(70v) A Multilevel Safety EARLY Warning Method Based on Bayesian Networks with Applications for Petrochemical Process

Process systems in petrochemical industries have becoming increasingly large and automatic. There exist strong interdependencies between various facilities and components. Once any unit in the system fails, it often triggers cascade failures as a chain reaction, resulting in significant production losses, and even leads to catastrophic accidents. They are the result of a number of failures at different technical and organizational levels.

Proactive methodologies for the development of early warning methods can unveil early deviations in the causal chain. Two main issues exist that are how to describe the interdependency in the complex process system, and how to ratiocinate and demonstrate the reasons at each level of the current abnormal event and also foresee the possible consequence.  In this way, a multilevel safety early warning method is proposed for petrochemical process to overcome above problems. The approach combines the advantages of multilevel flow modeling (MFM), HAZOP analysis and dynamic Bayesian network (DBN) as well, which become a special and effective mechanism of safety early warning.

Firstly MFM is used to represent a system in terms of goals, objectives, functions and components each of which can be described at different levels of part-whole decomposition. Secondly HAZOP study is carried out based on the MFM models, by which all the possible deviations and their corresponding potential fault causes and consequences are analyzed carefully. Thirdly dynamic Bayesian Network (DBN) is used to build the fault causal relationships which represent the fault interdependencies in the complex process system. Finally by the inference mechanism of DBN, the most possible initial reason(s) happened in the fault interdependency network with multi reasons and consequences can be found out accurately and also the future possible consequences can be predict for proactive maintenance or emergency decision making.

The analysis shows that the proposed model fits its purpose enabling causal reasoning that explains causes and consequences derived from deviations. Examples from case studies with respect to the Fluidized Catalytic Cracking unit (FCCU) in petrochemical process system are included to illustrate the effectiveness and accuracy of the proposed approach. Based on the proposed multilevel safety early warning method, an expert system is developed and applied to practice in the petroleum industry. Its application is also illustrated to show the effect of accident prevention in the petrochemical enterprises. The salient features of this expert system and its performance on practical petrochemical systems is presented.