Research On Wear State Monitoring Technology Of Relief Tooth Turning Tool Based On Machine Learnin

The shovel tooth turning tool is the turning tool of the gear hob,and its wear state directly affects the machining accuracy,efficiency and production safety of the hob.At present,manual monitoring of tool wear status relies heavily on machining experience and has a high cost.Therefore,it is of great significance to realize intelligent identification of the wear status of shovel-tooth turning tools.This paper takes the wear state of the shovel tooth turning tool as the research object,collects and processes the vibration signal during the machining process,and then uses the machine learning method to learn the wear characteristics of the shovel tooth turning tool,so as to realize the analysis of the wear state of the shovel tooth turning tool in engineering practice.Precise identification.The main work is as follows:1)In a processing workshop of an enterprise in Guizhou,an experimental platform for monitoring the wear condition of shovel tooth turning tools is built and an experimental data set is produced.Among them,the wavelet hard threshold denoising method is used to denoise the original vibration signal;on the basis of denoising,Principal Component Analysis(PCA)is used to extract the tool wear characteristics,and the characteristics of tool wear under different machining conditions are extracted.The collected shovel-tooth turning tool feature data is visualized and analyzed,and then,a data set of unbalanced identification of shovel-tooth turning tool wear status is produced.2)Aiming at the problem of unbalanced data of shovel-tooth turning tool wear state,a Regenerate-Adaptive Synthetic Sampling(Re-ADASYN)method is proposed.First,based on the k nearest neighbor(k-NN)samples of minority classes,calculate their comprehensive feature expression as the "base point",and use the "base point" as the direction regulator for sample generation;then,slightly move the generation of samples generated by ADASYN to the "base point".Move to generate second-generation samples;finally,use the second-generation samples with high-quality feature retention as the final generated samples.At the same time,a new method(SFROR)is proposed to evaluate the quality of the samples generated by the sampling algorithm,that is,after the vectorized dimension reduction of the data matrix,the quality of the generated samples is comprehensively evaluated based on the offset and feature retention.The proposed ReADASYN algorithm is verified on PHM2010 and self-collected shovel tooth turning tool data sets.The results show that Re-ADASYN is significantly better than other 7mainstream sampling algorithms in 8 classifiers and 6 evaluation indicators.3)An online recognition model of MFO-LightGBM shovel tooth turning tool wear state is constructed.Considering that the Lightweight Gradient Boosting Machine(LightGBM)has fast training speed and high accuracy,it has many hyperparameters,and manual parameter adjustment has a certain blindness.Therefore,using intelligent optimization algorithms such as GA,PSO and MFO to optimize the five hyperparameters of LightGBM in parallel,comparative experiments show that MFO is more suitable for this engineering scenario.Finally,the MFO-LightGBM shovel tooth turning tool wear state recognition model based on equalized samples is constructed,and the online recognition of the shovel tooth turning tool wear state is realized by means of incremental learning.4)A prototype system for identifying the wear state of the shovel-tooth turning tool is developed,which realizes the accurate identification of the wear state of the shoveltooth turning tool,and laid a foundation for the application in the industrial environment.