A Novel Diagnostic Strategy Based on Closed Neighborhood Fault Pattern
Qifan Zhang, Shuming Zhou, Sun‐Yuan Hsieh, Weixing Zheng · IEEE Transactions on Network Science and Engineering · 2025
Fault diagnosis in multiprocessor systems is crucial for ensuring reliable and efficient operation in various applications, including industrial automation, Big Data processing, and high-performance computing. Both internal and external factors have the potential to inflict different levels of damage on system units, with the worst scenario being closed neighborhood invalidation–where the failure of a single unit results in the simultaneous failure of all neighboring units. Robust diagnostic strategies paired with effective diagnostic algorithms enable the early detection and isolation of faults, which will minimize downtime and prevent cascading failures. However, existing diagnostic strategies, which assume independent vertex failure, are ineffective in measuring the multiprocessor system's diagnostic capability in the presence of closed neighborhood invalidation. In response, we introduce a novel diagnostic strategy termed neighbor diagnosability, which is characterized as the maximum number of faulty closed neighborhood structures that can be diagnosed within the system. This strategy is then implemented in hypercubes to determine the neighbor diagnosability under both PMC and MM$^*$models. Furthermore, we suggest two diagnostic algorithms,ND-PMCandND-MM$^*$, tailored to these two diagnostic models. Finally, the experimental simulations are performed under two closed neighborhood failure modes: random failure mode and neighbor failure mode. The experimental results demonstrate that the accuracy of diagnosis exceeds 99.999% by these two algorithms.