Temporal Profiles Of The Atomic Emissions In High Repetition Rate Laser-ablation Spark-induced Breakdown Spectroscopy And A Related Machine Learning Approach For Alloy Classification

High-repetition-rate laser-ablation spark-induced breakdown spectroscopy(HRR LA-SIBS)is a new spectroscopic technique which uses high repetition frequency laser as excitation source and combines with spark discharge to analyze material elements with high sensitivity.In order to achieve quantitative analysis,it is usually required that the plasma is in local thermodynamic equilibrium within the time window of spectral observation,which requires a gated detector.However,the existing detectors cannot realize the gating detection under the condition of high repetition frequency.In order to study the feasibility of quantitative analysis of elements using spectra recorded by non-gated detectors,the time domain characteristics of atomic radiation in HRR LA-SIBS technique are firstly studied in this paper.The time domain signals of the atomic radiation lines of silver,copper and zinc atoms in925 silver alloy and lead-brass standard samples were observed and analyzed experimentally,and the effect of capacitance on the time domain signals of the atomic emission was studied.The experimental results show that in HRR LA-SIBS,due to the effect of spark discharge,the time domain diagrams of different transition atomic radiation of different elements in plasma are very similar in linear line,and the main influencing factor is the time characteristics of spark discharge.After the termination of spark discharge,the free decay life of atomic radiation depends on the energy of the upper level of the transition,and the higher the energy of the upper level,the shorter the free decay life of atomic radiation.However,because the atomic radiation intensity after the stop of spark discharge is relatively weak compared with the atomic radiation intensity during the process of spark discharge,it does not have a significant effect on the time integral intensity of the atomic radiation during the whole cycle.Since all the atomic radiation in HRR LA-SIBS presents a similar time-domain distribution,the time integral signal strength in the whole cycle will be approximately proportional to the time integral signal strength obtained in the gated signal recording mode.In the spectral analysis,it can be approximately considered that the two are equivalent.Therefore,in HRR LA-SIBS,ordinary CCD can be used to replace the expensive ICCD to record spectral data in the non-gated signal recording mode and carry out element quantitative analysis.In order to realize the rapid classification of aluminum alloys,the HRR LA-SIBS technique was combined with machine learning to collect the spectral data of 10 kinds of aluminum alloy standard samples with different roundness,the spectral data are preprocessed by threshold cleaning and PCA algorithm,which can reduce the dimension of the data and preserve the characteristic information of the spectral data.K-nearest neighbor(KNN),random forest(RF)and support vector machine(SVM)were used to classify the processed spectral data,and the accuracy of three different machine learning models was compared.The results show that the combination of HRR LA-SIBS and machine learning is feasible for aluminum alloy classification,which has the advantages of fast reading and spectral data acquisition,fast classification speed and high classification accuracy.By comparing the classification results of three different machine learning models,it is concluded that the support vector machine(SVM)model is the best choice for aluminum alloy classification.HRR LA-SIBS combined with machine learning is a novel object classification technology,which combines the advantages of spectral analysis and machine learning,and is expected to be widely used in the future artificial intelligence era.