A Comprehensive Data-driven Fault Diagnosis Method for Electric Vehicles

Recent years have witnessed a transition in energy structure, where large number of electronic devices and systems are introduced in multiple fields including industry, academia, commerce, and so forth. Safe and efficient operations of these systems are critical to ensure productivity as well as to avoid hazard, which poses strict demands on fault diagnosis. Most of traditional methods tend to focus on algorithm design or certain types of hardware or software flaws under given operational conditions, thus not suitable for modern electronic systems that may suffer from a variety of different faults. In this paper, on top of our previous work on big data, a systematic way of fault diagnosis in electric vehicles is put forward, which covers data processing, feature extraction, model-based diagnosis, and model fusion. The proposed method is trained and validated using data from real-world electric vehicles, which are representative examples of modern complex systems. Results show a diagnosis accuracy over 95% can be achieved with a comprehensive consideration of fault modes under a variety of operational scenarios. The proposed algorithm can also be used to indicate key features leading to faults so that system level upgrade can be performed accordingly. The design criteria and idea of the algorithm is also adaptable to other systems or applications with minor changes.