A multi-agent approach to a networked fault detection system

Abstract

Nowadays, fault diagnosis systems design should have into account the distribution and complexity of the process and must be able to cooperate and communicate with other systems to achieve satisfactory performance. To this end a new agent based fault detection system for networked control process is proposed in this paper. A hybrid architecture based on horizontal layers as fault detection (FD) agents architecture is adopted. The reactive layer of the FD agents is based on neural networks models for residuals generation with adaptive threshold. The deliberative layers take into account the agent knowledge and goals, and also the process distribution and network. The new agent based FD system has been applied and tested with good results in a three tank process using a star Ethernet network topology.