Analysis Of Hob Vibration Signal And The Method Of Fault State Identification

Hob status has a very important effect on hobbing,hob damage will not only lead to the processing of gear precision and quality decline,the rate of failure to improve;What is more,in the long run will cause hobbing machine bed irreparable damage.The hob vibration signal can clearly reflect the hob state and is easier to collect than the force signal.In order to realize the vibration control of hobbing machine and improve the quality of hobbing,this paper takes hob of hobbing machine as the research object,explores the response mechanism of hob vibration,researches the denoising method of hob vibration signal in hobbing process and the fault status identification of hob.The main work of the thesis includes:(1)Establishment of mathematical model of hob vibration response.A hob vibration balance equation is established based on the euler beam theory.The formula for calculating the vibration displacement of the hob along any direction in the hob coordinate system is derived by modal analysis,and the mathematical model of the vibration displacement and acceleration response in the X and Z directions of the workpiece coordinate system is established based on the coordinate transformation principle.(2)Research on noise reduction method of non-stationary vibration signal based on improved CEEMDAN and rough set.In view of the characteristics of complex components and random noise of hob vibration signals in actual hobbing conditions,a new method for non-stationary vibration signals denoising is proposed,which combines CEEMDAN,rough set feature evaluation and wavelet threshold method,to mine the weak fault information hidden by noise.This method aims at the mode aliasing of EMD decomposition and the default parameters of CEEMDAN which are not applicable to hob vibration signal.The selection criteria of CEEMDAN decomposition parameters are established,and the key parameters of CEEMDAN can be selected based on the characteristics of the signal itself.Aiming at the problem that a single index can easily lead to the recognition deviation,the multi-category features reflecting the IMFs status are extracted,and the multi-category features are evaluated based on the feature evaluation method of rough set to further obtain the comprehensive evaluation of IMFs.Based on the comprehensive evaluation index,the noise-containing IMFs was distinguished from the noise-free IMFs.Finally,different wavelet basis functions were used to denoising the noise-containing IMFs.The envelope spectrum and time-frequency spectrum of denoising signal are analyzed,and the denoising components of hob vibration signal are judged.(3)Fault diagnosis method of hob based on firefly parameter optimization and stacking sparse autoencoder.An intelligent fault diagnosis methods based on deep neural network with stack sparse autoencoder are put forward based on the hob vibration signal after denoising with different fault types to avoid the traditional artificial feature extraction and state recognition algorithms rely heavily on a priori knowledge of the problem,realize the hob fault type intelligent identification end-to-end hob.Firstly,the sample data and corresponding labels for training are constructed and the frequency domain information is obtained.Then,taking the prediction error rate of the minimum test set as the objective function,the superparameter of the network structure of the stacking sparse self-encoder as the independent variable coordinate,the network structure parameters of the stacking sparse self-encoder were obtained by firefly parameter optimization algorithm.Finally,the training samples were input into the first sparse autoencoder for training,and the output of the above layer was input to the lower layer.A single autoencoder was trained layer by layer.Finally,the global weight and bias parameters were reversely adjusted by BP algorithm to complete the model training.Finally,the trained model is used for fault identification of hob,and compared with other fault identification classifiers,the experimental results show that the model has a high fault identification rate and identification stability,which lays a foundation for its potential application in engineering.