Classification of Polluted Silicone Rubber Insulators by Using LIBS Assisted Machine Learning Techniques

Abstract

Silicone rubber (SR) samples are coated with various types of artificially prepared pollutants, in order to identify and distinguish them by employing laser induced breakdown spectroscopy (LIBS). LIBS analysis is successful in identifying the elemental composition of the various types of pollutants. The presence of copper sulphate as well as carbon-based compounds such as fly ash, coal and calcium-based compounds such as cement and calcium phosphate (fertilizer) have been identified by the increment in the normalized intensity ratio of the copper, carbon and calcium peaks respectively. LIBS spectral data has been used in conjunction with several machine learning (ML) algorithms such as linear discriminant analysis, decision tree, K-nearest neighbors and various gradient boosting techniques to classify seven different types of contaminated SR samples. When compared to the other ML approaches utilized in the present study, classification using the Light gradient boosting technique has reflected better classification accuracy of 97.43% with a reasonable computation time of 5.1 s.