Application Of Laser-induced Breakdown Spectroscopy Combined With Machine Learning Algorithm In Classification

Laser-induced breakdown spectroscopy(LIBS)is a popular technique for the analysis of composition and content.In recent years,it has been widely used in classification.The wavelength of the spectrum line can determine the types of elements contained in the material,the intensity of the spectrum line can clarify the different element content,so as to achieve the purpose of material identification and classification.LIBS has the advantages of real-time and rapid detection.The key application in classification is the accurate recognition and extraction of LIBS spectral information.Fortunately,machine learning has the unique advantages of working with large amounts of data and mining useful information,can filter and extract effective information from LIBS spectra.This paper mainly aims to explore the application of LIBS combined with machine learning in classification,and conducts research on the classification and recycling of aluminum alloys,which is currently hot in research.At the same time,this paper also studies the origin identification of daily-use ceramics,which is less researched.It makes sense to choose to study the sorting and recycling of aluminum alloys.With the development of industry,there are large number of aluminum alloy products accumulated,such as industrial scraps and discarded scraps.The recycling,identification and reasonable classification of aluminum alloys can not only realize the recycling of resources,but also can help to protect the environment,and has economic and social benefits.Determining the source of daily-use ceramics and distinguishing daily-use ceramics from different origins is helpful for testing and supervising product quality.The main research contents and conclusions of this article are as follows:(1)LIBS combined with generalized regression neural network(GRNN)was employed for classification and identification of different types of daily-use ceramics.Firstly,four different categories of daily-use ceramics were excited by LIBS technology.The experiment mainly consisted of two aspects: firstly,discussed the influence of spectral data extraction on model efficiency;secondly,optimized the performance of the model.To discuss the influence of spectral data extraction on model efficiency,two spectral data inputs were proposed: the whole spectral range and several primary characteristic spectral lines of the main elements.The GRNN classification model was constructed,the analyses demonstrated that the screening of valid data for LIBS spectra can greatly increase the modeling efficiency and shorten the time by about 45 times.Before modeling,the spectral data were normalized.And then the abnormal spectral data was eliminated using mahalanobis distance,so as to reduce the adverse impact of poor spectra on the GRNN classification model.Next,principal component analysis(PCA)was used to further simplify the LIBS spectral data.Under the condition of ensuring the efficiency of the model,the single prediction accuracy rate of the test sets can reach 100%,and the average prediction accuracy rate is 99.74%.On all these counts,the correct classification of daily-use ceramics by their LIBS spectra combined with GRNN can be achieved.(2)LIBS combined with particle swarm optimization(PSO)and extreme learning machine(ELM)was employed for classification and identification of 5 different types of aluminum alloys.After collecting the LIBS spectra of samples,PSO was used to extract 7 characteristic spectrum segments from LIBS spectra,and then,segments were combined together as the input of ELM.The results show that the feature extraction of the spectra by PSO can improve the average prediction accuracy of ELM to 99.65%.The experimental results show that LIBS combined with PSO-ELM can realize the identification and classification of different brands of aluminum alloys.