Fault Diagnosis And Health Status Classification Of The Gear Of The Gearbox Based On Deep Learning

The gearbox is an important component for transmitting power in rotating machinery and equipment,and its condition determines the safe and stable operation of the entire equipment because of its extremely harsh operating environment and high failure rate.It is found that the gear is one of the parts of transmission most prone to failure.In many accidents caused by transmission,the most common problem is caused by gear failure.When gears fail,it will inevitably cause the gearbox to fail to operate normally,which in turn will affect the safe operation of the equipment and lead to serious economic loss or personal injury.In this thesis,the pre-processing method of vibration signal was firstly studied,and the denoise method combining Empirical Mode Decomposition(EMD)and wavelet thresholding was used to denoise the gear fault vibration signal,and the effectiveness of this method was verified by comparing with the de-noise effect of wavelet thresholding method and EMD.Then,the time domain,frequency domain and entropy features were extracted from the de-noise vibration signal to form a multi-domain original feature set,and the Fisher score evaluation criterion is used to further filter out the feature subset that can reflect the gear fault information from the original feature set.In order to investigate the fault diagnosis problem of gears in gearboxes,this thesis proposed two different methods for gear fault diagnosis and identification based on deep learning algorithms.Firstly,a Convolutional Neural Network(CNN)model was established,and the original feature set and the feature subset after feature selection processing were input to the network model for classification to examine the effect of feature selection on the classification accuracy is investigated.And the influence of different parameter settings of CNN model on the classification effect was investigated to analyze the classification performance of CNN fault diagnosis model.Finally,DBN-SVM fault diagnosis model is established to realize the classification and recognition of different fault types of gears,and it is compared and analyzed with DBN and SVM models.To address the problem that fault diagnosis is only applicable to achieve after the occurrence of a fault and cannot know the fault generation in advance,this thesis establishes a CNN-LSTM based gearbox healthy operation status recognition model,studies the influence of different network layers and learning rate on the classification results of the model,determined the optimal structural parameters,and realized the prediction of gearbox operation status.The study was also compared and analyzed with the classification effects of CNN and LSTM models.The contents and results of the present work are of great significance for predicting non-healthy operation status in advance,reducing the failure rate and ensuring the safe operation of mechanical equipment.