Low-redundancy implementation of safety-critical functions in autonomous electric vehicles
The typical approach for designing distributed real-time systems with a high degree of safety in order to guarantee the error-free execution of mission-critical functions, even in case of data or component outages, is the introduction of massive system redundancy, meaning the use of two or three times the number of critical system components. This approach is unsuitable for cost-sensitive markets such as vehicle manufacturing however. Despite an enormous number of costly electronic systems that require extensive space and energy and take up weight, in case of subcomponent failure the concrete remaining runtime, and thus the risk, still remains unknown. Against this background, the project is focused on the issue of creating efficient, scalable and in particular functionally safe, reliable and fault-tolerant electronic systems for use in electric autonomous vehicles that can get by without conventional, multichannel redundancy architectures or related approaches.
As part of the alliance project, the Machine Learning research division is examining issues related to preventive and predictive maintenance. The focus is on developing strategies for early warning and for forecasting the remaining runtime of the concrete electronic system by supporting the use of machine learning methods.
The Model-based Systems Engineering (MbSE) research division examines processes for the predictive reconfiguration of the system in which crucial functions can be shifted to other electronic control units (ECU) in case of a predicted fault. The underlying strategies are calculated based on system and fault prediction models during the development phase and then integrated into the ECU configuration. The strategies factor in a time horizon longer than methods based on fault detection, and allow the user of the vehicle to postpone a visit to the repair shop for instance (if appropriate by forgoing the use of less-important, non-critical functions).
Development of methods for the automatic detection of imminent (fault) patterns in the complex interaction between dynamic system parameters and operational conditions, fault detection notifications and environmental influences (sensor measurement values) for the purpose of creating improved diagnoses, analyses, decisions and communications in other systems, as well as the development of model-based methods for synthesizing predictive reconfiguration and degradation strategies.
Germany Federal Ministry of Education and Research (BMBF)
German government framework program for Research and Innovation 2016-2020: “Microelectronics Made in Germany „Mikroelektronik aus Deutschland – Innovationstreiber der Digitalisierung“
Publication: "Elektronik für autonomes elektrisches Fahren (Elektronom)"
01.01.2019 - 30.06.2022
