Using Safety Contracts to Verify Design Assumptions During Runtime

Abstract

A safety case comprises evidence and argument justifying how each item of evidence supports claims about safety assurance. Supporting claims by untrustworthy or inappropriate evidence can lead to a false assurance regarding the safe performance of a system. Having sufficient confidence in safety evidence is essential to avoid any unanticipated surprise during operational phase. Sometimes, however, it is impractical to wait for high quality evidence from a system’s operational life, where developers have no choice but to rely on evidence with some uncertainty (e.g., using a generic failure rate measure from a handbook to support a claim about the reliability of a component). Runtime monitoring can reveal insightful information, which can help to verify whether the preliminary confidence was over- or underestimated. In this paper, we propose a technique which uses runtime monitoring in a novel way to detect the divergence between the failure rates (which were used in the safety analyses) and the observed failure rates in the operational life. The technique utilises safety contracts to provide prescriptive data for what should be monitored, and what parts of the safety argument should be revisited to maintain system safety when a divergence is detected. We demonstrate the technique in the context of Automated Guided Vehicles (AGVs).