Based on domain adversarial neural network with multiple loss collaborative optimization for milling tool wear state monitoring under different machining conditions

Abstract

In machining, it is crucial to monitor the tool wear status in real time to guarantee the quality of the workpiece being machined. Tool wear monitoring technology mainly reflects the tool state through the physical signals generated during the machining. At present, the technology faces many challenges in practical applications. When facing different machining scenarios, the model is difficult to adapt to new machining scenarios. Therefore, this study proposes a method to monitoring the tool wear state under different machining conditions based on Domain Adversarial Neural Network with multiple loss collaborative optimization (MLCODANN). This method takes the domain adversarial neural network as the framework and uses a multiple loss collaborative optimization method to adjust the optimization direction of the loss. It avoids the problem of conflict between the domain alignment and the classification loss, improves the convergence of model loss. In addition, this study used ResNet18 as a feature extraction network to extract features of the cutting signal. Meanwhile, the horizontal and vertical convolutional kernels $1\times k$ and $k\times 1$ are used instead of the convolutional kernel $k\times k$, which reduces model parameters and training time the and improves the model performance. Finally, through comparative experiments, it is proved that MLCODANN model has high accuracy in recognizing tool wear state under different machining conditions.