwati, J. R. Maze, M. M. Duvenhage, W. D. Roos and H. C. Swart, J. Alloys Compd., 2022, 909, 164789 CrossRef CAS.
  151. V. Mishra, O. Gedeon and T. Gavenda, J. Am. Ceram. Soc., 2023, 106(6), 3426–3437 CrossRef CAS.
  152. G. Agnello, C. Cushman, A. C. Antony, N. Smith, J. Banerjee, H. H. Park, D. Cabrera, M. Geiger, C. Simons, H. Jing and W. Walczak, J. Non-Cryst. Solids, 2022, 592, 121766 CrossRef CAS.
  153. G. Pintori, S. Panighello, O. Pinato and E. Cattaruzza, Int. J. Appl. Glass Sci., 2023, 14(3), 468–479 CrossRef.
  154. Z. H. Jiang, X. Ding, M. S. Song and M. X. Ling, Ceram. Int., 2022, 48(19), 28404–28412 CrossRef CAS.
  155. Z. L. Chen, P. Huang, Y. Yang, W. X. Jiang, C. F. Lai and S. M. Peng, Fusion Eng. Des., 2022, 184, 113303 CrossRef CAS.
  156. W. H. Zhang, Q. Z. Zhao, P. Hu, M. He, Y. W. Bao and K. N. Li, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 530, 39–42 CrossRef CAS.
  157. I. Jogi, J. Ristkok, J. Raud, J. Butikova, K. Mizohata and P. Paris, Fusion Eng. Des., 2022, 179, 113131 CrossRef CAS.
  158. J. Kadi, M. Palankai, J. Szoke and G. Veres, Fusion Eng. Des., 2023, 191, 113519 CrossRef CAS.
  159. X. Y. Ding, H. Y. Zheng, P. P. Zhang, L. M. Luo, Y. C. Wu and J. H. Yao, J. Mater. Eng. Perform., 2023, 32, 7869–7878 CrossRef CAS.
  160. L. Y. Sun, D. Wu, C. Li, J. Wu, S. H. Hong, E. Bang, Z. H. Hu, F. Ding, G. N. Luo and H. B. Ding, Nucl. Mater. Energy, 2022, 31, 101174 CrossRef CAS.
  161. M. Imran, Z. H. Hu, F. Ding, M. H. Li, G. N. Luo, H. Sattar, M. Iqbal and H. B. Ding, Spectrochim. Acta, Part B, 2022, 198, 106568 CrossRef CAS.
  162. J. M. Liu, D. Wu, X. H. Hu, S. M. Liu, H. C. Wu, R. Hai, C. Li and H. B. Ding, Spectrochim. Acta, Part B, 2023, 199, 106569 CrossRef CAS.
  163. E. Wust, T. Dittmar, C. Kawan, J. Romazanov, S. Brezinsek and W. X. Team, Nucl. Mater. Energy, 2022, 33, 101287 CrossRef.
  164. S. K. Lee, S. Paek, J. H. Lee, Y. H. Kim, E. Y. Choi and C. H. Lee, J. Nucl. Mater., 2022, 572, 154076 CrossRef CAS.
  165. Z. Y. Wang, Y. B. Shen, Y. S. Zhang, Y. H. Liu, Z. M. Dong, M. Y. Quan, F. T. Yu, X. H. Cao, Z. B. Zhang, Y. C. Wang and Y. H. Liu, J. Radioanal. Nucl. Chem., 2023, 332(4), 1353–1365 CrossRef CAS.
  166. Y. C. Wang, Q. Liu, S. Zhang, Y. H. Liu, Y. Q. Wang, Y. Dai, Z. M. Dong, Z. P. Cheng, X. H. Cao, Y. Q. Chen, Z. B. Zhang and Y. H. Liu, Sep. Purif. Technol., 2022, 294, 120972 CrossRef CAS.
  167. D. D. Wang, Y. L. Liu, S. L. Jiang, Y. K. Zhong, W. Han, M. Li, L. Wang, Z. F. Chai and W. Q. Shi, J. Electrochem. Soc., 2023, 170(3), 032501 CrossRef CAS.
  168. C. Y. Lee, J. M. Lim, H. Kim, J. Y. Park and J. H. Lee, Anal. Sci. Technol., 2022, 35(6), 256–266 Search PubMed.
  169. V. C. Bradley and J. D. Brockman, Talanta, 2022, 250, 123743 CrossRef CAS PubMed.
  170. A. Saha, S. B. Deb and M. K. Saxena, Talanta, 2022, 246, 123529 CrossRef CAS PubMed.
  171. B. C. Russell, P. E. Warwick, H. Mohamud, O. Pearson, Y. Yu, H. Thompkins, S. L. Goddard, I. W. Croudace and Z. Zacharauskas, J. Anal. At. Spectrom., 2023, 38(1), 97–110 RSC.
  172. C. Carrier, A. Habibi, L. Ferreux, L. Solier, D. Hebert, C. Augeray, M. Morin, D. Maro and L. Benedetti, J. Radioanal. Nucl. Chem., 2023, 332(6), 2003–2015 CrossRef CAS.
  173. M. Matsueda, T. Kawakami, K. Koarai, M. Terashima, K. Fujiwara, K. Iijima, M. Furukawa and Y. Takagai, Chem. Lett., 2022, 51(7), 678–682 CrossRef CAS.
  174. M. He, Q. Z. Zhao, Y. W. Bao, K. N. Li, W. H. Zhang, P. Hu, S. Y. Su, J. Gong and Q. B. You, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 527, 52–57 CrossRef CAS.
  175. M. Michel, A. Filby, S. Heinze, R. Margreiter and E. Strub, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 533, 70–75 CrossRef CAS.
