Research on Tool Remaining Useful Life Prediction Based on Bayesian Optimization of CNN-BiLSTM

In the prediction of the remaining useful life of milling cutters, the direct method relies on offline monitoring, and the measured data cannot be automatically and timely fed back into the prediction of the remaining life. To address this issue, this paper develops a milling cutter remaining life prediction system based on Bayesian optimized CNN-BiLSTM. Firstly, the in-machine wear data of the tool is obtained based on the optical image method, and an interface for the in-machine vision device is established to automatically transmit the data to the cloud server, enabling persistent storage of the data. Secondly, a deep learning platform is built on the cloud server, using wavelet packet decomposition technology and the least squares method to fit and denoise the data. The CNN-BiLSTM model is used to automatically extract features and predict the life, and the Bayesian optimization algorithm is used to optimize the hyperparameters of the model. Finally, experiments are conducted to compare with other models, and the results show that this system can timely obtain the latest collected data to correct model deviations, resulting in higher prediction accuracy.