| As a key component of mechanical system,the status of gear box plays a very important role in the normal operation of mechanical equipment,so it is of practical significance to carry out fault diagnosis and identification analysis for gear box.In this thesis,aiming at the problem that it is difficult to distinguish the fault diagnosis of rolling bearing and gear in gearbox,based on deep learning technology and combining with traditional signal processing method,a multi-channel neural network model is built to realize the intelligent diagnosis of compound fault of gearbox.The main contents of the thesis are as follows:Firstly,the thesis introduces the common fault types and vibration mechanism of rolling bearings and gears in gear boxes.And the use of relevant experimental equipment under the fault of the parts for signal acquisition,a more detailed description of the experimental operation required flow.Secondly,the feature extraction capability of multi-channel convolutional neural network is studied.First,the Ensemble Empirical Mode Decomposition algorithm was combined with MC-CNN to achieve feature extraction of gearbox signals through the integration strategy.Secondly,a variety of time-frequency analysis methods are used as data preprocessing,combined with MC-CNN,and the proposed pyramid splitting attention mechanism is used for deeper feature mining.The experimental results show that the proposed two methods can extract signal features efficiently and have good robustness and generalization.Thirdly,the diagnosis method of multi-sensor and multi-channel data fusion is studied.The multi-sensor vibration signals were preprocessed,and the multi-channel two-dimensional feature surface set was constructed as the input of the diagnostic network model.CNN network is used to realize adaptive feature extraction of multichannel feature surface.The results show that this method has a high ability of fault classification.Finally,a Generative method based on Conditional Generative Adversarial Networks is proposed to solve the problem of small or imbalanced fault data sets.The model uses the feature extraction ability of neural network to generate data consistent with real sample distribution through generator,and uses relevant means to detect the quality of generated data and real data.The expanded data is taken as the original input set of the multi-channel data fusion diagnostic model,and the feature classification ability and CGAN data generation ability of the diagnostic model are verified through experimental analysis. |
