Physics-guided Gaussian Process Model And Its Application In Tool Wear Monitoring

With the development of science and technology,the intelligent processing of manufacturing industry has become one of the most attractive research topics.As an important part of intelligent machining,tool wear condition monitoring is conducive to ensuring machining quality and machining accuracy,shortening production cycle and reducing production costs.However,the related research on tool wear condition monitoring is still not perfect,which has shortcomings in both physical and data-driven models.The physical models often apply empirical or simplified formula to model the wearing process which often loss prediction accuracy under changing cutting parameters,and faces the identification of unknown parameters.On the other side,the data-driven models estimate the tool wear by monitored data without considering the mechanisms of tool wear,resulting in low model generalization and results interpretation.In order to solve the shortcomings of physical models and data-driven models,this study adopts the method of data-physical fusion model.The main research contents and innovations of this paper are summarized as follows:1)In this paper,the wear mechanism and form of tool are studied,and the tool wear label is expanded.The cutting force signal is preprocessed by signal clipping and wavelet denoising.Then,features are extracted from the time domain,frequency domain and time-frequency domain of the force signal.Finally,Pearson correlation analysis is used for feature selection,which the results show that the selected features can better reflect the change of tool wear.2)Based on the mechanism of cutting force,the influence of shear force and friction force on milling force is analyzed,which the micro-milling force model is established considering the influence of tool jump on undeformed chip thickness.Subsequently,the least squares method and the recursive least squares method are used to solve the tool runout parameters and the milling force coefficients,respectively.The accuracy of the milling force model is verified by multi-condition experimental data.The relationship between milling force and tool wear is analyzed,and the physical model of tool wear is established based on the micro-milling force model.The effectiveness of the physical model of tool wear is verified by multi-condition experimental data.3)In order to solve the shortcomings of physical models and data-driven models,this paper proposes a physical-based Gaussian process model.Using the Gaussian process latent force model(GPLFM)technique,the Gaussian process(GP)is used to model the unknown input of the tool wear physical model,and the physical model is combined into the GP covariance function to construct a GP covariance function with physical information.By analyzing the behavior of unknown input in the physical model of tool wear,the appropriate GP covariance is selected to model the unknown input correctly.Due to the guidance of physical information,the proposed model has higher prediction accuracy and generalization than the traditional data-driven model under multi-condition experiments.