Research On Wear Condition Monitoring And Fault Diagnosis Method Of High-speed Dry Cutting Hob Based On Spectral Graph Wavelet Transform

High-speed dry cutting CNC gear hobbing machine is widely used in aviation,shipping,automobile,engineering machinery and other fields of gear manufacturing.As one of the core components of high-speed dry cutting CNC gear hobbing machine,the high-speed dry cutting hob is an important tool for gear rough machining.In the process of machining,the high-speed dry cutting hob will bear the periodic alternating impact cutting load.At the same time,it is affected by the mechanical,thermal,chemical and other factors caused by the fierce friction between the hob teeth and the gear workpiece,resulting in wear and faults on the high-speed dry cutting hob,thus affecting the machining quality and safety performance of CNC gear hobbing machine.Therefore,the research on wear condition monitoring and fault diagnosis methods of high-speed dry cutting hob is of great significance to guarantee the safe,reliable and efficient operation of CNC gear hobbing machine,and to improve the machining quality and efficiency of gears.However,in the high-speed dry gear hobbing process,due to the influence of high speed,complex operating conditions,multi vibration source coupling,random interference,white background noise and other factors,the monitoring signal is doped with strong background noise,resulting in low signal-to-noise ratio and poor accuracy of state features extracted from the monitoring signal,which seriously affects the reliability and stability of hob condition monitoring.Therefore,aiming at the problem of weak wear feature extraction and sensitive fault feature extraction of high-speed dry cutting hob under strong background noise,this paper combines the spectral graph wavelet transform theory to carry out in-depth research from the perspective of graph for the vibration signal denoising preprocessing,wear and fault feature extraction,as well as the wear condition monitoring and fault diagnosis of the hob.The main research work of this paper is as follows:(1)Aiming at the problem that the traditional time series processing methods ignore the local relationship between discrete points,the multi-scale analysis of time series signal based on graph representation is studied.On the one hand,according to the matching relationship between the time-ordered characteristics of time series signal and the sequence structure of one-dimensional path graph,the one-dimensional path graph is introduced to represent time series signal,and then it can be decomposed into components of different frequency intervals in the spectral graph domain combined with spectral graph wavelet transform,so as to realize multi-scale analysis of time series signal from the perspective of one-dimensional graph domain.On the other hand,the amplitude of the time series signal is normalized to grayscale pixel intensity,which is further converted into a two-dimensional grayscale image,and then the twodimensional neighbor graph is introduced to represent the two-dimensional grayscale image,which can be decomposed into multi-resolution subgraphs using spectral graph wavelet transform,so as to realize the multi-scale analysis of time series signal from the perspective of two-dimensional graph domain.The introduction of graph takes into account the similarity information between adjacent discrete points,which greatly enriches the input information in the time series signal analysis process.(2)In light of the problem that there exists strong background noise in vibration signal during gear hobbing,and the traditional denoising methods cannot balance the dual requirements of strong background noise elimination and signal detail feature retention,the research on the denoising method of hob vibration signal is carried out.A threshold denoising method combining one-dimensional path graph representation and spectral graph wavelet transform is proposed.The threshold criterion is used to perform one-time filtering on the spectral graph wavelet coefficients,and a model for selecting optimal decomposition levels of spectral graph wavelet transform is established based on detrended fluctuation analysis.Also,the influence of optional parameters such as weighting function,threshold and threshold function on the denoising performance of the proposed method is further discussed.The denoising experiments of simulated signals and measured vibration signals prove that this method can effectively eliminate the noise interference components of the non-linear and non-stationary signal,and can retain the fine structure and detail characteristics of the signal as much as possible.(3)In view of the problem that there is no effective detection for hob faults,and the fault features extracted by single-scale analysis has lower accuracy,the research on the hob fault diagnosis method based on multi-scale feature extraction is carried out.On the basis of using one-dimensional path graph to represent the vibration signal,a hob fault diagnosis method based on multi-scale feature extraction of one-dimensional path graph combined with random forest is proposed.The multi-scale symbol dynamic entropy of each scale is extracted from the spectral graph domain to form the feature vector,which is input into the random forest for hob fault identification.The beetle antennae search is used to determine the optimal parameters of random forest,and an adaptive strategy is proposed to improve the global optimization ability and convergence speed of the beetle antennae search algorithm.The experimental results of typical hob faults under two groups of operating conditions show that the method can effectively identify the fault types of hob,and has high classification accuracy and generalization performance.(4)Direct at the problem that the hob replacement only depends on the personal experience of operators,and the weak wear features are difficult to effectively extract,the research on the hob wear condition monitoring method based on two-dimensional domain feature extraction is carried out.On the basis of using two-dimensional neighbor graph to represent the vibration signal,a hob wear condition monitoring method based on fine texture feature extraction of two-dimensional neighbor graph and machine learning classification model is proposed,and the multiple dimensional scaling algorithm is applied to reduce the dimension of fine texture features,so as to avoid the drawbacks that the high-dimensional features will bring difficulty in computing and storage,and reduce the learning generalization ability of classification model.Then the monitoring performance of different texture feature descriptors combined with various classification algorithms is compared,and the optimal feature dimension and anti-noise ability of the monitoring system are analyzed.The experimental results of hob life cycle wear indicate that the proposed method can accurately monitor the hob wear condition with fewer feature dimensions and strong background noise,and has high robustness.In this paper,the application of spectral graph wavelet transform in mechanical signal analysis and processing is studied deeply and systematically.It is applied to the denoising preprocessing and feature extraction of hob vibration signal in high-speed dry cutting gear hobbing process.Combined with random forest,linear discriminant analysis and other classification algorithms,a complete set of wear condition monitoring and fault diagnosis methods of high-speed dry cutting hob are established.The research work and methods in this paper have important theoretical significance and engineering application value to improve the maintenance support capability and optimize the hob replacement strategy for high-speed dry cutting CNC gear hobbing machine.