Application Research Of Machine Learning Algorithms In Tool Wear Status Assessment

Intelligent production is a hot topic in today's mechanical processing industry.And the realization of a fully automated production model of prod ucts has beco me the inevitable trend in the development of modern factories.Under this condition,how to make the production equipment independently judge its own operating status and make timely adjustments is an important part of ensuring its production quality and efficiency.The intelligent evaluation of the wear state of the tool can make an early warning before the tool is scrapped to ensure the production precision and equipment safety,which is of great significance to the intelligent production industry.In this paper,the cutting force signal and acceleration signal which are with strong correlation with the degree of the wear during the cutting are selected.And several machine learning algorithms are applied to the evaluation of tool wear state.The research mainly includes feature extraction of the original signals,the optimization of Support Vector Machine(SVM)model and the application of Random Forest(RF)algorithm in tool wear assessment.Since the original acceleration signal and the cutting force signal are inconvenient for direct analysis,the feature factors are extracted in the time domain and the frequency domain.Then the wavelet packet decomposition is used to obtain the energy percentage of the frequency bands.For the excessive feature quantity,there may be redundancy problems.In order to improve the analysis efficiency,the Principal Conponent Analysis(PCA)is applied to decrease the dimension of the eigenmatrix to reduce the correlation between the features.Cons:idering that the SVM model has a good processing ability of small sample and nonlinear data,its application in tool wear state evaluation is studied.Three machine learning algorithms are used to optimize the parameters in SVM model.The Cross Validation method is used to verify the recognition results of the tool wear state of the optimized SVM models.The results show that the SVM models optimized by these algorithms have the higher accuracy in evaluating the wear state of the toolFurthermore,aiming at the insufficiency of the time to build a SVM model,a tool wear state assessment model based on Random Forest algo rithm is established.Due to the good high-dimensional data processing ability of the Random Forest algorithm,the eigenmatrix without PCA is used as the data input to obtain the evaluation model The evaluation result shows that using the Random Forest algorithm to deal with the tool wear state assessment problem can take into account both faster response speed and higher recognition accuracy.The research shows that the SVM model optimized by the machine learning algorithms and the Random Forest algorithm model can effectively evaluate the tool wear state and lay a solid foundation for the establishment of online evaluatibn system.