Intermittent Fault Diagnosability and Fault Diagnosis Algorithm for Irregular Diagnosable Network Under the MM Model

To address the significant challenges in diagnosing intermittent faults in multiprocessor systems, which have randomness, intermittency, nonrepeatability, high impact, frequency, and concealment, this study delves into an underexplored area. While most research efforts have concentrated on fault diagnosis in regular diagnosable networks, irregular diagnosable networks have received less attention. Recognizing the wide application potential of irregular graph networks, this article presents a novel approach to diagnose intermittent faulty nodes within such networks. We propose a probabilistic self-diagnosis method tailored for irregular diagnosable networks under the comparison model, employing a system-level diagnostic framework to bolster the reliability of multiprocessor systems. First, we establish the intermittent fault diagnosability t^MM*_I (G_IR) of an irregular diagnosable network G_IR as t − 1, where t represents the network’s minimum degree. This determination is based on the indistinguishability of constructed sets and the application of linear multiple faults analysis. Furthermore, we develop a widely applicable probabilistic intermittent fault diagnosis algorithm, named PIFDIRGMM*, designed to identify faults in irregular diagnosable networks with lower time complexity. This algorithm serves as a versatile framework that can be adapted for other network types. To validate our approach, we assess the performance of the PIFDIRGMM* algorithm using two real-world datasets WSN-DS and TON_IoT, and compare it against current mainstream algorithms. The experimental results indicate that as the execution stages of PIFDIRGMM* progress, the accuracy, recall, and F1 achieve minimum levels of 99.92%, 98.16%, and 99.07%, respectively. The false negative rate peaks at 1.94%, while the false positive rate remains consistently at 0. In comparison to leading methods, our scheme improves accuracy from 0.7% to 1.95% and reduces the false alarm rate by 1.7% to 3.2%, demonstrating its effectiveness and superiority.