Tool wear monitoring based on an improved convolutional neural network

Tool condition is the key factor affecting the quality and efficiency of precision cutting of parts. As tool wear is inevitable during machining, tool wear status during machining must be regularly monitored. This study proposes a combined convolutional neural network and support vector machine (SVM) approach for tool wear status monitoring. First, 1D cutting force data are wavelet-transformed and converted into 2D spectrogram. Second, the leaky-ReLU activation function is adopted to enhance network robustness. Third, an SVM classifier is used to replace the traditional Softmax function to improve the model generalization capability. Finally, the cutting force signal of the tool used for the machining of the aero-engine integral blisk is verified. The accuracy of the constructed network model can reach 98.28 %. Moreover, the proposed model has a simple structure, requires a small number of parameters, and has good robustness and reliability.