An Online Monitoring Method for Intelligent Tool Wear Prediction Based on Deep Learning

In response to the shortcomings of traditional monitoring methods, which often rely on empirical approaches or signal processing techniques and suffer from poor real-time performance and low accuracy, a milling tool wear monitoring model combining CNN with LSTM based on attention mechanism (ATT) is proposed. Firstly, the wavelet scalogram of the axial cutting force and its corresponding tool wear status are obtained. CNN is used to adaptively extract deep multidimensional features from the scalogram, which avoids the issue of information loss that might occur with traditional feature extraction methods. The multidimensional features are then input into the LSTM, which is well-suited for handling time series, and regression is used to accurately predict tool wear during the machining process. The ATT mechanism is introduced into the LSTM model, assigning different probability weights to the hidden layers. This addresses the issue of information loss in long sequences, enhances the influence of important features, and reduces the impact of less relevant features on predictions, thereby improving the model's prediction accuracy.