Intelligent Classification of Wear Particles Based on Deep Convolutional Neural Network

Abstract

Abstract The intelligent classification of wear particles has remained a high priority research area for ferrography technology and industrial tribology. In this study, five deep convolutional neural network (DCNN) models are used for identification of seven kinds of wear particles. Instead of manually designing and selecting the features, the proposed DCNN realizes an end-to-end processing. Wear particles’ dataset is built by various kinds of tribology experiments. The experimental results show that the accuracy of DenseNet121 on the test set is 88.39%. A conclusion can be drawn that DCNN is suited for wear particles’ classification and can be put into practical use in condition monitoring systems in the future.