Microstructure classification of γ-TiAl alloy using an MLP deep learning analysis model of LIBS spectra

This study proposes a new strategy to accurately classify γ-TiAl samples with different microstructures using laser-induced breakdown spectroscopy (LIBS) combined with deep learning techniques. We first observed the microstructure of six groups of γ-TiAl treated with different solid solution temperatures and found that the percentage of lamellae increased with increasing temperature, while the percentage of γ phase substantially decreased. Next, the elemental characteristic spectral lines were collected by a coaxial acquisition device. Then we performed baseline correction and normalization on the LIBS spectra to eliminate the background signals. Principal Component Analysis (PCA) was then used to reduce the dimensionality to simplify the data structure. Finally, the processed data were fed into three different deep learning models, namely, Multilayer Perceptron (MLP), Long Short-Term Memory Network (LSTM), and Convolutional Neural Network (CNN), for training and classification. The classification accuracy using MLP, LSTM, and CNN was 83.33%, 81.87%, and 80.42%, respectively. The effect of material microstructure characterization by LIBS spectroscopy combined with the PCA-MLP model is particularly remarkable. This study provides a new solution for the rapid analysis of microstructures of engineering materials.