Improved Residual Network and LSTM Hybrid Model Based Tool Health Monitoring

Health monitoring is crucial for intelligent manufacturing systems to reduce downtime, avoid major hazardous accidents, improve work efficiency, and save costs. The current mainstream health monitoring methods, such as CNN and LSTM, donot consider sequence and time dependencies together. To solve the above problem, an improved residual network and LSTM hybrid model based tool health monitoring is proposed in this paper. First, in the experiment, Wavelet transform was employed as the initial step for time series reconstruction, with the aim of achieving both dimensionality reduction and denoising. The resulting reconstructed time data was fed into the enhanced hybrid model. Then, the improved hybrid model overcomes the shortcomings of the single model, and avoids the mutual interference problem of the conventional combination model in feature extraction. Finally, hybrid model utilizes parallel Deep Residual Networks (DRN) and Long Short- Term Memory (LSTM) networks to extract high-dimensional features from the data. These extracted features are subsequently fed into the fully connected layer via an attention mechanism module to yield the predicted results. The universality and accuracy of the method in the paper were verified through the PHM20 10 dataset.