Tool Wear Prediction Based on BiLSTMA Networks

As a key part of CNC machine, the degradation state of tools directly affects the quality of workpieces processing. It is an urgent problem to extract effective feature information from multiple sensors signals and establish an accurate tool wear prediction model. In this article, a tool wear prediction method based on bidirectional long short-term memory attention neural networks (BiLSTMA) is proposed to this problem. Firstly, the paper divides the time series data into data of different time periods and extracts the time domain, frequency domain, time-frequency domain features of the local time period. Secondly, BiLSTM neural network is used to further extract deep features of time dimension from all local features. Finally, an attention mechanism is introduced to reasonably allocate attention weights of the deep features for tool wear prediction. The effectiveness of this method in improving tool wear prediction accuracy is verified by experiments.