An integrated neural network/expert system approach for fault diagnosis

Abstract

The main thrust of this research is the development of an artificially intelligent system to be used as an operators' aid in the diagnosis of faults in large-scale chemical process plants. The operator advisory system involves the integration of two fundamentally different artificial intelligence techniques: expert systems and neural networks. A diagnostic strategy based on the hierarchical use of neural networks is used as a first-level filter to diagnose faults commonly encountered in chemical process plants. Once the faults are localized within the process by the neural networks, the deep knowledge expert system analyzes the results, and either confirms the diagnosis or else offers an alternative solution. The model-based expert system contains information of the plant's structure and function within its object-oriented knowledge base. The diagnostic strategy can handle multiple faults, novel or previously unencountered faults, and noisy process sensory measurements. The operator advisory system is demonstrated using a multi-column distillation plant as a case study.