Industrial Alarm Root Cause Analysis Based On Bayesian Network With Cyclic Structures

The increasingly serious problem of "alarm floods" affects the normal operating order in modern process industries.As the alarm system can not respond effectively to hundreds of alarm signals in a short period of time,the operator will not be able to effectively deal with the alarm.The alarm variables of modern process industry are many,complex,and nonlinear.The variable is not only influenced by the abnormal signals,but also related to its own historical state.Therefore,it becomes more critical to establish an accurate causal topology model and an effective alarm traceability method.Bayesian Networks can express causality and infer uncertainty well.However,most of the modern process industries have introduced feedback mechanism,which produces cyclic phenomenon in both material flow and energy flow.The Bayesian Network is directed acyclic graph.If Bayesian Network is used to model the system variables,it cannot match the cyclic phenomenon in practice.To solve the problems,the paper proposed a novel multimodule Bayesian Network with cyclic structures through both knowledgedriven and data-driven methodology.By considering multi-valued case,dividing the node based on the specific conditions.The novel Bayesian Network with cyclic structures can express the causal relationship of each alarm variable at high and low alarms.Then by merging the high and low alarm state nodes of the same node,a Bayesian Network with cyclic structures can be obtained.At the same time,event reasoning is adopted to analyze alarm root causes,and a series of alarm traceability analysis procedures are proposed.TE process is as an experiment to establish the network and analyze the alarm root cause.Experimental results show that the methodology can learn the network model matching the cyclic phenomenon and correctly analyze the alarm root cause according to the traceability method to complete the alarm traceability analysis.