Remaining Useful Life Prediction Of Grinding Wheel Based On Deep Learning

Grinding wheel is the core grinding part of CNC grinding machine.Its working state and wear degree directly affect the processing quality,processing cost and operating efficiency of CNC grinding machine.Based on the data collected during the operation of the grinding wheel,combined with the feature selection algorithm,this thesis studies the prediction method of the remaining useful life of the grinding wheel based on deep learning.Aiming at the characteristics of high noise and complex correlation of the original data,a two-stage filtering feature selection algorithm based on Copula entropy and Max-Relevance and Min-Redundancy(m RMR)is proposed,and fusion variational mode decomposition(VMD)to preprocess the data.In order to predict the remaining useful life of the grinding wheel,a RUL prediction model based on the multi-scale AM-LSTM network is proposed,and the reliability of the prediction results is estimated by uncertainty quantification.Finally,combined with the predictive maintenance requirements of grinding wheel,a grinding wheel health monitoring and life prediction system is designed and implemented.The main research contents and innovations are as follows:(1)Aiming at the problems of high dimensionality,noise and redundant information of the original grinding wheel data,a data preprocessing method based on feature selection is proposed.Firstly,a two-stage feature selection algorithm combining Copula entropy and m RMR is proposed,which searches for the feature set with the highest correlation with the predicted target and the lowest redundancy.Secondly,in view of the problem that traditional methods tend to ignore low-scale features,the non-stationary data in the collection are decomposed and reconstructed in combination with VMD decomposition,so as to improve the data’s ability to represent the degradation of grinding wheels.Finally,the proposed method is applied to the grinding wheel data set,and the original data is reduced from 55 dimensions to 15 dimensions.At the same time,the Copula entropy between the data set and the grinding wheel RUL is increased from 0.81 to 3.22,which verifies the effectiveness of the proposed method.(2)Aiming at the problem that the degradation mechanism of grinding wheel is complex and it is difficult to effectively predict the life of grinding wheel,a grinding wheel RUL prediction algorithm based on multi-scale AM-LSTM network is proposed.Aiming at the difference in time series characteristics of data on multiple time scales,a multi-scale learning strategy and attention mechanism are introduced on the basis of traditional long short-term memory neural network(LSTM)to improve the feature mining ability of the model for multi-parameter long time series data.The experimental results show that compared with the Bi-LSTM model,the root mean square error and absolute average error of the proposed model in the RUL prediction of grinding wheel are reduced by 31.07% and 39.54%,respectively,which has higher prediction accuracy.The prediction effectiveness of the model is further verified in the aero-engine RUL prediction experiment.(3)Aiming at the credibility evaluation of the prediction results of the deep learning model,on the basis of the multi-scale AM-LSTM network,the MC Dropout method and the ensemble model are combined to build a hybrid model to estimate the uncertainty of RUL prediction.The experimental results show that compared with the MC Dropout method and the ensemble model,the hybrid model can further improve the prediction accuracy of the model while obtaining a reliable prediction uncertainty estimate,and the ability to distinguish out-of-distribution data is more reliable.(4)In order to meet the predictive maintenance requirements of the grinding wheel during the working process,the selection of the database,the deep learning network framework and the design of the algorithm program were comprehensively considered.At the same time,a visual graphical operation interface was designed and developed based on Py Qt,and the realization and testing of the grinding wheel health monitoring and life prediction system based on deep learning was completed.The grinding wheel health monitoring and life prediction system based on deep learning can provide data reference for operation and maintenance personnel to make grinding wheel maintenance decisions,and has certain practical value in improving the quality and efficiency of workpiece grinding and reducing processing costs.