Research On Predictive Evaluation Method Of Tool Wear Based On Multi-information And Multi-model Fusion

In this paper,the predictive evaluation method of CNC milling cutter wear is deeply studied and analyzed by using the data-driven method under the background of Industrial Big Data and Prognostics Health Management application.In this study,firstly the signal collected in the milling process is fully explored and analyzed,including truncating the signal's invalid data,filtering the abnormal values,and the analysis of the periodicity,the stability and the power and energy characteristics of signal.And then,the statistical method?FFT transform and WT transform are used respectively to extract the statistical features from the time domain?the spectrum and energy features from the frequency domain and the features of the wavelet coefficients and energy distributions ratio from the joint time-frequency domain.Besides,the extracted features are classified and merged according to the signal type and its axial direction,and subsequently are used for the research of the multi-information feature fusion.In this paper,the features selected by the F-test score and the Mutual Information(MI)score be used to improve the speed and accuracy of the model fitting.In this article,the characteristics of multi-information fusion and the main factors influencing the tool wear are experimentally verified and analyzed by using the Decision Tree Regression(DTR)and the Support Vector Regression respectively.The Results show that the effect of multi-information fusion is better than single information generally,and the force signal features and the features on the X axis are the main factors of influencing the tool wear.Finally,the integration method in machine learning field is introduced as a strategy of multi-model fusion,and it uses DTR as its basic learner.In the evaluation process of the multi-model method,the multi-model methods and the single model are analyzed by experiments and comparative analysis in three aspects of the accuracy,stability and applicability of the model.The result is that the multi-model fusion strategy based on the integrated method is superior to the single model in the all above three indexes,which verifies the effectiveness of the multi-model fusion strategy on the tool wear predictive evaluation.