Research On Fault Diagnosis Of Rolling Bearing Based On Adaptive Convolutional Neural Network

Rotating machinery,along with the development of science and technology,has become an indispensable equipment in modern industrial production.Rolling bearings,as one of its key components,often fail due to impacts,alternating loads,thermal fatigue and so on during long-term high-speed operation.It will affect seriously the stability and reliability of the equipment once a fault occurs,causing the equipment to fail to operate normally and resulting in huge economic losses and even major safety accidents.Thus,it is extremely vital to study the fault diagnosis of rolling bearings.Firstly,aiming at the noise reduction of bearing vibration signal,a noise reduction algorithm of bearing vibration signal is studied.When using variational modal decomposition(VMD)to process vibration signals,the quality of noise reduction is mainly dependent upon whether the penalty factor and the of modes of VMD are appropriate.So,the Gravity Algorithm(GSA)is applied to optimize the penalty factor and the number of modes of the variational mode decomposition.The experimental results showed that the penalty factor and the number of modes of VMD can be optimized extraordinary well and by this method is provided with a good noise reduction effect.Secondly,an adaptive convolutional neural network model(ACNN)is studied for the problem of a great many hyperparameters in convolutional neural networks.The number of layers of the network,the size of the convolution kernel,the type of pooling layer,the number of convolution kernels and neurons in the fully connected layer are the keys in the convolutional neural network to build the model.It is labor-intensive and spends a great deal of time to debug these hyperparameters.Hence,these hyperparameters of convolutional neural networks are optimized well by PSO to optimize and it is applied to diagnose the faults of bearings,which improves the efficiency of model building.Thirdly,the bearing fault diagnosis method based on GSA-VMD and ACNN is studied.It adopts the universal gravitational algorithm to optimize the variational modal decomposition,and then decomposes the vibration signal of the bearing to obtain several modal components,which are combined with the vibration signal to construct a feature matrix as the input of the adaptive the adaptive convolutional neural network model,and then identifies the type of bearing fault.The bearing data of Western Reserve University is applied to conduct experiments,the results of which showed that the method,compared with other selected methods,has higher accuracy,better stability and stronger adaptability.Finally,the bearing fault diagnosis system is developed with Python programming language.It has four functions: intelligent diagnosis function,user authority management,bearing model management and vibration signal processing,which is convenient to realize bearing fault diagnosis based on deep learning theory.