Research On Prediction Method Of Tool Wear And Remaining Useful Life Based On Multi-Source Information Fusion

Machine tool is one of the most important processing equipment in modern machincry manufacturing industry,known as "industrial mother machine".As the "teeth" of machine tools,the wear state of cutting tools has a significant impact on the stable opcration of cutting,the quality of parts processing,and production efficiency.Therefore,the tool state can be scnsed in real time during the cutting process,and the corresponding state monitoring model can be built to realize the online monitoring of tool wear.With the increasing of tool wear,it is of important engineering application value to establish a reliable evaluation method of tool remaining useful life to accurately predict the remaining useful life(RUL)of the tool and ensure timely replacement of the tool before failure,which is helpful to reduce production cost,improve production efficiency and ensure processing quality.Due to the randomness of tool wear and the complexity of wear mechanism,the prediction method based on physical model has a large modeling error,and the model prediction accuracy is not ideal.Although the prediction accuracy of the data-driven prediction method is high,the prediction result is excessively dependent on the data quality.Aiming at the above problems,a tool wear prediction model based on the combination of physical model and data-driven model is established.Firstly,the wear model of milling cutter flank based on abrasive,adhesive and diffusion wear is established,and the change rate of wear width is used as the evaluation standard.Based on the monitoring data of tool wear state,one dimensional convolution neural network(1DCNN)prediction model is established.Finally,the particle filter fusion algorithm is used to transform the physical model of tool wear into a state transition equation,and the data-driven model into an observation equation,thus further constructing the state space model of tool wear.By using the fusion tool wear prediction method,the accuracy of tool wear prediction is improved.Aiming at the problems of low data utilization rate and large fluctuation of prediction accuracy when using single sensor data to predict the RUL of tools,this paper proposes a multisensor fusion method for predicting the RUL of tools.Firstly,feature extraction is carried out on the multi-sensor signals,and the monotonicity score of the feature time series is calculated based on Spearman’s rank correlation coefficient,and all the features whose monotonicity score is higher than the set threshold are screened out.Then,principle component analysis(PCA)is used to fuse and reduce the dimension of the selected multi-sensor feature information.The first principal component is selected and exponentially smoothed to construct a health indicator(HI)representing tool wear degradation.Finally,a tool wear degradation model is established based on the established health indexes.Markov chain monte carlo(MCMC)sampling algorithm is adopted to update the parameters of the degradation model in real time,and then the tool RUL at each moment is iteratively predicted.The proposed method can avoid the limitations of machine learning-based methods that need to rely on a large number of lifetime data to train the prediction model offline and the model has limited adaptability to new prediction tasks.Finally,by building an experimental platform for tool wear,three-dimensional cutting force signals and three-dimensional acceleration signals are collected in the whole life cycle,and the proposed method is verified and analyzed.The experimental results show that:compared with the single prediction model,the fusion tool wear prediction model has higher prediction accuracy;the RUL prediction method based on multi-sensor fusion has achieved good prediction accuracy.