  176. I. Stanciu, D. Pacesila, S. Bishop, M. Enachescu, A. Petre, M. Virgolici, A. Serban, F. Albota, I. Erhan, V. Fugaru, D. Iancu and V. Mosu, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 529, 1–6 CrossRef CAS.
  177. P. Kumar, L. Saikia, Meenakshi, D. Khandelwal, P. V. Kumar, R. Sharma, S. Ojha, S. Gargari, P. K. Mukherjee and S. Chopra, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 529, 29–37 CrossRef CAS.
  178. M. Gwozdz, R. Margreiter, T. F. Pabst, G. Hackenberg, K. Eberhardt, S. Heinze, S. Herb, M. Schiffer, A. Stolz, E. Strub, D. Mucher and A. Dewald, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 533, 76–80 CrossRef CAS.
  179. S. Pavetich, A. Wallner, H. Bottero, L. K. Fifield, M. B. Froehlich, Y. A. Huang, D. Koll, Z. Revay, Z. Slavkovska, J. H. Sterba and S. G. Tims, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 527, 45–51 CrossRef CAS.
  180. K. Sasa, Y. Ochiai, Y. Tosaki, T. Matsunaka, T. Takahashi, M. Matsumura and K. Sueki, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 532, 73–77 CrossRef CAS.
  181. M. B. Froehlich, Z. Slavkovska, D. Koll, S. Pavetich, F. Dastgiri, L. K. Fifield, M. A. C. Hotchkis, S. Merchel, S. G. Tims and A. Wallner, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 529, 18–23 CrossRef CAS.
  182. M. Veicht, I. Mihalcea, P. Gautschi, C. Vockenhuber, S. Maxeiner, J. C. David, S. H. Chen and D. Schumann, Radiochim. Acta, 2022, 110(10), 809–816 CrossRef CAS.
  183. Y. C. Fu, L. Y. Zhang, Y. T. Bi, Q. Liu and P. Cheng, J. Radioanal. Nucl. Chem., 2023, 332(4), 1253–1261 CrossRef CAS.
  184. K. Hain, S. Adler, F. Guelce, M. Martschini, J. Pitters and R. Golser, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 531, 109–114 CrossRef CAS.
  185. Z. W. Zhang, R. Qiu, Y. X. Yao, Q. Wan, G. W. Pan and J. F. Shi, Spectrosc. Spectral Anal., 2023, 43(1), 57–61 Search PubMed.
  186. G. C. Y. Chan, L. R. Martin and R. E. Russo, Appl. Spectrosc., 2023, 77, 819–834,  DOI:10.1177/00037028221112953.
  187. A. van Veelen, S. M. Hickam, N. P. Edwards, S. M. Webb, D. L. Clark, M. P. Wilkerson, A. L. Pugmire and J. R. Bargar, Anal. Chem., 2022, 94(19), 7084–7091 CrossRef CAS PubMed.
  188. V. C. Bradley, T. L. Spano, C. V. Thompson, B. W. Ticknor, D. R. Dunlap, S. C. Metzger, C. R. Hexel and B. T. Manard, Anal. Methods, 2022, 14(44), 4466–4473 RSC.
  189. J. Wimpenny, K. M. Samperton, P. Sotorrio, M. S. Wellons, S. M. Scott, D. Willingham and K. Knight, J. Anal. At. Spectrom., 2023, 38(4), 827–840 RSC.
  190. G. Wallner, H. Uguz, M. Kern, F. Jirsa and K. Hain, J. Environ. Radioact., 2022, 255, 107030 CrossRef CAS PubMed.
  191. S. Xing, C. Y. Peng, M. Christl, K. L. Shi, H. A. Synal and X. L. Hou, Anal. Chem., 2023, 95(7), 3647–3655 CrossRef CAS PubMed.
  192. E. E. Groopman, T. L. Williamson and D. S. Simons, J. Anal. At. Spectrom., 2022, 37(10), 2089–2102 RSC.
  193. N. R. Lee, T. H. Kim, C. G. Lee and J. Park, J. Radioanal. Nucl. Chem., 2022, 331(12), 5605–5612 CrossRef CAS.
  194. A. Diacre, A. L. Faure, M. Cornaton, F. Pointurier and O. Evrard, Talanta, 2023, 252, 123848 CrossRef CAS PubMed.
  195. J. Gao, M. He, Y. Shen, C. K. Xu and Y. G. Zhao, J. Mass Spectrom., 2023, 58(2), e4905 CrossRef CAS PubMed.
  196. R. Park, C. G. Lee, K. H. Chung and J. Park, J. Radioanal. Nucl. Chem., 2023, 332(7), 2833–2840 CrossRef CAS.
  197. C. L. Beck, E. S. Arnold, E. J. Warzecha, K. Koh, R. E. Stene, T. R. Pope, N. J. Henson, S. K. Vogt, M. S. Wellons and C. A. Barrett, J. Radioanal. Nucl. Chem., 2022, 331(12), 5237–5240 CrossRef CAS.
  198. K. L. LeBlanc, K. Nadeau, J. Meija, L. Yang, P. A. Babay, M. A. Bavio, C. Boome, D. Chipley, R. S. C. Leguizamon, J. Denton, D. L. Drew, M. A. Fernandez, V. Fugaru, V. D. Genetti, F. Gonzalez, J. D. Inglis, S. Jovanovic, E. Keegan, T. Kell, Y. Kimura, W. Kinman, S. Kiser, R. E. Lindvall, E. Loi, K. Mayer, J. F. Mercier, R. Millar, A. Nicholl, L. Orlovskaya, J. L. Ramella, A. Serban, M. A. Sharp, Y. Q. Shi, C. Tobi, L. Valenzuela, Z. Varga, A. Vesterlund, M. Virgolici, H. Yamazaki, E. N. Zubillaga, A. El-Jaby and Z. Mester, J. Radioanal. Nucl. Chem., 2022, 331(9), 4031–4045 CrossRef CAS.
