Uncertainty based fault type identification for fault knowledge base generation in system of systems

Abstract

A system of systems (SoS) is a large-scale complex system composed of constituent systems (CSs) that interact organically to achieve the goals of the SoS, which cannot be achieved by individual CSs. Each CS can have operational and management independence, and interactions between CSs can result in various emergent behaviors. Due to these characteristics, various uncertainties may arise in SoS in addition to the uncertainties of each CS. Uncertainty in SoS is manifested in the form of faults, and new unobserved faults can exist in the fault knowledge of each CS. To debug these faults efficiently, a high-quality fault knowledge base must be created for the SoS. A high-quality fault knowledge base should be able to express as many different types of faults as possible. However, existing studies are conducted under the assumption that there is sufficient data when creating an initial fault knowledge base or cannot consider various faults related to the characteristics of the SoS. These studies created a fault knowledge base using only a limited amount of collected fault data and were highly dependent on the quality of the fault data. Unfortunately, when a generated fault knowledge base is dependent on the current fault data, it becomes very vulnerable to fault types that are not included in these data. Therefore, this study proposes an approach to create a fault knowledge base in SoS considering uncertainty, determines the quality of fault data, and accordingly creates a high-quality fault knowledge base. The proposed approach categorizes faults based on the nature of uncertainty and the manifestation location, making it possible to easily find and add fault types that are not considered currently. Through a case study of an advanced driver assistance system (ADAS), a type of SoS, we followed the method by which domain experts create a fault knowledge base. Extracting and classifying knowledge from more than 9,000 fault data entries revealed that only 7 out of 10 fault types were observed. In addition, an analysis of fault classification for an autonomous vehicle test scenario in EuroNCAP revealed fault types that were not considered.