Fault Diagnosis Of Chemical Process Using Mechanism Correlation Based Bayesian Network

With the increasing integration and automation of chemical process,the relationship between variables becomes more and more complicated.Once a failure occurs in the production process,the chain effect between variables will cause more serious consequences than before.Moreover,the measured variables in the process usually cannot represent the entire process.If there is a failure of unobserved variables,there will be undetectable problem or serious lag results after the failure,which will have a serious impact on the chemical plant.Therefore,in the chemical production process,how to effectively monitor the process,detect the occurrence of failures in time,and determine the true root cause of the failures are the keys to ensure the smooth operation and the stable product quality of the chemical process.This paper proposes a mechanism correlation dependent Bayesian network method for fault diagnosis of chemical processes.Frist,based on mutual information feature engineering,fault detection and identification are performed on the process.A subset of features most relevant to the fault state is constructed to monitor the state change of the process and find the variables that contribute mostly to the fault state.The kernel extrem learning machine network is constructed to identify the fault type after fault occurs.The performance of the proposed method is tested by the missed diagnosis rate,misdiagnosis rate and fault identification accuracy.Then the mechanism correlation based Bayesian network is constructed to diagnose the chemical process fault.The internal mechanism of chemical process variables was analyzed to find the causal relationship between variables,and the causal relationship between variables is further determined by the correlation coefficient to obtain the structure of Bayesian network.Then Bayesian interval estimation is used to learn the parameters of the network and a priori causal relationship is combined to build Bayesian Network.After the fault is detected,the Bayesian network is used to explain the fault propagation path and diagnose the root cause.For an unobserved variable or a broken-chain fault propagation path,it is represented as an observed variable by adding virtual nodes.This method can effectively perform fault detection,show the mechanism relationship between variables,determine the root cause of the fault,and finally achieve the purpose of fault diagnosis.The method proposed in this paper is applied to the TE process and an industrial ammonia synthesis case.The results show that the method can effectively reduce the rate of missed diagnosis while ensuring a small misdiagnosis rate.And when the fault occurs,the achievement of propagation path aroung root cause benefit the purpose of fault diagnosis.