Fault Diagnosis in Discrete-Event Systems with Incomplete Models: Learnability and Diagnosability

Area of Science:

• Control Systems Engineering
• Computer Science
• Artificial Intelligence

Background:

• Model-based fault diagnosis for discrete-event systems often requires complete system models.
• Incomplete models, arising from abstraction or data-driven methods, limit diagnostic accuracy.
• Previous work introduced a learning diagnoser to address incomplete models by generating hypotheses.

Purpose of the Study:

• To investigate the learnability and diagnosability properties of a learning diagnoser for discrete-event systems.
• To establish conditions under which missing model information can be learned.
• To define and analyze weak and strong diagnosability for fault detection and isolation.

Main Methods:

• Analysis of learnability: determining if missing model components can be inferred.
• Definition and analysis of diagnosability: assessing the ability to detect and isolate faults.
• Development of conditions for successful learning and fault diagnosis with incomplete models.

Main Results:

• Conditions for the learning diagnoser to effectively learn missing discrete-event system model information are provided.
• Formal definitions for weak and strong diagnosability are introduced.
• Conditions under which weak and strong diagnosability hold are established.

Conclusions:

• The learning diagnoser demonstrates potential for fault diagnosis even with incomplete system models.
• The study provides theoretical foundations for learning missing model information and ensuring fault detectability.
• This research advances the robustness of fault diagnosis in complex discrete-event systems.