Research On Fault Diagnosis Of Valve Based On Bayesian Network

With the improvement of science and technology,more and more attention has been paid to the technology of mechanical fault diagnosis.Reciprocating compressor,as a typical reciprocating machine,has a complex internal structure and a large number of excitation.The traditional fault diagnosis technology can not meet the needs of actual Engineering.Bayesian network(BN)has unique advantages in encoding and reasoning uncertain knowledge,which has been used in speech recognition,image processing,financial analysis and other fields.Therefore,a fault diagnosis method based on BN is proposed.The method is used to study the common faults of the compressor valve.Based on the research of BN theory,the paper establishes some structure learning methods of BN,which provides strong evidence for its application in fault diagnosis.At the end of this paper,two kinds of BN classification models are constructed and applied to the fault diagnosis of compressor valve.The main works can be summarized as follows:1.This paper expounds the basic theory of BN and introduces the learning algorithm of the BN and the four kinds of common Bayesian classifiers.2.According to the acceleration signal of the compressor valve,the wavelet threshold denoising method is performed on the original signal,and the fault feature vectors are extracted by the wavelet packet algorithm.The sample which is composed of the feature vector and the class variable is discretized and used as the input of the Bayesian classifier.3.In this paper,the algorithm of BAN classification is proposed for the common faults of the compressor valve.Firstly,the genetic algorithm and K2 algorithm are used to construct the network structure of the attribute nodes which is evaluated by the K2 scoring function.The parent nodes(class nodes)of these nodes are added to construct the classification model and the Bayesian estimation algorithm is used to obtain the conditional probability table of each node.According to the test sample set,the posterior probability of each test sample can be obtained and the class label corresponding to the maximum posterior probability is the classification result of the sample.4.In this paper,a general Bayesian classification algorithm(GBNC)is proposedfor the common faults of compressor valve.The algorithm firstly combines the conditional independence algorithm and the greedy algorithm.In the method,the variables which are independent of the current node variable are removed and the initial candidate parent nodes can be reduced by the conditional independence test.The greedy algorithm is used to update the candidate parent nodes of each node,and the optimal substructure of each node is finally combined into the optimal structure of BN.In this paper,the method of sparse score is used to select the fault features.5.The conclusion of the whole thesis and the suggestion of further research direction is proposed.The different sets of features are extracted,and the GBN classifier is used to classify the test samples.The experimental results show that the method can improve the accuracy of fault diagnosis and reduce the computational complexity.