Hybrid system for fault diagnosis using scanned input: A tutorial

It is well known that expert systems are useful in capturing expertise and applying knowledge to chemical engineering problems such as diagnosis, process control, process simulation, and process analysis. Traditionally, expert system applications are limited to knowledge domains that are heuristic and involve only simple mathematics. Neural networks, however, represent an emerging technology capable of rapid recognition of patterned behavior without regard to mathematical complexity. Although useful in problem identification, neutral networks are not very efficient in pointing to in-depth solutions and typically do not promote a profound understanding of the problem or the reasoning behind its solutions. This paper explores the potential for expanding the scope of expert system applications by combining expert systems with neural networks. Any computer system that is partly an expert system and partly a neural network can be called a hybrid system. This pairing is a natural one because where one system falls short the other excels. Imprecise (or even incomplete) data can submitted to a neutral network for classification. Once these data are classified, the results are not always as precise as need dictates. It is at this point that the expert system can be invoked to do what it does best: takemore » definite and complete, but general, input and produce a definite and precise output. This paper presents a relatively new approach--one that expands the scope of artificial intelligence (AI) applications by combining expert systems and neural networks to form a hybrid system. A general methodology for developing hybrid systems is given. The methodology and merits of developing hybrid systems are illustrated through a case study.« less