Research On Tool Wear Prediction Method Based On Multi-sensor Feature Fusion And Ensemble Gaussian Model

Ensuring machining quality and improving machining efficiency is an effective way to enhance the core competitiveness of enterprises.However,Tool wear and breakage during the machining will cause dynamic error and unplanned shutdown which will affect the quality and production efficiency.The cutting force,vibration and acoustic emission during machining can reflect the cutting tool wear from different dimensions.How to extract effective features from them to predict tool wear is the key link to obtain the relationship between state features and quality features,which can realize the effective control of machining parameters to ensure quality and improve efficiency.In this paper,the correlation between signal features of multi-sensors and tool wear and the redundancy between features are analyzed to realize the fusion of signal features.The optimal feature subset of tool wear prediction model is obtained.An ensemble prediction model for tool wear is proposed to solve the shortcoming of poor robustness of single prediction model The main tasks are as follows:(1)The modeling flow of tool wear prediction model,tool wear form and its quantization method,and the advantages of multi-sensor fusion are described.The milling experiment of tool life cycle is carried out to collect multi-sensor monitoring data and tool wear data.The collected multi-sensor data were preprocessed.The above work lays a foundation for the data analysis and the construction of data-driven model.(2)The method of sensor feature extraction and analysis is studied.The 165-dimensional features are extracted from the signals of multi-sensors.The correlation and redundancy of these 165-dimensional features are analyzed by using Pearson correlation coefficient(PCC)and maximum information coefficient(MIC).The advantages of MIC in data mining is proved.So it was used to screen out the insensitive features.(3)A feature fusion algorithm is constructed by combining the MIC and kernel principal component analysis to reduce feature dimensions and redundancy.The proposed algorithm is used to conduct feature fusion for different combinations of multi-sensors.The correlation and redundancy after fusion are analyzed and compared with those before fusion,which proves the effectiveness of the fusion features.(4)An Gaussian ensemble model(Bagging-GPR)is improved by improving the diversity factor and accuracy of the submodel.The superiority of the model constructed in this paper is proved and the shortcoming of poor robustness of single prediction model is solved.The priority of feature fusion under different combinations of three sensors was determined by Gaussian regression model.Based on the priority,the tool wear prediction was carried out in the situation of sensor data miss.Its R~2 could reach more than 0.96,which has a very great guiding significance for guiding the actual tool wear prediction.