ResGRUA Model for Tool Wear Prediction Based on Encoder-Decoder

Tool wear prediction is crucial to ensure machine parts quality and machining efficiency. Many machine learning methods, especially deep learning, have been applied to the researches of tool wear intelligent prediction. Aiming at the problems of poor adaptability and low long-term prediction accuracy in traditional tool wear prediction methods, based on encoder-decoder framework, we combine residual GRU network and self-attention, proposing the ResGRUA model to realize intelligent prediction of tool wear. Firstly, features from the input time series are extracted. Then, feature encoder based on bidirectional LSTM network is designed and applied on the generated feature vectors to learn the dynamic laws of two directions. The ResGRUA network as the model decoder is finally trained to predict machine condition. Experimental studies on an open source dataset are performed to validate the effectiveness and superiority of the proposed model. Meanwhile, the experimental results demonstrate that this method has high stability in training process and can predict tool wear more accurately.