Low-Redundancy Realization of Safety-Critical Automotive Functions

We propose a low-redundancy architecture for safety-critical automotive functions that is motivated by the discrepancy of fail-operational requirements of autonomous driving (AD) applications, and the market’s cost pressure. We base on low-overhead monitoring structures for thermomechanical fatigue that generate data to predict the remaining useful lifetime (RUL) for individual elements. It is analysed in a cloud backend by means of a machine-learning model trained with data from accelerated aging tests and finite elements (FE) simulations. We employ model-based engineering to automatically synthesize a reconfiguration strategy that maximizes the remaining system utility by relocating software components from processing elements with impending failures. For evaluation, we consider the reconfiguration of a critical software component, and the integration of the low-redundancy monitoring concept into an electric power steering (EPS).