Study Of Tool Remaining Life Prediction Method With Deep Fusion Of Multi-source Information

As a key part of CNC machine tools,the wear and degradation state of the tool directly affects the quality of the product processing.In the machining process,online direct measurement of tool wear requires frequent pauses in the machining process,which is costly.A feasible method is to monitor vibration,spindle current,working condition and other status information online during machining process,and analyze and process the realtime collected multi-source status information to indirectly predict the tool wear status and remaining service life.Effective prediction of tool wear status and remaining service life allows timely tool repair or replacement,which is of great significance for maintaining tool machining accuracy.In this research,we adopt a data-driven approach to model a tool's remaining life and offer a technique for predicting a tool's remaining life based on deep fusion of multi-source data,which primarily covers the following characteristics.1.A detailed analysis of multiple signals characterizing the tool wear state is carried out,vibration signals and current signals are selected as data sources for analyzing the tool wear state,a data pre-processing method combining wavelet threshold denoising and outlier processing is proposed for the characteristics of the data sources,and a multi-dimensional correlation feature screening method is proposed around the time-domain,frequencydomain and time-frequency-domain characteristics of the data.2.A tool remaining life prediction model combining principal component analysis(PCA)and multicore weighted least squares support vector machine(W-LSSVM)is proposed to obtain the health index Hotelling T2,which reflects the tool wear status,using PCA,and then modeling and analysis using multicore W-LSSVM to achieve effective prediction of tool remaining life.3.Aiming at the uncertainty caused by artificial feature extraction in the tool remaining life prediction method based on traditional machine learning,deep learning can automatically refine the learning ability of multi-level features,and a fusion neural factorization machine(NFM),convolution The tool remaining life prediction method based on neural network(CNN)and long short-term memory(LSTM)network models combines the characteristics of the tool's collected data and the respective characteristics of different neural networks,which can effectively perform low-frequency operating condition data and high-frequency state data of the tool.Feature learning and mining,in the case of sufficient training samples,a higher prediction accuracy of the remaining tool life can be obtained.