A multi-agent model for fault diagnosis in petrochemical plants

Abstract

Petrochemical plants are extremely complex systems with many dynamically interconnected components. Traditional approaches to fault detection and diagnosis of these complex systems follow a centralized design in which huge and sophisticated models (e.g., first principle models) are constructed to process sensor data acquired from the entire plant. These systems are very difficult to design due to their complexity. Maintaining such a system to reflect any plant changes (e.g., equipment replacement), is also very challenging. In this article, we introduce a multi-agent model for fault detection and diagnosis which exploits the concept of leadership; that is, when a fault is detected one agent emerges as leader and coordinates the fault classification process. The proposed model is flexible, modular, decentralized, and portable. Our experimental results show that even using simple detection and diagnosis methods, the model can achieve comparable results to those from sophisticated centralized approaches.