A data augmentation method based on cGAN for tool wear state monitoring

In the process of tool wear, the amount of sample data collected in the normal wear stage is usually more than that in the initial wear stage and the sharp wear stage, which leads to the imbalance of the tool wear state data set, thus reducing the accuracy of the deep learning network model in predicting the tool wear state. To solve this problem, this paper proposes a data augmentation method based on cGAN for tool wear state monitoring. The category information as condition in cGAN is helpful for the generator to better capture the data distribution characteristics of tool wear samples, so as to generate samples similar to the real tool wear sample distribution. Specifically, the vibration signals in the milling process are collected. The vibration signals are converted into spectral data and input into the cGAN. The cGAN is trained by adversarial training between the generator and discriminator to learn data distribution characteristics and generate tool wear state sample data. The enhanced data set is input into the deep learning network model for classification to test the availability of the generated sample data. Experimental results show that training the deep learning network model with the enhanced tool wear states dataset can effectively improve the accuracy of the model in monitoring tool wear states, and the prediction accuracy reaches 98.1%.