A hybrid framework for large scale process fault diagnosis

Abstract

Real-time process fault diagnosis deals with the timely detection and diagnosis of abnormal process conditions. Industrial statistics estimate the economic impact due to abnormal situations to be about 20 billion dollars per year in the petrochemical industries alone in the U.S. Thus, it is an important part of safe and optimal operation of chemical plants. A promising alternative approach is proposed in this paper, that of a hybrid, blackboard-based framework, called DKit. The motivation for development of hybrid framework lies in the fact that no single diagnostic method satisfies all the requirements of complex, industrial-scale diagnostic problems. A hybrid framework in which different diagnostic methods perform collective problem solving shows a lot of promise. The current version of DKit, implemented in G2, combines causal model-based diagnosis with statistical classifiers and syntactic pattem recognition. The salient features of this system and its performance on an simulated Amoco FCCU is presented.