  199. K. Nishiizumi, Nucl. Instrum. Methods Phys. Res., Sect. B, 2022, 530, 43–47 CrossRef CAS.
  200. K. J. Mathew, C. Ottenfeld and N. Butterfield, J. Radioanal. Nucl. Chem., 2022, 331(12), 4881–4889 CrossRef CAS.
  201. K. J. Mathew, C. Ottenfeld, N. Butterfield and S. Levesque, J. Radioanal. Nucl. Chem., 2022, 331(12), 4873–4879 CrossRef CAS.
  202. K. J. Mathew, C. Ottenfeld and R. Keller, J. Radioanal. Nucl. Chem., 2022, 331(12), 4897–4903 CrossRef CAS.
  203. A. Priebe and J. Michler, Materials, 2023, 16(5), 2090 CrossRef CAS PubMed.
  204. I. H. Cho and H. J. Kim, Appl. Sci. Converg. Technol., 2022, 31(6), 128–132 CrossRef.
  205. A. G. Revenko, A. L. Tsvetyansky and A. N. Eritenko, Radiat. Phys. Chem., 2022, 197, 110157 CrossRef CAS.
  206. E. Scialla, J. Brocchieri and C. Sabbarese, Appl. Radiat. Isot., 2023, 191, 110517 CrossRef CAS PubMed.
  207. T. A. Alrebdi, A. Fayyaz, H. Asghar, A. Zaman, M. Asghar, F. H. Alkallas, A. Hussain, J. Iqbal and W. Khan, Molecules, 2022, 27(12), 3754 CrossRef CAS PubMed.
  208. M. Finsgar, npj Mater. Degrad., 2023, 7(1), 1 CrossRef CAS.
  209. O. De Castro, J. N. Audinot, H. Q. Hoang, C. Coulbary, O. Bouton, R. Barrahma, A. Ost, C. Stoffels, C. G. Jiao, M. Dutka, M. Geryk and T. Wirtz, Anal. Chem., 2022, 94(30), 10754–10763 CrossRef CAS PubMed.
  210. S. Imashuku, H. Narita and K. Wagatsuma, Spectrochim. Acta, Part B, 2022, 194, 106457 CrossRef CAS.
  211. A. Priebe, A. Aribia, J. Sastre, Y. E. Romanyuk and J. Michler, Anal. Chem., 2023, 95(2), 1074–1084 CAS.
  212. T. R. Bure, O. Renault, E. Nolot, T. Lardin, C. Robert-Goumet and N. Pauly, Appl. Surf. Sci., 2023, 609, 155317 CrossRef CAS.
  213. D. M. Zheng, P. Volovitch and T. Pauporte, Small Methods, 2022, 6(11), 2200633,  DOI:10.1002/smtd.202200633.
  214. A. F. Castro-Mendez, C. A. R. Perini, J. Hidalgo, D. Ranke, J. N. Vagott, Y. An, B. R. Y. Lai, Y. Q. Luo, R. P. Li and J. P. Correa-Baena, ACS Appl. Mater. Interfaces, 2022, 14(30), 34269–34280 CrossRef CAS PubMed.
  215. K. Vegso, A. Shaji, M. Sojkova, L. P. Slusna, T. Vojtekova, J. Hrda, Y. Halahovets, M. Hulman, M. Jergel, E. Majkova, J. Wiesmann and P. Siffalovic, Rev. Sci. Instrum., 2022, 93(11), 113909 CrossRef CAS PubMed.
  216. I. Kallquist, R. Le Ruyet, H. D. Liu, R. Mogensen, M. T. Lee, K. Edstrom and A. J. Naylor, J. Mater. Chem. A, 2022, 10(37), 19466–19505 RSC.
  217. K. Kimura, J. Phys. Soc. Jpn., 2022, 91(9), 091005 CrossRef.
  218. X. M. Lin, X. T. Yang, H. N. Chen, Y. L. Deng, W. H. Chen, J. C. Dong, Y. M. Wei and J. F. Li, J. Energy Chem., 2023, 76, 146–164 CrossRef CAS.
  219. T. Y. Guselnikova, Talanta, 2023, 251, 123792 CrossRef CAS PubMed.
  220. Y. Nomura and K. Yamamoto, Adv. Energy Mater., 2023, 13(13), 2203883,  DOI:10.1002/aenm.202203883.
  221. X. Wang, H. Zhou, Z. H. Chen and X. B. Meng, Energy Storage Mater., 2022, 49, 181–208 CrossRef.
  222. T. Y. Guselnikova, A. R. Tsygankova and N. S. Medvedev, Spectrochim. Acta, Part B, 2022, 197, 106544 CrossRef CAS.
  223. N. S. Medvedev, N. A. Orlov and A. I. Saprykin, At. Spectrosc., 2022, 43(6), 430–436 CAS.
  224. M. C. M. Paradis, F. R. Doucet, S. Rousselot, A. Hernandez-Garcia, K. Rifai, O. Touag, L. C. Ozcan, N. Azami and M. Dolle, Batteries, 2022, 8(11), 231 CrossRef.
  225. J. Y. Song, X. R. Qin, Q. Lyu, W. A. Fan, Q. Wang, X. L. Wang and Z. D. Jia, Electr. Eng., 2023, 105(3), 1775–1782 CrossRef.
  226. M. Kappeler, C. Basler, A. Brandenburg, D. Carl and J. Wollenstein, Sensors, 2022, 22(22), 8816 CrossRef CAS PubMed.
