Intelligent identification of material removal behavior during scratching of 4H-SiC based on acoustic emission sensing and unsupervised deep learning

This research presents an intelligent methodology to identify the material removal behavior during scratching of 4H-SiC based on acoustic emission sensing and unsupervised deep learning. A dedicated high speed scratching setup is developed, and the scratching depth is adjusted to selectively activate different material removal modes of 4H-SiC during scratching process. The material removal behavior changes from ductile deformation to brittle fracture with the scratching depth increasing from 50 nm to 400 nm at a constant scratching speed of 20 m/s. To differentiate various acoustic emission sources, an unsupervised convolutional auto-encoder and k-mean clustering methodology is employed. One-dimensional convolutional autoencoder is used for adaptive extraction of acoustic emission signal features, while k-means clustering algorithm is used to analyze material removal modes and damage types based on the extracted signal characteristics. Combined with material removal characterization, the appearance of specific clusters in different scratching tests is leveraged to map acoustic emission data to machining patterns. The results show that different acoustic emission signal features exhibit varying sensitivity to the damage formation during high-speed scratching of 4H-SiC. The parameters of amplitude, energy, and count are identified as the optimal characteristic parameters for distinguishing material removal modes and damage types for 4H-SiC. The acoustic emission sensing and deep learning approach presented in this study can be used to construct a real-time nondestructive tool to characterize and monitor material removal behavior during manufacturing processes.