Research On Tool Wear Monitoring System Based On Multi-source Information Fusion

The machining accuracy of the machine tool mainly depends on the quality of the machine tool,control algorithms and cutting tools,etc.,and is the core index for evaluating the performance of the machine tool.As the direct implementer of the machining workpiece,the cutting tool plays a vital role in the accuracy and quality of the workpiece surface.Timely and accurate grasp of the tool wear state can effectively ensure the safety of personnel and equipment and production efficiency during the cutting process.This thesis takes tool wear monitoring as the starting point,takes multisensor signals and multi-monitoring models as the main research objects,adopts the method of multi-source information fusion to carry out research,and builds a set of signal analysis and processing,feature extraction optimization,tool wear state identification.Based on the comprehensive model system,a tool wear monitoring model based on multi-sensor feature fusion and a tool wear monitoring model based on multi-model weight distribution fusion are established.The research contents of this paper mainly include the following aspects:(1)Selection of smooth processing signal and wavelet threshold denoising of multi-sensor signals.Aiming at the invalid signals at the beginning and end of the cutting-in and cutting-out process of the tool,the data is eliminated by using the criterion,which improves the quality of the data.In view of the large amount of noise doped in the signal,the wavelet threshold denoising algorithm is used.Based on analyzing the effect of the threshold and the vanishing moment order on the signal denoising,the optimal parameters of the wavelet threshold denoising are established,and the multi-sensor signal is de-noised.Noise processing,so that the noise signal in the multi-sensor has been well suppressed.(2)Feature extraction and feature selection are carried out on the multisensor signals,which provide input conditions for the subsequent modeling process and wear prediction.The original signal data does not correspond to the wear value one-to-one.It is necessary to "reduce" the data information to extract the features that effectively reflect the wear amount of the tool.Therefore.statistical-based features are extracted in the time domain,power spectral features based on fast Fourier transform are extracted in frequency domain,and the wavelet energy ratio feature based on wavelet transform is extracted in the timefrequency domain.The correlation between the proposed feature and the tool wear amount is analyzed,and the Pearson correlation coefficient is used to select the feature,and the correlation between the tool wear amount and the tool wear amount is obtained.Strong feature space.(3)A tool wear prediction method based on multi-sensor feature fusion is proposed.Aiming at the problem of one-sidedness,limitation and low precision in predicting the tool wear amount from a single sensor signal,a tool wear prediction model based on multi-sensor feature fusion was proposed and built.The multi-sensor features are fused in the feature layer,the dimensionality of the fusion features is reduced by principal component analysis,and the BP neural network is used for training and learning,and a fusion model is established.Finally,a comparative experiment with a single sensor signal model is designed to verify the superiority of the proposed fusion model.(4)A tool wear prediction method based on multi-model weight distribution fusion is proposed.Aiming at the problem that a single monitoring model is difficult to accurately predict the amount of tool wear and is one-sided,a tool wear prediction model based on multi-model weight allocation and fusion is proposed and built.Through the training and learning of regression tree,BP neural network and support vector regression,the weights of each model are extracted,and the three models are fused at the decision layer to establish a fusion model.By comparing the design with a single prediction model,it is verified that the fusion model is superior to each single model.