Tool Wear State Identification Based on Frequency Domain Denoising and Frequencies-Separation Attention Networks

Real-time and accurate monitoring of tool wear is an important link to realizing intelligent manufacturing, which has an important impact on product quality and production cost. Due to its convenience and strong nonlinear mapping capability, deep learning has attracted extensive attention in the field of tool wear monitoring. However, the monitoring signal in the machining process will inevitably be interfered with by the noise from the machine tool, which will affect the models' reliability and optimization. In this work, the effectiveness of monitoring signals is analyzed, and a machine tool inherent characteristic frequency interception method and a time-frequency domain decoupling technique based on wavelet transform with tunable Q-factor are introduced to perform targeted noise removal during machining. The method is naturally interpretable. Meanwhile, a frequency-separation attention mechanism is proposed to weigh the frequency at different stages adaptively. The validity of the proposed method is verified by using real machining data. And the interpretability of the model is briefly analyzed.