Research On Cutting Tool Remaining Useful Life Prediction Of CNC Lathe Based On Bayesian Updating

As an important executive component in machining systems,cutting tools have been highly valued by academia and industry.Tool wear condition affects not only machining accuracy and surface quality,but also machining cost.To balance tool availability and machining cost,accurate remaining useful life prediction is of great importance.The accuracy of tool remaining useful life prediction for new tool samples is low by using only historical data to build a model,so it is necessary to find out the online prediction method of tool remaining useful life in combination with tool wear monitoring model,and then improve the accuracy of new tool remaining useful life prediction.In this paper,the CK6140 CNC lathe tool is used as an example to design tool degradation orthogonal experiment,and collect tool wear data and vibration signal data.A deep transfer learning-based tool wear monitoring model is constructed by considering tools machined under variable working conditions,so as to obtain the relational mapping between vibration signals and tool wear values.Considering differences of individual tools and wear degradation rates,a tool degradation model is developed based on the Wiener process model.A tool RUL online prediction model is developed by combining tool wear monitoring output and Bayesian update method.The main research elements of this paper are as follows:Firstly,in order to obtain data to support the research method of this paper,orthogonal experiment is designed to establish the vibration signal dataset and the tool wear dataset.The L₉（3³）orthogonal experiment of tool degradation at three levels of cutting speed,feed rate and cutting depth was designed to collect the vibration signals of the tool during the cylindrical turning and to measure the tool wear value periodically.The tool vibration signal is processed and analyzed,and the sliding window method is used to the vibration signal downsampling.Use EMD-wavelet thresholding-based noise reduction method for signal noise reduction,and the processed signals are analyzed and extracted in the time domain,frequency domain and time-frequency domain.This section sets the data basis for online monitoring of tool wear and remaining tool life prediction.Secondly,to describe the relational mapping between vibration signals and tool wear values,a tool wear monitoring model based on deep transfer learning is developed.Deep learning methods are used to develop a model for monitoring tool wear under a working condition.A monitoring model based on CNN-LSTM by combining the strengths of CNN and LSTM is developed.The optimal values of the learning rate and the number of neurons in the hidden layer of the LSTM module in the model are solved iteratively by using the PSO algorithm to further improve the accuracy of the tool wear monitoring.The CNN-PSO-LSTM model under a working condition is used as a pre-trained model,and the model is trained by the model adaptation-based method with machining parameters in the fully connected layer,so as to realize tool wear monitoring under variable working conditions and provide a sample of online observations for online prediction of remaining tool life.Finally,to address the randomness of tool degradation and the variability of tool degradation rates,the online prediction model for tool remaining useful life based on the Bayesian method is developed.A single-stage Wiener process-based model and a three-stage Wiener process-based model are developed for tool degradation.And the EM algorithm is used to solve for the maximum likelihood estimates of the model parameters.The estimation of tool RUL is obtained accordingly.Two indicators,RMSE and MAPE,are used to evaluate the models and select the best model as the tool remaining life prediction model based on historical data.The tool wear output online from the tool wear monitoring model is used as the observations,and the model parameters are updated online using the Bayesian method,thus achieving online prediction of the remaining tool life.