An Efficient Method to Address Dynamic Component Irrelevance in Coverage Models

SUMMARY & CONCLUSIONSFault tolerance is an essential feature in many critical applications of digital systems to ensure uninterrupted operation even in the presence of some faults. Automatic fault recovery mechanisms (fault detection, location, isolation, and correction) play a crucial role in implementing fault tolerance because an uncovered fault may propagate and contaminate other non-faulty components, which in turn leads to a system or subsystem failure even when adequate redundancy exists. An accurate reliability analysis of these systems is critical for effective implementation of fault tolerance, and it must account for system failure logic and the effectiveness of recovery mechanism expressed as coverage factor. When the coverage factors of components are independent, the Simple and Efficient Algorithm (SEA) based on a separable method provides an accurate and efficient reliability analysis, and these models are called Traditional Coverage Models (TCM).Recent publications on fault tolerant systems have been focusing on a new class of coverage model called Irrelevancy Coverage Model (ICM) that addresses an important issue that has been overlooked in the Traditional Coverage Models (TCM) of redundant structures where an initially relevant component could become irrelevant (to system success or failure) after failure of a set of other components. Once a component becomes irrelevant, any uncovered failure of the component has no impact on the system. This introduces complex stochastic dependencies because coverage factors of components depend on the states of other components. Existing methods to solve ICM are too complex and involve operations that demand significant computational times to address underlying temporal and non-monotonic (non-coherent) failure logic. As a result, these methods are limited to small-scale systems and often approximations that overestimate the system’s reliability. In this paper, we propose a preprocessing step and an associated modularization method to create an equivalent reduced model that can be solved easily. In most cases, the reduced model can be solved efficiently using SEA. In case it cannot be solved by SEA, we provide an efficient solution to solve the reduced model. We demonstrate the effectiveness of the proposed solution using published benchmark examples originating from well-known safety-critical applications. We also provide an exact solution to a published benchmark that was considered too complex to solve accurately in a recent article.