Research On Intelligent Diagnosis Of Rotating Machinery Fault Based On Deep Learning

Rolling bearings and gears are key components of equipment manufacturing and important rotating parts,they are called mechanical joints.The normal operation status directly affects the safety of personnel and economic benefits.Therefore,finding an effective fault diagnosis method is important particularly.The paper mainly uses the deep learning Sparse Autoencoder model and wavelet transform to diagnose the fault of the gearbox.Combined with the stochastic resonance diagnosis of bearing faults,two new intelligent diagnosis methods are proposed.At the same time,the stochastic resonance model is improved,making the extracted fault features more obvious.The effectiveness of the newly proposed algorithm is verified by examples of gear and roller bearing fault diagnosis.Firstly,deep learning Sparse Autoencoder model is studied and the structure of Sparse Autoencoder model is studied particularly.The gradient descent algorithm,noise coefficient Dropout are studied and analyzed.The activation function sigmoid are also studied.The method is studied,which provides theoretical guidance for the experimental phenomena that appear below,and helps to understand the deep learning Sparse Autoencoder model from the aspect of algorithm.Secondly,for the traditional fault diagnosis method of gears,there is a problem of low accuracy rate of manual extraction features.A fault diagnosis method that combines automatic extraction features of deep learning and manual extraction features is proposed.Firstly,the wavelet transform Mallat algorithm is used to perform five-level decomposition of the vibration time-domain signal of the gearbox,and the time-domain statistical characteristics of each signal are extracted manually.The time-domain statistical characteristics are input into the trained noise reduction self-encoding model to further extract features,and the model.The data of the last hidden layer is used as the input of the K nearest neighbor algorithm for diagnosis and identification.Finally,the diagnosis example of multi-stage gear transmission system test bench proves that this method can effectively improve the fault diagnosis accuracy of gearbox.Then,the stochastic resonance model is improved,and a weak fault signal enhancement detection method based on novel nonlinear coupled bi-stability stochastic resonance is proposed.Firstly,the stochastic resonance theory is studied.Then the characteristics of the model are obtained by observing its characteristic curve.Using the obtained model characteristics,the time domain diagram and power spectrum diagram of the bearing fault signal and the time domain diagram after the coupled bistable system are processed.Comparing with the power spectrum,and it was proved that this model can effectively extract the weak fault signal of the bearing.Finally,a rolling bearing fault diagnosis method based on an improved nonlinear coupled bistable stochastic resonance model and Sparse Autoencoder model is proposed.Firstly,the characteristics of weak fault signals of bearings are extracted by modified nonlinear coupled bi-stability stochastic resonance method.Then,the frequency spectrum of the fault features extracted is input into deep learning Sparse Autoencoder for classification and recognition.Finally,through the example of bearing experimental data diagnosis of Western Reserve University,it is proved that this method can effectively improve the bearing fault diagnosis accuracy.