Solving the PMC-Based System-Level Fault Diagnosis Problem Using Hopfield Neural Networks

This paper presents a modified Hop field neural network (HopfieldNN) for solving the PMC-based system-level fault diagnosis problem of multiprocessor systems which aims at identifying the set of faulty processors. The PMC-based diagnosis model assumes that each processor is tested by a subset of the other processors, and that at most a bounded subset of these processors can permanently fail at the same time. The problem of efficiently identifying the set of faulty processors of a diagnosable system, especially when not all the testing outcomes are available to the diagnosis algorithm at the beginning of the diagnosis phase, i.e., partial syndromes, remains an outstanding research issue. The new HopfieldNN-based diagnosis algorithm does not require any prior learning or knowledge about the system or about any faulty situation, hence, providing better generalization performance. Results from a thorough simulation study demonstrate the effectiveness of the HopfieldNN-based fault diagnosis algorithm, in terms of diagnosis correctness, diagnosis latency, and diagnosis scalability, for randomly generated diagnosable systems of different sizes and under various fault scenarios.