Diagnosis

Diagnosis is a logical process consisting of two different phases: first, the inference mechanism must recognize if the current observed behaviour of the system is actually the correct response to environmental conditions and to internal commands, or if, instead, it is affected by some anomalies due to failures occurred within a system component. Then, once a failure state has been detected, the inference mechanism must identify the failure (or the set of failures, in case of multiple faults scenarios) which is actually affecting the system. The natural logical approach to infer the potential explanations for a given set of manifestations is abduction; an abductive failure identification algorithm is an Expert System which requires a reliable Knowledge Base (KB).

The combination of Model Based techniques for Autonomous Failures Detection and Abductive Reasoning for their Identification represents the hybrid approach to diagnosis task actually used in our group. Hybrid diagnosis couples the main advantages of both approaches and exploits each of them only for the task which is most appropriate of it. In particular, research on methodologies for a reliable and robust autonomous diagnosis is based on the following approaches:

• Failure Detection – Fuzzy Inductive Reasoning (FIR) coupled with Envelope Technique are used for creating a dynamical model of the system under nominal conditions and for producing its forecasted behaviour with a degree of acceptability;
• Failure Identification – Possibilistic Abductive algorithm exploiting Zadeh’s Possibility Logic as uncertainty handling formalism.

Our research activities are focused on:

• improvement of FIR performances by means of NeuroFIR algorithms, on-line adaptation and reconfiguration, automatic learning algorithms applied to off-line system modelling;
• development of FIR real-time algorithms and tools for off-line qualitative modelling;
• methodologies for automatic KB acquisition and/or automatic KB tuning;
• KB representation and methodologies for KB reliability and completeness improvement;
• development of tools for off-line KB management as support for system design process;
• improvement of abductive reasoning performances by means of contextualization of events and management of temporal information about manifestations, management of multiple failures, reduction of potential solutions research space;
• development of algorithms and tools for real-time diagnosis simulation.