  227. C. Basler, M. Kappeler and D. Carl, Sensors, 2023, 23(3), 1082 CrossRef CAS PubMed.
  228. S. M. Xu, D. Haddad and M. P. Balogh, Manuf. Lett., 2023, 35, 1–5 CrossRef.
  229. A. Franquet, V. Spampinato, S. Kayser, W. Vandervorst and P. van der Heide, Vacuum, 2022, 202, 111182 CrossRef CAS.
  230. P. P. Michalowski, M. Anayee, T. S. Mathis, S. Kozdra, A. Wojcik, K. Hantanasirisakul, I. Jozwik, A. Piatkowska, M. Mozdzonek, A. Malinowska, R. Diduszko, E. Wierzbicka and Y. Gogotsi, Nat. Nanotechnol., 2022, 17(11), 1192–1197,  DOI:10.1038/s41565-022-01214-0.
  231. J. W. Hou, W. C. Wu, L. F. Li, X. Tong, R. J. Hu, W. B. Wu, W. Z. Cai and H. L. Wang, J. Energy Storage, 2022, 55, 105369 CrossRef.
  232. X. F. Luo, J. Patra, W. T. Chuang, T. X. Nguyen, J. M. Ting, J. Li, C. W. Pao and J. K. Chang, Advanced Science, 2022, 9(21), 2201219 CrossRef CAS PubMed.
  233. S. Kim, M. W. Jeong, K. Kim, U. G. Kim, M. Kim, S. Y. Lee and Y. C. Joo, RSC Adv., 2023, 13(3), 2131–2139 RSC.
  234. O. V. Lundovskaya, A. R. Tsygankova, N. A. Orlov and F. V. Yatsunov, Inorg. Mater., 2022, 58(9), 990–998 CrossRef CAS.
  235. O. J. Marques, C. L. Chen, E. V. Timofeeva and C. U. Segre, J. Power Sources, 2023, 564, 232852 CrossRef CAS.
  236. S. Touze, V. Laperche, A. Hubau and P. Moreau, Waste Manage., 2022, 146, 66–76 CrossRef CAS PubMed.
  237. Q. Ai, Z. Y. Chen, B. Y. Zhang, F. Wang, T. S. Zhai, Y. F. Liu, Y. F. Zhu, T. Terlier, Q. Y. Fang, Y. L. Liang, L. H. Zhao, C. S. Wu, H. Guo, Z. Fan, M. Tang, Y. Yao and J. Lou, ACS Energy Lett., 2023, 8, 1107–1113 CrossRef CAS.
  238. G. Quintana-Silva, H. Sobral and J. Rangel-Cardenas, Chemosensors, 2023, 11(1), 4 CrossRef CAS.
  239. J. S. Wu, X. Y. Xia, Q. Gao, Y. M. Xin, Z. Li, J. S. Xiu and H. Q. Liu, Mater. Sci. Semicond. Process., 2023, 153, 107162 CrossRef CAS.
  240. W. Hempel and T. M. Friedlmeier, Surf. Interface Anal., 2023, 55(6–7), 437–442 CrossRef CAS.
  241. J. Labuda, J. Barek, Z. Gajdosechova, H. Goenaga-Infante, L. J. Johnston, Z. Mester and S. Shtykov, Pure Appl. Chem., 2023, 95(2), 133–163 CrossRef CAS.
  242. J. Sanmartin-Matalobos, P. Bermejo-Barrera, M. Aboal-Somoza, M. Fondo, A. M. Garcia-Deibe, J. Corredoira-Vazquez and Y. Alves-Iglesias, Nanomaterials, 2022, 12(14), 2501 CrossRef CAS PubMed.
  243. A. Fuentes-Cervantes, J. R. Allica, F. C. Celis, J. M. Costa-Fernandez and J. R. Encinar, Nanomaterials, 2023, 13(6), 1132 CrossRef CAS PubMed.
  244. F. Laborda, I. Abad-Alvaro, M. S. Jimenez and E. Bolea, Spectrochim. Acta, Part B, 2023, 199, 106570 CrossRef CAS.
  245. X. W. Tian, H. W. Jiang, L. G. Hu, M. Wang, W. B. Cui, J. B. Shi, G. L. Liu, Y. G. Yin, Y. Cai and G. B. Jiang, TrAC, Trends Anal. Chem., 2022, 157, 116746 CrossRef CAS.
  246. S. Zhu, Y. L. Wang and C. Y. Chen, Curr. Anal. Chem., 2022, 18(6), 723–738 CrossRef CAS.
  247. A. C. Gimenez-Ingalaturre, K. Ben-Jeddou, J. Perez-Arantegui, M. S. Jimenez, E. Bolea and F. Laborda, Anal. Bioanal. Chem., 2023, 415(11), 2101–2112 CrossRef CAS PubMed.
  248. T. E. Lockwood, R. G. de Vega and D. Clases, J. Anal. At. Spectrom., 2021, 36(11), 2536–2544 RSC.
  249. J. H. Liu, X. Wei, C. X. Wu, L. N. Zheng, M. Wang, M. L. Chen and J. H. Wang, Anal. Chim. Acta, 2023, 1254, 341114 CrossRef CAS PubMed.
  250. M. Aramendia, J. C. Garcia-Mesa, E. V. Alonso, R. Garde, A. Bazo, J. Resano and M. Resano, Anal. Chim. Acta, 2022, 1205, 339738 CrossRef CAS PubMed.
  251. D. Ojeda, E. Bolea, J. Perez-Arantegui and F. Laborda, J. Anal. At. Spectrom., 2022, 37(7), 1501–1511 RSC.
  252. M. Tharaud, L. Schlatt, P. Shaw and M. F. Benedetti, J. Anal. At. Spectrom., 2022, 37(10), 2042–2052 RSC.
  253. S. Harycki and A. Gundlach-Graham, J. Anal. At. Spectrom., 2023, 38(1), 111–120 RSC.
  254. X. W. Tian, H. W. Jiang, M. Wang, W. B. Cui, Y. Y. Guo, L. N. Zheng, L. G. Hu, G. B. Qu, Y. G. Yin, Y. Cai and G. B. Jiang, Anal. Chim. Acta, 2023, 1240, 340756 CrossRef CAS PubMed.
