Research On Intelligent Fault Diagnosis Method Based On Deep Learning For Gearbox

With the rapid development of industrialization level,the function of mechanical equipment in modern industry has become more and more rich,powerful and efficient,the safety and reliability of mechanical equipment has become a problem that can not be ignored.Gearbox is one of the most common part in machinery transmission system.Because of heavy load intensity and long running and complex running condition,gearbox in the process of work may be affected by the influence of various abnormal problems and even may cause the stagnation of production,economic losses and casualties.Therefore,it is of great significance and application value to develop intelligent fault diagnosis method and technology of gearbox.This paper studies gearbox intelligent fault diagnosis method based on deep learning.The main research contents are as follows:(1)Aiming at the dependence of artificial feature selection and tedious pre-processing process of traditional gearbox fault diagnosis methods,and take full advantage of the adaptive feature extraction ability of deep learning,a gearbox intelligent fault diagnosis method based on one-dimensional convolution,bidirectional short and long time memory network and attention mechanism was proposed.The proposed method adaptively learns the fault features from vibration signals through one-dimensional convolution and bidirectional short and long time memory network,and combines the attention mechanism to enhance the recognition ability for effective fault features.Finally,the proposed method is applied to the gearbox dataset of Southeast University and the bearing dataset of Case Western Reserve University.The experiments show that the proposed method can directly carry out fault diagnosis based on the original signal,has excellent fault identification accuracy,achieves 99.91% and 99.80% fault identification accuracy respectively under the two working conditions of gear,and has stable fault diagnosis effect under different working conditions.In the mixed working condition of bearing,the fault recognition rate can reach 99.79%.(2)Aiming at the problem of poor diagnostic accuracy of the diagnosis model in the noisy environment and improving the training efficiency of model,a gearbox fault diagnosis method based on multi-scale Transformer,convolutional neural network and transfer learning was proposed.The method can obtain multiple scale signal data through multi-scale coarse-grained process which can filter random noise to a certain extent.Meanwhile,the method can perform multi-scale learning from multiple scales to get rich and complementary fault characteristics through the Transformer and the convolutional neural network.The way of transfer learning is adopted to train the model and improve the training efficiency of the model.Finally,several groups of experiments are carried out by using the gearbox dataset of Southeast University,the bearing dataset of Case Western Reserve University and the bearing data set of Paderborn University.The fault recognition rate of bearing and gearbox is over 99.50% which verified that the model has good fault diagnosis effect.Besides,the accuracy can reach 92.98% when the signal-to-noise ratio is-6 d B which verified that the method has excellent robustness in noise environment.