Bearing Fault Diagnosis Based On Empirical Wavelet Transform And Improved Deep Belief Network

The development of science and technology promotes the continuous progress of industrial technology,and mechanical equipment as one of the main parts of industrial process,its equipment status is related to the stable operation of the whole system.In order to ensure the normal operation of the system,it's necessary to diagnose the fault of bearing which is one of the important parts of mechanical equipment.Deep learning is a new machine learning method,and it has better performance in terms of signal and data processing,therefore,this paper studies the deep belief network model and combines it with feature extraction method to realize the fault diagnosis of bearing.To this end,this paper focuses on the following aspects of work.First of all,the research status of deep learning in the field of fault diagnosis is described in this paper.Aiming at the research object,the existing problems are analyzed,and then the significance of applying deep learning algorithm to fault diagnosis is expounded.In addition,this paper describes the structural composition and the fault types of bearing,and introduces some basic theory of the deep belief network.Besides,the influence of the parameters of deep belief network on the performance of fault diagnosis model is studied.Then,based on the research results,a diagnosis model based on deep belief network is constructed to diagnose the bearing fault,and this paper analyzes the diagnosis results in order to point out the shortcomings and improvement ideas.Bearing vibration signal is a typical non-stationary signal,and its fault data is easily affected by its working environment.Using the advantage of empirical wavelet transform in feature extraction of non-stationary signal,this method is used to decompose the original bearing vibration signal and extract useful signal components for its classification.And the signal extracted by empirical wavelet transform is input into the basic deep belief network model,then the bearing fault recognition is realized,therefore,the fault diagnosis effect of the method is compared and analyzed.Deep belief network has good classification performance in the field of fault diagnosis,but it is difficult to select parameters of deep belief network adaptively according to the research object.Therefore,it takes a lot of time to set parameters through some experiments,or we need to refer to the research results of others,and initialize parameters according to this empirical knowledge.However,these methods have certain limitations,and the operation process is complicated,so it does not make use of the actual application.Therefore,genetic algorithm is used to optimize the parameters of deep belief network for this issue,and then an improved fault diagnosis model is constructed by the optimization results.The feature extraction method is integrated with the improved deep belief network parameter model in order to establish a bearing fault diagnosis model which is based on EWT-GA-DBN,and the fault diagnosis performance of this model is verified by bearing fault data sets of the Western Reserve University and DC College.Finally,the fault classification performance and fault classification efficiency of this method are mainly analyzed in order to prove the validity and practicability of the fault diagnosis model.