Cutting tool wear state recognition based on a channel-space attention mechanism

Abstract

The identification of the cutting tool wear state is important to the cutting process. If the cutting tool is not replaced in time, it may have a negative impact on the processing quality of the workpiece. However, the identification of the cutting tool wear state is mainly achieved by establishing a mapping relationship between cutting tool information and cutting tool wear state. The feature extraction of signal during cutting tool wear directly affects the accuracy of wear state identification model. Poor selection of cutting tool wear signal features will lead to misjudgment of cutting tool wear state. This will lead to more fuzzy areas in the process of cutting tool state recognition. In this paper, the cutting force signal is transformed into time-frequency images through continuous wavelet transform to realize the feature extraction of cutting tool wear. In this paper, a channel-space attention mechanism is established to evaluate the effectiveness of different cutting tool wear signal features. The cutting tool wear states recognition model is capable of selective learning of different wear feature. Finally, a mapping relationship between cutting tool wear states and wear feature is established by means of a fully connected layer. The identification of the cutting tool's wear state is thus completed. The model proposed in this paper achieved 0.965, 0.965 and 0.966 in the three-classification metrics of Accuracy, Recall and F1score respectively. The model reduces misclassification due to difficult to classify areas (Fuzzy areas) between different wear stages. It has high recognition accuracy and anti-interference capability. The proposed model can accurately identify the cutting tool wear state based on the processing monitoring information. This will help in making more flexible cutting tool replacement decisions on smart manufacturing.