  255. K. H. Chun, J. T. S. Lum and K. S. Y. Leung, Anal. Chim. Acta, 2022, 1226, 340258 CrossRef CAS PubMed.
  256. Y. Gao, D. X. Li, R. Y. Zhang, Y. T. Guo, L. Chen, G. L. Tian, R. X. Liu and G. L. Ge, Spectrochim. Acta, Part B, 2022, 198, 106550 CrossRef CAS.
  257. D. Torregrosa, C. Gomez-Pertusa, G. Grindlay, L. Gras and J. Mora, J. Anal. At. Spectrom., 2023, 38(2), 403–413 RSC.
  258. C. D. Koolen, L. Torrent, A. Agarwal, O. Meili-Borovinskaya, N. Gasilova, M. Li, W. Luo and A. Zuttel, ACS Nano, 2022, 16(8), 11968–11978 CrossRef CAS PubMed.
  259. J. Kocic, D. N. Dirin, R. Kagi, M. V. Kovalenko, D. Gunther and B. Hattendorf, J. Anal. At. Spectrom., 2022, 37(8), 1738–1750 RSC.
  260. G. Kajner, A. Keri, A. Belteki, S. Valkai, A. Der, Z. Geretovszky and G. Galbacs, Lab Chip, 2022, 22(14), 2766–2776 RSC.
  261. Z. C. Zhou, A. Al Hejami, M. J. Burgener, J. Burgener and D. Beauchemin, J. Anal. At. Spectrom., 2022, 37(7), 1450–1454 RSC.
  262. M. Stiborek, L. Jindrichova, S. Meliorisova, A. Bednarik, V. Prysiazhnyi, J. Kroupa, P. Houska, B. Adamova, J. Navratilova, V. Kanicky and J. Preisler, Anal. Chem., 2022, 94(51), 18114–18120 CrossRef CAS PubMed.
  263. S. Fernandez-Trujillo, A. Sanchez-Cachero, F. J. G. Bernardo, N. Rodriguez-Farinas, M. Jimenez-Moreno and R. C. R. Martin-Doimeadios, Anal. Bioanal. Chem., 2023, 415(11), 2113–2120 CrossRef CAS PubMed.
  264. M. Bierwirth, V. Olszok, A. Wollmann and A. P. Weber, J. Aerosol Sci., 2022, 163, 105983 CrossRef CAS.
  265. Y. C. Hsieh, Y. P. Lin, T. C. Hsiao and W. C. Hou, Sci. Total Environ., 2022, 838, 156444 CrossRef CAS PubMed.
  266. D. Cassano, A. Bogni, R. La Spina, D. Gilliland and J. Ponti, Nanomaterials, 2023, 13(3), 594 CrossRef CAS PubMed.
  267. N. D. Donahue, V. Sheth, A. N. Frickenstein, A. Holden, S. Kanapilly, C. Stephan and S. Wilhelm, Nano Lett., 2022, 22(10), 4192–4199 CrossRef CAS PubMed.
  268. M. J. Eller, J. M. Sandoval, S. V. Verkhoturov and E. A. Schweikert, Anal. Chem., 2022, 94(22), 7868–7876 CrossRef CAS PubMed.
  269. K. Finch, A. Hernandez and G. Gamez, Anal. Chem., 2023, 95, 2269–2277,  DOI:10.1021/acs.analchem.2c03940.
  270. K. Flores, L. N. Rand, C. Valdes, A. Castillo, J. M. Cantu, J. G. Parsons, P. Westerhoff and J. L. Gardea-Torresdey, Environ. Sci. Technol., 2022, 56(19), 13719–13727 CrossRef CAS PubMed.
  271. Y. U. Hachenberger, D. Rosenkranz, C. Kromer, B. C. Krause, N. Dreiack, F. L. Kriegel, E. Koz'menko, H. Jungnickel, J. Tentschert, F. S. Bierkandt, P. Laux, U. Panne and A. Luch, Nanomaterials, 2023, 13(5), 922 CrossRef CAS PubMed.
  272. K. Ilmiah, J. Sumranjit, T. Wutikhun and A. Siripinyanond, ACS Appl. Nano Mater., 2023, 6, 1250–1260,  DOI:10.1021/acsanm.2c04825.
  273. L. I. Kaynarova, D. L. Georgieva and V. M. Stefanova, J. Anal. At. Spectrom., 2022, 37(7), 1484–1500 RSC.
  274. C. K. Y. Law, E. Bolea-Fernandez, T. Liu, L. Bonin, E. Wallaert, K. Verbeken, B. De Gusseme, F. Vanhaecke and N. Boon, Microb. Biotechnol., 2023, 16(5), 901–914 CrossRef CAS PubMed.
  275. M. Mateo, J. Becerra, A. P. Zaderenko, P. Ortiz and G. Nicolas, J. Cult. Herit., 2022, 58, 237–244 CrossRef.
  276. M. Momcilovic, J. Petrovic, M. Nemoda, J. Ciganovic, N. Krstulovic, M. Ognjanovic and S. Zivkovic, Appl. Phys. B: Lasers Opt., 2023, 129(4), 62 CrossRef CAS.
  277. B. Moreira-Alvarez, A. L. Larraga-Urdaz, A. Fuentes-Cervantes, M. L. Fernandez-Sanchez, J. M. Costa-Fernandez and J. R. Encinar, Talanta, 2023, 256, 124309 CrossRef CAS PubMed.
