Abstract
With increasing emphasis on renewable energy and depleting oil reserves along with rising emission levels, the world is moving towards the use of electric vehicles (EV). However, to ensure safety and reliability in fast evolving systems like EV which is subjected to various operating conditions, there is a need for a systematic fault diagnosis approach. In highly complex systems like EV where accurate representative physics-based models seem challenging, the use of machine learning techniques on the available extensive data from sensors can be useful for fault diagnosis. In this work, a fault diagnostic approach is developed and implemented for a real running EV's power system based on machine learning techniques (such as Random Forest, KNN, linear and logistic regression) applied on the acquired data log. A distinct part of the EV power system, the Auxiliary circuit (comprising of batteries, the distribution unit, motors, controllers, etc.), is considered for implementing the real-time fault detection and isolation scheme. For detecting starting instances of the electric vehicle faults, the required optimal time window is computed. Further, pre-fault and postfault analysis are done on the acquired data log to find symptoms of anomalies across different variables characterizing the electric vehicle behavior. Finally, consolidating the prefault and post-fault analysis the consensus on the resulting fault detection and isolation is validated for actual faults occurred. This approach of fault diagnosis can also be implemented in real time and be extended to the entire EV power system or any other systems where extensive data capturing all possible operating points are available through sensors under normal and faulty conditions.
Keywords: Electric Vehicle, Power system, Machine Learning, Fault Diagnosis, Auxiliary Circuit