836449598 / interpretable-attention-based Goto Github PK
View Code? Open in Web Editor NEWTool wear monitoring plays an important role in improving product quality and machining efficiency of high-speed milling. As a typical data-driven algorithm, deep learning has been widely studied in tool wear monitoring, but it is rarely applied in practice as an independent algorithm up to now. This is mainly because the interpretability of deep learning does not meet the requirements of the industrial field. In this study, a novel attention-based dual-scale hierarchical LSTM (ADHL) is proposed. The ADHL can not only accurately monitor tool wear, but also access the interpretability from aspect of structure design and feature extraction. Because the prior knowledge of periodicity is introduced into the structure design, the extracted multi-scale features can cover almost all the characteristic periods. In addition, the periodicity of the features of interest can be analyzed based on the attention distribution. The effectiveness and feasibility of this method are verified from many perspectives on the high-speed milling experiments. To our best knowledge, this is the first attempt on the interpretability of deep-learning based tool condition monitoring algorithms.