  278. P. E. Peyneau, J. Anal. At. Spectrom., 2022, 37(12), 2683–2690 RSC.
  279. F. Pinakidou, K. Simeonidis, E. Myrovali, M. Brzhezinskaya, E. C. Paloura, M. Angelakeris and M. Katsikini, J. Phys. Chem. C, 2022, 126(24), 10101–10109 CrossRef CAS.
  280. P. Purohit, F. J. Fortes and J. Laserna, Nano Res., 2023, 16(5), 7470–7480 CrossRef CAS.
  281. D. Skroblin, A. F. Herrero, T. Siefke, K. Nikolaev, A. Andrle, P. Honicke, Y. Kayser, M. Krumrey, C. Gollwitzer and V. Soltwisch, Nanoscale, 2022, 14(41), 15475–15483 RSC.
  282. M. Snellman, P. Samuelsson, A. Eriksson, Z. S. Li and K. Deppert, J. Aerosol Sci., 2022, 165, 106041 CrossRef CAS.
  283. S. Staeck, A. Andrle, P. Honicke, J. Baumann, D. Grotzsch, J. Weser, G. Goetzke, A. Jonas, Y. Kayser, F. Forste, I. Mantouvalou, J. Viefhaus, V. Soltwisch, H. Stiel, B. Beckhoff and B. Kanngiesser, Nanomaterials, 2022, 12(21), 3766 CrossRef CAS PubMed.
  284. Z. A. Temerdashev, O. A. Galitskaya and M. A. Bolshov, Molecules, 2022, 27(22), 7748 CrossRef CAS PubMed.
  285. A. Wahlisch, R. Unterumsberger, P. Honicke, J. Lubeck, Y. Kayser, J. Weser, G. L. Dai, K. Hahm, T. Weimann, C. Seim, S. Rehbein and B. Beckhoff, Small, 2023, 19(9), 2204943,  DOI:10.1002/smll.202204943.
  286. D. Charlton, C. Costa, G. F. Trindade, S. Hinder, J. F. Watts and M. J. Bailey, Sci. Justice, 2023, 63(1), 9–18 CrossRef PubMed.
  287. Z. Siketic, I. B. Radovic, M. Barac, M. Brajkovic and M. P. Hadzija, Anal. Chem., 2023, 95(5), 3069–3074 CrossRef CAS PubMed.
  288. T. Trejos, Wiley Interdiscip. Rev.: Forensic Sci., 2023, 5(2), e1476,  DOI:10.1002/wfs2.1476.
  289. O. Kaspi, O. Israelsohn-Azulay, Y. Zidon, H. Rosengarten, M. Krmpotic, S. Gouasmia, I. B. Radovic, P. Jalkanen, A. Liski, K. Mizohata, J. Raisanen, O. Girshevitz and H. Senderowitz, Forensic Sci. Int., 2022, 333, 111216 CrossRef CAS PubMed.
  290. O. Kaspi, O. Israelsohn-Azulay, Z. Yigal, H. Rosengarten, M. Krmpotic, S. Gouasmia, I. B. Radovic, P. Jalkanen, A. Liski, K. Mizohata, J. Raisanen, Z. Kasztovszky, I. Harsanyi, R. Acharya, P. K. Pujari, M. Mihaly, M. Braun, N. Shabi, O. Girshevitz and H. Senderowitz, J. Chem. Inf. Model., 2023, 63(1), 87–100 CrossRef CAS PubMed.
  291. R. Corzo, T. Ernst, J. Insana, C. Martinez-Lopez, J. Webb, E. Haase, P. Weis and T. Trejos, Forensic Chem., 2022, 31, 100447 CrossRef CAS.
  292. O. Ovide, R. Corzo and T. Trejos, Forensic Sci. Int., 2023, 343, 111550 CrossRef CAS PubMed.
  293. A. Ishimi, T. Nakanishi, Y. Seto and Y. Nishiwaki, J. Forensic Sci., 2022, 67(5), 1825–1835 CrossRef CAS PubMed.
  294. R. Ahmed, M. J. Altamimi and M. Hachem, Molecules, 2022, 27(19), 6602 CrossRef CAS PubMed.
  295. C. Douvris, E. Bentil, I. Ayensu, C. O. Akoto, I. K. Amponsah, J. Adu and D. Bussan, Toxics, 2022, 10(10), 567 CrossRef CAS PubMed.
  296. T. Fischer, M. Marchetti-Deschmann, A. C. Assis, M. L. Elad, M. Algarra, M. Barac, I. B. Radovic, F. Cicconi, B. Claes, N. Frascione, S. George, A. Guedes, C. Heaton, R. Heeren, V. Lazic, J. L. Lerma, M. Garcia, M. Nosko, J. O'Hara, I. Oshina, A. Palucci, A. Pawlaczyk, A. Z. Pospiskova, M. de Puit, K. Radodic, M. Repele, M. Ristova, F. S. Romolo, I. Safarik, Z. Siketic, J. Spigulis, M. I. Szynkowska-Jozwik, A. Tsiatsiuyeu, J. Vella, L. Dawson, S. Rodiger and S. Francese, Sci. Justice, 2022, 62(4), 433–447 CrossRef PubMed.
  297. E. Pigorsch, B. Kiessler and M. Huls, J. Forensic Sci., 2022, 67(4), 1505–1512 CrossRef CAS PubMed.
  298. M. A. A. Al-Ameri, B. Ciylan and B. Mahmood, Electronics, 2022, 11(23), 4036 CrossRef.
  299. V. Merk, W. Werncke and L. Pfeifer, Analyst, 2022, 147(23), 5470–5476 RSC.
  300. C. Manfriani, M. Zuena, L. Ruggiero, C. Lucarelli, S. Ridolfi, L. Tortora and A. Sodo, Dyes Pigm., 2022, 207, 110755 CrossRef CAS.
