Abstract
The number of samples with various wear states gathered in processing is unbalanced because the wear rate of end mills fluctuates nonlinearly. There are still some limitations to the current approaches for identifying the end mill wear condition in this situation. It is primarily manifested in the dominance of the majority class samples in forming the classification decision boundary. The minority class samples and the samples with a more significant influence on the decision boundary in the majority class samples are closer and difficult to separate, making it challenging to identify the wear state of minority classes. Herein, a deep feature-weighted convolutional neural network (DFWCNN)-based end mill wear state identification approach is proposed to overcome the above limitations. First, the balance influence factor selectively reduces the weight of the samples that significantly influence the decision boundary to lessen the majority classes' dominance on the classification decision boundary. Then, the variance influence factor reduces the intraclass distance to lessen confusion between the samples at the classification decision boundary. The experiments conducted herein have proved that the proposed approach improves the identification accuracy of samples with various wear states, particularly accelerated wear state samples, in the case of sample imbalance.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have