Tool Wear Condition Monitoring And Remaining Useful Life Prediction Based On Information Fusion

With the continuous advancement of global industrial reform in the 21 st century,the intelligent development of CNC machine tools has become one of the keys to the sustainable development of manufacturing industry.Tool as an important part of CNC machine tool,its damage will affect the machining quality and production efficiency.Therefore,in order to ensure the high efficiency and accuracy of machining,real-time monitoring tool state,formulate effective tool change strategy,is very necessary.Aiming at the problem that a single sensor signal cannot fully reflect the changing information of tool wear,this paper discusses how to use multi-source information fusion technology and deep learning method to realize intelligent monitoring of tool wear state and prediction of remaining useful life,and verifies the feasibility of the method through experiments.The main research contents of this paper include:(1)Tool wear mechanism analysis and data processing.To begin,the tool’s wear process is outlined,followed by the tool was damage process and characteristics.Secondly,the source and collection process of experimental data were briefly introduced.and the collected data are preprocessed.Then,the multi-domain features were extracted from the processed data.Finally,Spearman correlation coefficient was used to screen the multi-domain feature set,and the feature space with high sensitivity to tool wear was obtained.(2)The tool wear condition monitoring method was studied.Firstly,machine learning method was used to establish tool wear condition monitoring model,and then the model performance was verified by experimental data.A multi-scale attention mechanism residual network(MSA-Res Net)tool wear condition monitoring method was proposed,taking into account the intricate feature engineering and low recognition accuracy of machine learning methods.This method directly fused the features of multi-source information and reduced the dependence on manual processing.Verifying the model,three groups of experiments were conducted,yielding results of 97.2%,94.9% and 96.2% accuracy respectively.Comparing and analyzing these results with other models,the MSA-Res Net model was found to be superior,thus confirming its superiority in tool wear state recognition.(3)The prediction method of tool remaining useful life was studied.To begin,a model of tool residual service life was created,combining a support vector regression machine and genetic algorithm,and its performance was tested through experiments.Subsequently,a prediction model of tool remaining useful life,based on deep residual shrinkage network and bidirectional long and short memory network(DRSN-Bi LSTM),was proposed.Verifying the model,three groups of experiments were conducted,with RMSE and MAE as the evaluation indexes.The results revealed that the average comprehensive indexes of the three groups of experimental results were 0.041,0.034 and 0.979,respectively.Compared with other types of prediction methods,the prediction accuracy of the remaining useful life of the proposed method was significantly improved.