Research On Tool Wear Prediction Based On Multi-dimensional Time-frequency Characteristics

The wear of NC tool in the cutting process will directly affect the surface quality and dimensional accuracy of the workpiece.After the tool reaches a certain wear stage,if the tool is not replaced in time,it will seriously affect the production quality of the part,and the premature replacement of the tool will cause the increase of production cost and the waste of resources.The traditional processing workshop mainly depends on the operator to judge and choose the right time to change the tool according to the experience.However,the modern intelligent manufacturing mode gradually improves the requirements of automation,unmanned and punctuality of the production system,and the emphasis of each manufacturing enterprise on lean production gradually deepens,which can no longer meet the actual needs.Based on the theoretical basis that the mechanical signal changes with time in the process of tool wear and the technical support brought by the development of modern signal processing technology,the realtime prediction of tool wear through multi-dimensional time-frequency characteristics provides a new research idea for the key decision-making process of tool replacement in the whole manufacturing system.In this paper,the application of multi-dimensional time-frequency characteristics in tool wear prediction is studiedFirstly,this paper summarizes the important work in the field of tool wear monitoring at home and abroad based on wear mechanism and prediction method,analyzes the bottleneck problems,and provides theoretical basis for further research.Then,aiming at the problem that there are a large number of features unrelated to tool wear extracted after signal processing,the maximum mutual information coefficient(MIC)is introduced into the effective sorting and screening of features,and feature fusion is carried out through KPCA,finally low-dimensional and high correlation features are obtained,which provides reliable input data for the subsequent establishment of prediction model.Thirdly,this paper studies the application of an integrated learning algorithm,gradient lifting regression tree,in tool wear prediction.Compared with traditional boosting,which directly weights multiple samples,gbrt can more effectively approach the real value by calculating the residual in each iteration and building a model in the direction of reducing the residual.In addition,traditional machine learning algorithm can not In this paper,the index of model prediction reliability is given,and a multi kernel correlation vector machine model(multikernel)is proposed RVM),on the premise of ensuring its prediction accuracy,can also give its confidence interval,so that the reliability of the model can be quantified.Finally,this paper studies the application of deep learning technology in tool wear prediction.Compared with traditional machine learning algorithm,deep neural network can adaptively extract features and save complex feature dimensionality reduction process.Through a large number of parameter optimization process and performance comparison of various algorithms,the proposed two-way gate control cycle unit(Bi GRU)and noise reduction self encoder(DAE)are applied to the tool wear prediction Good performance was found in wear prediction.