Research on Carbon Content Classification Method of PrNd Alloy Based on Acoustic Vibration Detection and SDP-VGG Network

Aiming at the problems of long period and high cost in the detection of carbon content in praseodymium neodymium alloys by traditional chemical analysis methods, a soft measurement method for carbon content in praseodymium neodymium alloys based on acoustic vibration detection and SDP-VGG network was proposed. The praseodymium neodymium alloy is restrained by parallel support and stimulated by middle percussion so that the bending vibration can generate acoustic signals. Using the SDP acoustic signal feature visualization method, the acoustic signal of the praseodymium neodymium alloy is converted into a two-dimensional image after preprocessing. Classification of SDP images of praseodymium neodymium alloy acoustic signals with high carbon content (500×10^-6 and above) versus low carbon content (below 500×10^-6) using a convolutional neural network model with a convolutional autoencoder. The research results show that the convolutional neural network with a convolutional autoencoder is capable of directly learning features from pre-processed SDP images, eliminating the need for manual feature extration. Compared with the SVM and BP neural network models that directly use the time-domain array or frequency-domain array of acoustic signals as input, our proposed model achieves a classification accuracy of up to 98.3%, enabling rapid and accurate differentiation between high-carbon and low-carbon praseodymium neodymium alloys.