  301. R. Brunjes, J. Schuurman, F. von der Kammer and T. Hofmann, Forensic Sci. Int., 2022, 332, 111202 CrossRef PubMed.
  302. S. E. Szakas, K. Menking-Hoggatt, T. Trejos and A. Gundlach-Graham, Appl. Spectrosc., 2023, 77, 873–884,  DOI:10.1177/00037028221142624.
  303. K. Menking-Hoggatt, C. Ott, C. Vander Pyl, K. Dalzell, J. Curran, L. Arroyo and T. Trejos, Forensic Chem., 2022, 29, 100429 CrossRef CAS.
  304. H. L. Guo, H. C. Mei, P. Wang, C. Hu and J. Zhu, J. Forensic Sci., 2023, 68(2), 638–646 CrossRef CAS PubMed.
  305. M. Sharma, B. Sharma, A. K. Gupta and D. Pandey, Multimed. Tools Appl., 2023, 82(5), 6849–6865 CrossRef.
  306. P. Ippoliti, J. Werlich, C. Fuglsby, C. Yarnes, C. P. Saunders and J. Dettman, J. Forensic Sci., 2023, 68(2), 407–415 CrossRef CAS PubMed.
  307. S. Harikrishnan, S. D. George, S. Chidangil and V. K. Unnikrishnnan, Appl. Spectrosc. Rev., 2023 DOI:10.1080/05704928.2023.2189946.
  308. L. Ghervase and I. M. Cortea, Chemosensors, 2023, 11(2), 100 CrossRef CAS.
  309. V. Detalle and X. S. Bai, Spectrochim. Acta, Part B, 2022, 191, 106407 CrossRef CAS.
  310. D. M. Chen, L. L. Ma and X. M. Zhang, Spectrosc. Spectral Anal., 2023, 43(2), 334–341 Search PubMed.
  311. L. Sottili, L. Giuntini, A. Mazzinghi, M. Massi, L. Carraresi, L. Castelli, C. Czelusniak, F. Giambi, P. A. Mando, M. Manetti, C. Ruberto, L. Guidorzi, A. Re, A. Lo Giudice, R. Torres, F. Arneodo, S. M. E. Mangani, S. Calusi and F. Taccetti, Appl. Sci., 2022, 12(13), 6585 CrossRef CAS.
  312. R. McMillan, N. Waber, M. Ritchie and E. Frahm, J. Archaeol. Sci., 2022, 144, 105626 CrossRef.
  313. T. Trojek and D. Trojkova, Materials, 2023, 16(3), 1133 CrossRef CAS PubMed.
  314. V. Palleschi, L. Marras and M. A. Turchetti, Heritage, 2022, 5(4), 4089–4099 CrossRef.
  315. N. Herreyre, A. Cormier, S. Hermelin, C. Oberlin, A. Schmitt, V. Thirion-Merle, A. Borlenghi, D. Prigent, C. Coquide, A. Valois, C. Dujardin, P. Dugourd, L. Duponchel, C. Comby-Zerbino and V. Motto-Ros, J. Anal. At. Spectrom., 2023, 38(3), 730–741 RSC.
  316. E. Catelli, Z. L. Li, G. Sciutto, P. Oliveri, S. Prati, M. Occhipinti, A. Tocchio, R. Alberti, T. Frizzi, C. Malegori and R. Mazzeo, Anal. Chim. Acta, 2023, 1239, 340710 CrossRef CAS PubMed.
  317. P. Silveira and T. Falcade, J. Cult. Herit., 2022, 57, 243–255 CrossRef.
  318. J. Brocchieri, E. Scialla, A. Manzone, G. O. Graziano and C. Sabbarese, J. Archaeol. Sci.: Rep., 2022, 42, 103382 Search PubMed.
  319. J. Brocchieri, E. Scialla, A. Manzone, G. O. Graziano, A. D'Onofrio and C. Sabbarese, J. Cult. Herit., 2022, 57, 213–225 CrossRef.
  320. D. A. Chen, Y. D. Yang, T. Y. Wang, X. T. Wang and W. G. Luo, Heritage Sci., 2022, 10(1), 92 CrossRef CAS.
  321. M. Porcaro, A. Depalmas, S. Lins, C. Bulla, M. Pischedda and A. Brunetti, Materials, 2022, 15(11), 3879 CrossRef CAS PubMed.
  322. R. Schwab, M. Brauns, W. Fasnacht, S. W. Katzev, N. Lockhoff and H. W. Swiny, J. Archaeol. Sci.: Rep., 2022, 42, 103365 Search PubMed.
  323. J. Orsilli, A. Migliori, R. Padilla-Alvarez, M. Martini and A. Galli, J. Anal. At. Spectrom., 2023, 38(1), 174–185 RSC.
  324. R. A. Rezk, N. A. A. Ghany and A. M. Mostafa, Coatings, 2022, 12(10), 1548 CrossRef CAS.
  325. T. Calligaro, A. Banas, K. Banas, I. B. Radovic, M. Brajkovic, M. Chiari, A. M. Forss, I. Hajdas, M. Krmpotic, A. Mazzinghi, E. Menart, K. Mizohata, M. Oinonen, L. Pichon, J. Raisanen, Z. Siketi, Z. Smit and A. Simon, Forensic Sci. Int., 2022, 336, 111327 CrossRef CAS PubMed.
  326. M. Vermeulen, A. McGeachy, B. J. Xu, H. Chopp, A. Katsaggelos, R. Meyers, M. Alfeld and M. Walton, J. Anal. At. Spectrom., 2022, 37(10), 2130–2143 RSC.
