Research On Fault Diagnosis Method Of Rolling Bearing Based On Deep Learning

As an important part of load bearing and load transfer,rolling bearing is not only the basic key part of all kinds of rotating machinery equipment,but also one of the components most prone to failure.The overall operation effect of the equipment will be affected by the working state of the rolling bearing.Therefore,the co ndition monitoring and management of the health status of rolling bearings is of great significance in improving the operation efficiency of equipment,ensuring personnel safety and reducing the economic loss caused by maintenance costs.At the same time,with the development of high-performance data acquisition devices and signal monitoring technology,the order of magnitude of real-time condition monitoring data of mechanical equipment has been significantly improved.And the vibration signals collected by high-performance data acquisition devices mostly contain a certain degree of noise interference,and the characteristics of nonlinearity and non-stationarity are obvious,which increases the difficulty of traditional methods to extract fault features.In the rolling bearing health monitoring and management system driven by big data,deep learning algorithm is widely used because of its strong learning ability and feature self-extraction ability.This paper takes the rolling bearing as the equipment maint enance object,based on the case Western Reserve University CWRU rolling bearing vibration data set,through the combination of example verification and theoretical analysis,the rolling bearing health monitoring management method based on variational mode decomposition and improved depth belief network and the rolling bearing health monitoring management method based on one-dimensional convolution neural network are studied.The specific research contents of this paper are as follows:1)the condition monitoring and management method of rolling bearing based on vibration signal analysis is studied,starting with the common types of health degradation of rolling bearing,the structure of rolling bearing,the vibration mechanism in the process of health degradation and the characteristics of vibration signal.And common signal analysis methods are described in detail.2)A noise reduction method of rolling bearing vibration signal based on variational mode decomposition is proposed in this paper.The correla tion coefficient is used to characterize the correlation degree between the intrinsic mode function and the original vibration signal,and the real correlation component is selected for reconstruction.The experimental results show that the reconstructed s ignal can effectively filter out the noise while retaining the main characteristic information of the original signal.3)the input type of the restricted Boltzmann machine in the traditional deep belief network is defined as a binary variable,which is p oor for the reconstruction and fitting of the real data collected by each sensor in the industrial process.Therefore,in this paper,the Gaussian constrained Boltzmann machine is introduced to increase the ability of the model to represent the probability distribution in the real number domain.4)with the increase of network parameters,the model becomes more and more complex.This leads to the problems of long training time and over-fitting of the model in the process of reverse fine-tuning of deep belief network.Therefore,this paper proposes an optimization method which uses Adam algorithm to adaptively adjust the learning rate to accelerate the convergence speed of the model,and Dropout to regularize the model to improve the generalization ability o f the model.5)A fault diagnosis method of compound rolling bearing based on variational mode decomposition and improved depth belief network is proposed.The frequency domain data of the signal after noise reduction and reconstruction are used as the input vector of the improved depth belief network to diagnose the fault type,and the effectiveness of the method is verified by an example.The results show that this method can effectively improve the accuracy of fault diagnosis and accelerate the convergence speed of the model.6)because the two-dimensional structure of convolution neural network does not match the one-dimensional characteristic of rolling bearing vibration signal,the structure of convolution kernel and pool kernel is changed to one-dimensional.An one-dimensional convolution neural network suitable for vibration signal feature extraction is proposed,so that it can input one-dimensional vibration signal directly.Simulation results show that the 1DCNN fault diagnosis model can adaptivel y extract features and achieve high-precision identification of rolling bearing fault types.