Optimized fault diagnosis based on FMEA-style CBR and BN for embedded software system

Fault diagnosis is an important step for software-intensive manufacturing system and process. But because of the increased scale and complexity, as well as the uncertain condition of the running environment, embedded software fault diagnosis is still an open challenge for industry application. In this research, an optimized hybrid model is proposed by integrating a case-based reasoning (CBR) method and a BN-based diagnosis method. Initially, a FMEA-style case-based reasoning (F-CBR) method is proposed by collecting and formalizing the existed fault cases for the low-level similarity searching guided diagnosis. Then, by adopting a new designed algorithm, F-CBR can be further transferred to a deep-level Bayesian diagnosis network for the dynamic multi-fault diagnosis with uncertainty. Based on this framework, we implement a prototype of hybrid expert system for the diagnosis of embedded software by integrating CBR with Bayesian network (BN) through F-CBR by the corresponding failure spectra as the bridge. The feasibility and benefits of this hybrid diagnosis strategy are verified by examples and case studies in real industry applications with great promising results for different kinds of multi-level diagnosis.