Combined Deep and Shallow Knowledge in a Unified Model for Diagnosis by Abduction

Fault Diagnosis in real systems usually involves human expert’s shallow knowledge (as pattern causes-effects) but also deep knowledge (as structural / functional modularization and models on behavior). The paper proposes a unified approach on diagnosis by abduction based on plausibility and relevance criteria multiple applied, in a connectionist implementation. Then, it focuses elicitation of deep knowledge on target conductive flow systems – most encountered in industry and not only, in the aim of fault diagnosis. Finally, the paper gives hints on design and building of diagnosis system by abduction, embedding deep and shallow knowledge (according to case) and performing hierarchical fault isolation, along with a case study on a hydraulic installation in a rolling mill plant.