Spectrum-Based Fault Localization on a Collaboration Graph of a System-of-Systems

Abstract

A System-of-Systems (SoS) consists of independent and autonomous constituent systems (CSs) which collaborate to achieve an SoS goal. For SoS engineers, it is important to verify the results of the collaboration for an SoS goal. Statistical verification can be used to verify a large and complex SoS and to provide quantitative verification results. However, even when a failure of an SoS goal or a violation of a verification property is detected, it often requires a huge cost to find faults of an SoS because of the size of the SoS and the lack of information about independent CSs. In this paper, we propose a fault localization technique for an SoS to reduce the debugging cost by prioritizing suspicious entities of an SoS. This technique requires only an abstract model of the collaboration, named as a collaboration graph, which includes the presence or absence of CSs and their interactions. It extends a spectrum-based fault localization (SBFL) technique to utilize quantitative results of the statistical verification of an SoS. In the evaluation, we show that it is feasible to apply SBFL to SoS fault localization, and our approach is expected to effectively reduce the debugging space of an SoS.