Intelligent tool wear monitoring and multi-step prediction based on deep learning model

Abstract

In modern manufacturing industry, tool wear monitoring plays a significant role in ensuring product quality and machining efficiency. Numerous data-driven models based on deep learning have been developed to improve the accuracy of tool wear monitoring. However, tool wear monitoring under variable working conditions is rarely investigated. More importantly, for data-driven smart manufacturing, it is more meaningful and challenging to simultaneously achieve tool wear monitoring and multi-step prediction. To address the aforementioned issue, a novel framework based on feature normalization, attention mechanism, and deep learning algorithms was proposed for tool wear monitoring and multi-step prediction. Feature normalization was introduced to eliminate the dependence of local features on cutting conditions, and the attention mechanism was applied to enhance valuable information and weaken redundant information. Then a parallel convolutional neural network (parallel CNN) structure with different layers followed by Bi-directional long short term memory (BiLSTM) was developed for tool condition monitoring. Finally, based on the monitored tool wear values, a new tool condition prediction model based on the dense residual neural network (ResNetD) was proposed for short-term and long-term prediction of tool wear. Tool wear experiments under different combinations of cutting parameters were conducted to verify the proposed model, and the results showed that the proposed model has great advantages in efficiency and robustness compared with other data-driven models.