Decentralized diagnosis of permanent faults in automotive E/E architectures

This paper presents a novel decentralized approach for the diagnosis of permanent faults in automotive Electrical and Electronic (E/E) architectures. Both, the safety-critical real-time requirements and the distributed nature of these systems make fault tolerance in general and fault diagnosis in particular a crucial and challenging issue. At the same time, high unit numbers in manufacturing add cost efficiency as an important criterion during system design, which is conflicting with the use of often expensive explicit fault diagnosis hardware. To address these challenges, we propose a diagnosis framework that consists of two stages. In the first diagnosis determination stage, potential fault scenarios, such as defective Electronic Control Units (ECUs), are investigated to obtain a set of diagnosis functions. Specific diagnosis functions are used for each component in the network at runtime to determine whether a certain fault scenario is present. In the second diagnosis optimization stage, an optimization of diagnosis functions is proposed to determine trade-offs between diagnosis times and the number of monitored message streams. Experimental results based on 100 synthetic test cases give evidence of the feasibility and efficiency of the presented framework. Finally, an automotive case study demonstrates the practicability and details of our fault diagnosis approach.