Centralized Diagnosis of On-Board Power Supply Network Faults with LSTM-Networks

Electronic systems in automobiles for highly automated driving functions must be significantly more reliable than today. Improved diagnosis can help to archive this goal. Nowadays, electronic systems have implemented some diagnosis functions which refer to the functionality of the component based on the voltage at the input terminals. Faults in a vehicle power supply systems are often only detected by the systems when voltage levels are out of range. Advanced fault detection methods for the supply system that can identify a root cause or a fault location were not known. In this paper, a new diagnosis concept for the overall power supply system is proposed. A centralized diagnosis concept which could be implemented in a power supply distribution unit (PDU) consisting of electronic fuses is presented. By evaluating the current and voltage signals in the PDU, typical on-board power supply system faults can be classified and localized. The diagnosis approach is based on the training of an artificial intelligence model with simulation data. The trained model is then applied to real measurements in a laboratory setup. The results show good fault detection accuracy and short detection times of 50${\mu}$s for the investigated cases in a laboratory setup.