Integration of Fault Detection and Diagnosis in a Probabilistic Logic Framework

Abstract

In this paper we formalize an approach to detect and diagnose faults in dynamic industrial processes using a probabilistic and logic multiagent framework. We use and adapt the Dynamic Independent Choice Logic (DICL) for detection and diagnosis tasks. We specialize DICL by introducing two types of agents: the alarm processor agent, that is a logic program that provides reasoning about discrete observations, and the fault detection agent that allows the diagnostic system to reason about continuous data. In our framework we integrate artificial intelligence model-based diagnosis with fault detection and isolation, a technique used by the control systems community. The whole diagnosis task is performed in two phases: in first phase, the alarm processor agent reasons with definite symptoms and produces a subset of suspicious components. In second phase, fault detection agents analyze continuous data of suspicious components, in order to discriminate between faulty and non-faulty components. Our approach is suitable to diagnose large processes with discrete and continuous observations, nonlinear dynamics, noise and missing information.