Research on microstructure Classification of y-TiAl based on laser-induced breakdown spectroscopy and deep learning model

Online diagnosis and testing of y-TiAl components in aircraft engines during processing and man-ufacturing is an important part for aircraft engine manufacturing and intelligent testing. Due to the fact that laser-induced breakdown spectroscopy (LIBS) technology can only be used to detect the elemental composition of materials,there is a lack of direct judgment on the microstructure of materials. In this stud-y,the identification of y-TiAl microstructure was realized based on the combination of LIBS with deep learning algorithms. In experiments, y-TiAl samples were subjected to six diffcrent heat treatments to ob-tain different microstructures under electron microscope. Subsequently, LIBS experiments were conducted on y-TiAl with different microstructures,and the obtained spectra were denoised through baseline corrcc-tion and wavelet transform. In order to improve the simplicity and interpretability of the data, principal component analysis(PCA) was used to take the first 32 principal components as the dimensionality reduced data,which were used as the inputs for Classification by three deep learning modcls, i. e., BP neural net-work(BP),convolutional neural network(CNN),and long short-term memory neural network(LSTM). A-mong them,the LSTM model had the best Performance with accuracy of 96. 04%,while the BP and CNN modcls also had excellent results,with accuracy of 95. 57% and 93. 35%, respectively. Meanwhile, the training of three models was completed within 30 s. Therefore,the combination of LIBS and deep learning models could achieve the accurate Classification of y-TiAl with different microstructures, which provided new means and ideas for intelligent detection in industrial production in the future.