Research On Statistical Inference Algorithm For Bearing Fault Diagnosis

Bearings are one of the important core components and sources of failure in mechanical equipment.Rolling bearing condition detection and diagnosis is very important to avoid catastrophic failure of mechanical systems.With the advent of the era of big data,the intelligent diagnosis method based on machine learning has been applied on a large scale.At present,the intelligent diagnosis method is mainly based on the discriminant model in machine learning.However,with the wider application scenarios of rolling bearings,the working conditions faced by the equipment are becoming more and more complex,and the discriminant model can not adapt well to dynamic changes.In order to solve this problem,this paper proposes a fault diagnosis model for generating bearings based on hybrid Gaussian distribution and its statistical inference method.Aiming at the problem of generating diagnostic model selection,by analyzing the distribution of fault features,a hybrid Gaussian distribution is proposed as the probabilistic diagnosis model.For the problem that the probabilistic model is complex and difficult to infer precise distribution,this paper proposes the generation based on Gibbs sampling.Diagnostic model statistical inference method,the distribution is obtained by random sampling;for the problem that the Gibbs sampling statistical inference method has a slow convergence rate,this thesis further proposes a statistical inference method based on the variational Bayesian-based generation diagnosis model,and the variational Bayesian utilization The prior information constructs the distribution,and the method further improves the convergence speed under the premise of ensuring the accuracy.For the fault data to be diagnosed,the fault diagnosis is realized by combining the generative diagnosis model and the Bayesian theorem.By using the rolling bearing open standard data set and the actual data analysis of China Railway,it is shown that the fault diagnosis accuracy rate is improved by 11.1% compared with the support model based on the support vector machine based on Gibbs sampling statistics.The probabilistic diagnostic model based on the variational Bayesian statistical inference can also achieve the ideal diagnostic accuracy rate,and the convergence rate is significantly reduced compared with the Gibbs sampling method.This topic achieves a high diagnostic rate and a fast convergence rate,which has important practical significance.This method can be applied to other fault diagnosis fields.