  327. B. J. Xu, Y. A. Wu, P. X. Hao, M. Vermeulen, A. McGeachy, K. Smith, K. Eremin, G. Rayner, G. Verri, F. Willomitzer, M. Alfeld, J. Tumblin, A. Katsaggelos and M. Walton, J. Anal. At. Spectrom., 2022, 37(12), 2672–2682 RSC.
  328. A. Bombini, L. Anderlini, L. Dell'Agnello, F. Giaocmini, C. Ruberto and F. Taccetti, presented at 21st International Conference on Image Analysis and Processing (ICIAP), Lecce, Italy, 2022, vol. 13231, pp. 685–700,  DOI:10.1007/978-3-031-06427-2_57.
  329. S. C. R. Ferreira, M. Franchi, A. Pais and J. S. S. de Melo, Dyes Pigm., 2022, 205, 110519 CrossRef CAS.
  330. T. Gerodimos, A. Asvestas, G. P. Mastrotheodoros, G. Chantas, I. Liougos, A. Likas and D. F. Anagnostopoulos, J. Imaging, 2022, 8(5), 147 CrossRef PubMed.
  331. Y. M. Zhang, D. X. Sun, Y. P. Yin, Z. R. Yu, B. M. Su, C. Z. Dong and M. G. Su, Plasma Sci. Technol., 2022, 24(8), 084003 CrossRef.
  332. M. Raphael, B. D. Lucile, P. Laurent, M. Brice, G. Didier and C. Thomas, Eur. Phys. J. Plus, 2023, 138(1), 16 CrossRef.
  333. D. X. Sun, X. R. Li, Y. P. Yin, Y. M. Zhang, W. W. Han, Y. R. Wang, M. G. Su, C. Z. Dong, Z. R. Yu and B. M. Su, J. Cult. Herit., 2022, 55, 399–408 CrossRef.
  334. D. X. Sun, Y. M. Zhang, Y. P. Yin, Z. Zhang, H. L. Qian, Y. R. Wang, Z. R. Yu, B. M. Su, C. Z. Dong and M. G. Su, Chemosensors, 2022, 10(10), 389 CrossRef CAS.
  335. M. L. Amadori, G. Poldi, G. Germinario, J. Arduini and V. Mengacci, Appl. Sci., 2023, 13(6), 3581 CrossRef CAS.
  336. Y. Qi, M. Z. Qiu, H. Z. Jing, Z. Q. Wang, C. L. Yu, J. F. Zhu, F. Wang and T. Wang, Ceram. Int., 2022, 48(23), 34516–34532 CrossRef CAS.
  337. E. Beauvoit, N. Cantin, Q. Lemasson, R. Chapoulie and A. Ben Amara, Materials, 2022, 15(15), 5311 CrossRef CAS PubMed.
  338. E. C. Kracht, L. C. Bloch and W. F. Keegan, J. Archaeol. Sci.: Rep., 2022, 43, 103469 Search PubMed.
  339. C. S. Araujo, C. R. Appoloni, R. A. Ikeoka and L. C. P. Symanski, Appl. Radiat. Isot., 2023, 191, 110560 CrossRef CAS PubMed.
  340. J. Kim, J. W. Park, H. Kim, Y. Oh, J. Park, M. Conte and J. Kim, J. Archaeol. Sci.: Rep., 2023, 47, 103788 Search PubMed.
  341. A. Anglisano, L. Casas, I. Queralt and R. Di Febo, Sustainability, 2022, 14(18), 11214 CrossRef.
  342. R. C. Chaves, J. P. Veiga and A. M. Soares, Heritage, 2022, 5(3), 2422–2443 CrossRef.
  343. M. Vannoorenberghe, D. Teetaert, E. Goemaere, T. Van Acker, J. Belza, E. Meylemans, F. Vanhaecke and P. Crombe, J. Archaeol. Sci.: Rep., 2022, 42, 103413 Search PubMed.
  344. A. S. Maltsev, N. N. Umarova, G. V. Pashkova, M. M. Mukhamedova, D. L. Shergin, V. V. Panchuk, D. O. Kirsanov and E. I. Demonterova, Molecules, 2023, 28(3), 1099 CrossRef CAS PubMed.
  345. M. G. Li, F. Q. Ruan, R. R. Li, J. J. Zhou, T. L. Zhang, H. S. Tang and H. Li, Microchem. J., 2022, 182, 107928 CrossRef CAS.
  346. R. A. Ikeoka, C. R. Appoloni, R. B. Scorzelli, E. Dos Santos, M. D. Rizzutto and A. M. Bandeira, Minerals, 2022, 12(10), 1302 CrossRef CAS.
  347. L. Medeghini, M. Botticelli, A. C. Cadena-Irizar, B. Lepri, A. F. Ferrandes, M. Costa and P. Barrulas, Microchem. J., 2022, 179, 107526 CrossRef CAS.
  348. N. Schibille, V. A. Ruiz, J. D. Ares and S. G. Lloret, ChemPlusChem, 2022, 87(9), e202200147 CrossRef CAS PubMed.
  349. F. Armetta, M. L. Saladino, M. C. Martinelli, R. Vilardo, G. Anastasio, S. Trusso, V. M. Nardo, D. Giuffrida and R. C. Ponterio, RSC Adv., 2023, 13(7), 4495–4503 RSC.
  350. J. W. Lankton, C. Pulak and B. Gratuze, J. Archaeol. Sci.: Rep., 2022, 45, 103596 Search PubMed.
  351. O. Yatsuk, M. Ferretti, A. Gorghinian, G. Fiocco, M. Malagodi, A. Agostino and M. Gulmini, Molecules, 2022, 27(18), 6068 CrossRef CAS PubMed.
  352. R. Zanini, M. Roman, E. Cattaruzza and A. Traviglia, J. Anal. At. Spectrom., 2023, 38(4), 917–926 RSC.

This journal is © The Royal Society of Chemistry 2023