Estimation of the mechanical properties of steel via LIBS combined with canonical correlation analysis (CCA) and support vector regression (SVR)
Degradation of steel is a significant issue in the field of material aging, with the mechanical properties of steel degrading during service, affecting the safety of equipment. In this work, laser-induced breakdown spectroscopy (LIBS) was applied to investigate the mechanical properties of steel. T91 steel specimens with different degrees of microstructure aging were selected as model samples. Surface hardness was chosen as the key indicator of mechanical properties. The correlation between emission line intensity and hardness was analyzed in order to establish a calibration model of hardness. Multivariate analysis methods (principal component analysis [PCA] and canonical correlation analysis [CCA]) were introduced to identify the important variables from the whole spectrum. Then, two regression algorithms (partial least-squares regression [PLSR] and support vector regression [SVR]) were used to establish the calibration models with the selected variables. The results showed that it is feasible to couple CCA and SVR to estimate hardness, which can effectively identify the correlated variables and establish the correlation between emission lines and hardness, with maximum values for mean relative error (MRE), relative standard deviation (RSD) and root mean square error of prediction (RMSEP) of 2.47%, 2.94% and 6.14, respectively. In addition, the influence of collinearity variables on the established model was investigated in order to show that there is little multicollinearity issue in the calibration models constructed with CCA and SVR according to the values of RMSEP, RSD and MRE. This demonstrates that LIBS technology coupled with chemometrics (CCA and SVR) is an appropriate method to estimate the mechanical properties of steel.