Improving quantitative analysis of cement elements in laser-induced breakdown spectroscopy through combining matrix matching with regression

Abstract

Laser induced breakdown spectroscopy (LIBS) is a promising cement analysis technique due to its advantages of non-hazardous radiation and rapid elemental characterization. However, the presence of matrix effects in LIBS cement analysis results in unsatisfactory quantitative performance. In this study, we propose a matrix-matching quantitative analysis method to reduce the influence of matrix effects. This method combines hierarchical clustering with regression models, where samples with similar matrix properties are grouped together using hierarchical clustering, and regression models are constructed within each cluster using commonly used algorithms in LIBS quantitative analysis tasks (PLSR, SVR, and K-ELM). To evaluate the performance of the proposed method, experiments were conducted on a dataset of 58 LIBS cement samples involving measurements of six major elements (Ca, Si, Al, Fe, Mg, and K). The results demonstrated that the proposed combined models (HC-PLSR, HC-SVR, and HC-K-ELM) outperformed the baseline models in terms of the root mean square error of prediction (RMSE_P), mean absolute percentage error of prediction (MAPE_P), and relative standard deviation of prediction (RSD_P). The proposed matrix-matching method provided a feasible solution to alleviate the matrix effects in LIBS cement analysis and effectively improves the quantitative performance of LIBS cement analysis.