CDCNN: A Model Based on Class Center Vectors and Distance Comparison for Wear Particle Recognition

Abstract

Mechanical wear failure is the main cause of device failure, and wear particles can be utilized to judge the degree and mechanism of mechanical wear. However, with the further study by researchers, the classes of wear particles are likely to increase. As the number of classes increases, it is necessary to reconstruct and retrain the previous automatic recognition model of wear particles. To solve this problem, a novel convolutional neural network (CNN) model based on class center vectors and distance comparison, called CDCNN, is proposed in this study. The model can not only accurately classify the particles that appear in the training set, but also identify the new classes that were not seen during training. This work can show its advantages when the classification criteria for wear particles have changed or new classes of particles have appeared. We applied the proposed model and the previous convolutional neural network model for the wear particle dataset, and compared the classification results of the models. The comparison results show that the model has strong classification ability and generalization ability.