Probabilistic approaches to fault detection in networked discrete event systems

Area of Science:

• Computer Science
• Distributed Systems
• Fault Diagnosis

Background:

• Distributed systems can experience permanent faults altering their behavior.
• Faults in communication networks, especially during protocol execution, pose significant challenges.
• Existing fault detection methods may struggle with limited state observability.

Purpose of the Study:

• To develop a fault detection mechanism for distributed systems modeled as finite state machines.
• To address faults that cause permanent changes in system transitions.
• To minimize the probability of incorrect fault-free declarations (aliasing).

Main Methods:

• Utilizing an external observer/diagnoser modeled as a finite state machine.
• Employing a probabilistic framework to analyze synchronizing conditions.
• Developing adaptive strategies for optimal observer-system synchronization.
• Leveraging partial state information at intermittent intervals.

Main Results:

• Optimal selection of synchronizing conditions reduces aliasing probability.
• Adaptive strategies effectively detect permanent faults in system transitions.
• The approach demonstrates a low probability of misclassifying faulty systems.

Conclusions:

• The proposed method provides a robust approach to fault diagnosis in distributed systems.
• Effective fault detection is achievable even with limited and intermittent state access.
• The framework is applicable to practical scenarios like communication protocol testing.