Research On Multi-fault Diagnosis Based On Information Fusion

In the complex industrial processes,there are many causes for the system failure,such as the complication of system operation and the impact of multiple factors.Moreover,once one fault occurred,it would lead to a chain reaction easily,and then multiple faults emerge continuously.The diagnosis result from the single sensor is always vague,incomplete,uncertain,and even inaccurate since the harsh work environment.It's normal that the system provides faulty characteristic parameters and incomplete state of information in such situation.In addition,due to the limited priori knowledge and different diagnostic angles,each diagnostic algorithm has advantages and disadvantages,therefore,it is very difficult to make a comprehensive diagnosis of multi-fault based on a certain diagnosis algorithm.In order to perform the comprehensive and accurate fault diagnosis,it needs to make full use of the fault information from multi-source synthetically.Therefore,we focused on researching the fault diagnosis based on multi-level information fusion technique.The main research work is as follows:Firstly,we applied BP(Back Propagation)network and support vector machine to establish the model of rolling process fault diagnosis and obtain the preliminary diagnostic results in feature level.Aiming at the shortcoming of BP network is easy to fall into local extremum of complex data in the network learning and training process.We took advantages of mind evolutionary algorithm to optimize the original weights and thresholds,and established a more accurate fault diagnosis model.The traditional two class support vector machine is not suitable for the situation that the fault type of rolling process is complex.Therefore,the one versus one based multi-class support vector machine strategy for fault diagnosis and a posterior probability output model has been studied and established separately.What's more,the different parameters of the support vector machine affects the diagnosis results.On this basis,the grid search and cross validation based support vector machine parameter optimization method has been studied.At last,the BP neural network based preliminary diagnosis results of feature level and support vector machine are used as evidence in order to provide the basis for the decision level fusion.Secondly,we researched on the DSm(Dezert Smarandache)theory based multiple fault diagnosis.The DSm model based on the generalized identification framework and the concept of the super power set,which not only can be used to represent and deal with uncertain information,but also can deal with coupling information.Therefore,we established a multiple diagnosis model of decision level fusion based on DSm theory and the characteristics of multiple faults,and proved that the method of hybrid DSm is fit for the pure coupling multiple fault diagnosis by the rolling data and leads to an error diagnosis result.Last,regarding the problem that the hybrid DSm fusion rule does not satisfy the associative law,which makes the reliability decrease in the fusion process of the conflicts distribution,we researched on the classification and discount DSm based multiple fault diagnosis method for decision level fusion.First of all,we analysed the conflict degree of evidence,and then clustered the evidence to determine the fusion order of evidence for each category,which makes the conflict evidence combined with each other in the end and avoids the improper conflict distribution in the process of fusion.In addition,by taking into account of the individual abnormal evidence,we modified the evidence by the discount factor and proposes the based on classification and discount DSm based multiple fault diagnosis method for decision level fusion.The experimental results show that the proposed method is feasible and valid.