Fault Diagnosis Method Based On Dynamic Belief-Rule-Based Inference Methodology

As an important technology in the field of industrial safety,fault diagnosis/prediction methods play a huge role in ensuring the safe operation of industrial equipment.In the context of industrial big data,research in this direction has shifted from the processing of low-dimensional data to the processing of multidimensional data.Belief-Rule-Based(BRB)inference methodology is used to model various uncertain quantitative/qualitative knowledge,and it is also widely used to extract the nonlinear relationship between multi-dimensional input and output.In addition,Evidence Reasoning(ER)rule has a good effect on solving multi-attribute decision-making problems with subjective uncertainty and coexistence of quantitative and qualitative indicators.Based on BRB inference and ER rule,this paper studies the problem of fault diagnosis/prediction,the main content is as follows:(1)BRB dynamic update method based on particle filter.At present,most of the optimization methods of BRB model parameters are performed offline,which is not suitable for systems that change in real time.To solve this problem,a BRB dynamic update method based on particle filter is proposed to update the model parameters in real time.Assuming that the changing of BRB parameter satisfies the first-order Markov process,a series of weighted sample particles are used to approximate the posterior probability distribution of model parameters.Then the model parameters are updated by weighted particle samples.Finally,a typical nonlinear model case is used to verify the proposed method.(2)Multivariable fault detection method based on dynamic BRB.Aiming at the fault detection of condensate-water electrical pumps in large-scale thermal power plants,based on(1),a fault detection method for multivariable time-varying systems is presented.According to the relationship among the characteristics of the monitoring variables of the pump system,the system model based on BRB is established.Then the BRB dynamic update method based on the model estimation error is designed,and the alarm decision is made based on the updated value.Finally,the fault detection performance of the proposed method is verified by a water pump fault detection case.(3)A fault prediction method based on dynamic BRB and ER rule.In engineering practice,predicting the fault in advance can effectively reduce the loss caused by the deterioration of the fault.Aiming at the problem of fault prediction,based on the research in(1),a fault prediction method for rotating machinery is designed.For each fault characteristic variable,build its corresponding fault characteristic value multi-step prediction model based on dynamic BRB to obtain the fault characteristic prediction value.Then the fusion decision model based on ER rule is constructed,the predicted value is fused and the fault prediction is made according to the fusion results.Finally,the performance of the proposed method is verified through the failure prediction case of rotating machinery equipment.