Research On Rolling Bearing Fault Diagnosis Based On Improved Bayesian Network

Motor bearings,as the core parts of rotating machinery,play a pivotal role in ensuring the safe operation of machinery and equipment.Therefore,the fault diagnosis technology in the bearing field is of vital importance to the working characteristics of mechanical equipment and the accuracy of fault diagnosis.Bearings are subjected to strong vibration and shock,and the resulting oscillation information is non-uniform,nonlinear,low signal-to-noise ratio and multi-component,which makes it difficult to extract fault characteristics from the original vibration signal for bearing fault identification and diagnosis.In order to improve the accuracy of bearing fault identification,a fault diagnosis method based on fully integrated empirical modal decomposition(CEEMDAN)energy entropy and improved Bayesian network is proposed.Firstly,according to the motor bearing is vulnerable to the interference of its surrounding environment noise due to its working operation,the vibration signal generated has the characteristics of nonlinearity and nonstationarity,and the original vibration signal of the bearing is processed by the method of wavelet thresholding to achieve the purpose of noise reduction,and the feature components are selected by CEEMDAN decomposition,combined with the correlation coefficient and variance contribution rate and their energy entropy is calculated according to the energy of each order IMF component,and the The energy entropy is used as the intrinsic modal function to obtain a series of characteristic signal characteristic time scales,and the signal nonlinearity and non-smoothness characteristics in each order IMF component are extracted and analyzed as the feature vector;the node variables of the fault diagnosis model based on the improved Bayesian network are determined.Then,the Bayesian network and the sparrow search algorithm are described.To optimize the Bayesian network structure,a hybrid learning method of constraint and fractional learning is used,with the PC algorithm as constraint learning,and the sparrow search algorithm with better various characteristics is applied as fractional learning for the optimization of the Bayesian network structure.By analyzing the limitations of the sparrow search algorithm,and analyzing the problem that the position and update strategy of the sparrow search algorithm in the process of finding the best are easily influenced by the poor sparrow individuals to enter the local optimum,a random wandering strategy based on the genetic algorithm is proposed to improve the sparrow search algorithm.The genetic algorithm is used to select sparrow individuals probabilistically,use the uniform crossover principle and UMAD mutation factor for crossover mutation to improve the biological diversity of sparrow population;and the random walk strategy is used to expand the exploration scope of sparrow individuals to strengthen the global search ability of the sparrow search algorithm in order to prevent entering into the local optimal solution.The results show that the improved sparrow search algorithm has better global search capability and faster convergence speed.And the optimized sparrow search algorithm and PC algorithm are jointly optimized for Bayesian network structure,and tested on three classical Bayesian networks,showing the effectiveness of the sparrow search algorithm and PC algorithm for Bayesian network structure optimization.Finally,the initial network for bearing fault diagnosis is generated by optimizing the Bayesian network with PC algorithm,and the improved sparrow search algorithm uses the Bayesian information criterion as the fitness function and the energy entropy eigenvalues of the selected components as the input,and the divided training set is input to the network to construct the model and derive the optimal network structure.The rolling bearing fault diagnosis model based on Bayesian network is constructed according to PC-GASSA,thus realizing the fault diagnosis of rolling bearings.There are 48 figures,21 tables,and 75 references.