Self-attention Tool Condition Monitoring Based On Appending-inspired Feature Selection

The rapid development of science and technology in the 21st century has continuously affected the industrial processing and manufacturing industry.Driven by new technologies,intelligence is one of the important needs.In the industrial processing industry,tool wear is indispensable,especially in some complex cutting processes,the tool wear is more serious,and the life of the tool is shorter than expected,which may affect the processing quality and even cause part failure.Therefore,it is necessary to detect and monitor the wear of the tool in time.At present,most tool processing equipment monitors tool wear by establishing a mapping relationship between tool wear status and signals collected at the current cutting stage.How to extract effective characteristic information representing the tool wear state from the massive sensor signals and quickly and accurately predict the tool wear state through the prediction model is the most urgent problem to be solved in intelligent tool monitoring.In the process of signal feature extraction,this paper proposes an additive extended feature selection method.By performing additional feature selection on the statistical features extracted from the signal in multiple domains,the feature information with time correlation can be obtained,which can make up for the lack of time correlation of the statistical features themselves,and retain the gap between the original signal and the new input signal to the greatest extent.without losing the relevant information of the tool wear status carried in the original signal.In order to realize the online monitoring of tool wear status,this paper designs a tool wear prediction model based on self-attention convolution neural network:in the process of tool processing,the signals collected by multiple sensors at different time points and the new signals continuously input are analyzed.In multi-domain extended feature selection,the obtained features are fused based on the self-attention convolution neural network,and the fused features are used as the input of the predictor to realize the tool wear status prediction with multi-sensor signals as the original input.The experimental validation of the proposed method on PHM2010 and NASA milling tool wear datasets shows that the model proposed in this paper can effectively predict the tool wear state and achieve better prediction results than other comparative methods,thus verifying the feasibility and effectiveness of the method in this paper.