A Tool Wear State Recognition Model Based on Multi-Scale Convolutional Neural Network and Transformer Encoder

It is essential to keep track of the status of cutting instruments to guarantee the quality of machining, uphold safety standards, and boost productivity. With this in mind, the present paper introduces a model for identifying the wear state of tools that leverages a multi-scale convolutional neural network (MSCNN) and a Transformer encoder. Leveraging the multi-scale feature extraction of MSCNN along with the Transformer’s capacity for modeling global dependencies, the model achieves improved precision and robustness in recognition tasks. Initially, the model utilizes MSCNN to glean features across various scales from vibration signals, subsequently employs the Transformer encoder to discern broader contextual relationships, and concludes with classification u sing a dense layer. The findings from experiments show the model’s capability to accurately differentiate among initial wear, normal wear, and severe wear conditions, offering an innovative approach to tool health surveillance and fault detection.