Machine Learning-Based Fault Injection for Hazard Analysis and Risk Assessment

Abstract

Current automotive standards such as ISO 26262 require Hazard Analysis and Risk Assessment (HARA) on possible hazards and consequences of safety-critical components. This work attempts to ease this labour-intensive process by using machine learning-based fault injection to discover representative hazardous situations. Using a Simulation-Aided Hazard Analysis and Risk Assessment (SAHARA) methodology, a visualisation and suggested hazard classification is then presented for the safety engineer. We demonstrate this SAHARA methodology using machine learning-based fault injection on a safety-critical use case of an adaptive cruise control system, to show that our approach can discover, visualise, and classify hazardous situations in a (semi-)automated manner in around twenty minutes.