Tool wear estimation using a CNN-transformer model with semi-supervised learning

In the machining industry, tool wear has a great influence on machining efficiency, product quality, and production costs. To achieve accurate tool wear estimation, a novel CNN-transformer neural network (CTNN) model is proposed in this paper. In the CTNN model, the transformer model and convolutional neural networks (CNN) are used to process condition monitoring (CM) data in parallel, such as cutting force. The motivations are as follows. For one thing, both the transformer model and CNN can extract useful temporal features from CM data, and the learned temporal features by these two parts are fused to achieve accurate tool wear estimation. For another, CNN contributes to enhancing the transformer's ability to capture the sequence order. In addition, data noise introduces the aleatoric uncertainty to the estimation results. To quantify the aleatoric uncertainty, a negative log-likelihood loss function is employed to enable the model to output the probabilistic distribution associated with tool wear. In such cases, the model outputs both the tool wear and variance, and the variance is learned within the model in an unsupervised manner. Finally, the effectiveness and superiority of the proposed method are validated on a public milling dataset. It is found by experiments that both the transformer model and CNN play important roles in tool wear estimation, and better performance can be obtained when they are used in parallel. In summary, the experimental results suggest that the proposed model can obtain promising results in tool wear estimation.