A hybrid model combining wavelet transform and recursive feature elimination for running state evaluation of heat-resistant steel using laser-induced breakdown spectroscopy
Heat-resistant steel is widely used in various industries, and the running state is of great importance for equipment function and safety. In this work, laser-induced breakdown spectroscopy (LIBS) is applied to evaluate the running state of steel using indicators of micro and macro properties. The hybrid model based on wavelet threshold denoising (WTD) and K-fold-support vector machine-recursive feature elimination (K-SVM-RFE) is proposed to estimate the different indictors of various service conditions of steel. Fourteen T91 specimens, including 4 industrial specimens obtained from different service conditions in the power plant boiler, were used as the analytes. Firstly, the noise signal of the LIBS spectra of each specimen was analyzed and removed with WTD. Secondly, an improved approach K-SVM-RFE was applied to select the optimal feature subset and build the classification models of aging grade and hardness grade. The influence of denoising pretreatment on model performance was compared and discussed. Finally, the assessment matrix, established using the indicators from the aging grade and hardness grade, was used to evaluate the running state of steel. The results show that the test assessment matrix obtained with the hybrid model based on WTD and K-SVM-RFE is consistent with the reference matrix on the running state of steel.