Artificial Intelligence Supported EV Electric Powertrain for Safety Improvement

As an environmentally friendly transport option, electric vehicles (EVs) are endowed with the characteristics of low fossil energy consumption and low pollutant emissions. In today's growing market share of EVs, the safety and reliability of the powertrain system will be directly related to the safety of human life. Reliability problems of EV powertrains may occur in any power electronic (PE) component and mechanical part, both sudden and cumulative. These faults in different locations and degrees will continuously threaten the life of drivers and pedestrians, bringing irreparable consequences. Therefore, monitoring and predicting the real-time health status of EV powertrain is a high-priority, arduous and challenging task.
The purposes of this study are to develop AI-supported effective safety improvement techniques for EV powertrains. In the first place, a literature review is carried out to illustrate the up-to-date AI applications for solving condition monitoring and fault detection issues of EV powertrains, where recent case studies between conventional methods and AI-based methods in EV applications are compared and analysed. On this ground this study, then, focuses on the theories and techniques concerning this topic so as to tackle different challenges encountered in the actual applications. In detail, first, as for diagnosing the bearing system in the earlier fault period, a novel inferable deep distilled attention network is designed to detect multiple bearing faults. Second, a deep learning and simulation driven approach that combines the domain-adversarial neural network and the lumped-parameter thermal network (LPTN) is proposed for achieve IPMSM permanent magnet temperature estimation work. Finally, to ensure the use safety of the IGBT module, deep learning -based IGBT modules’ double pulse test (DPT) efficiency enhancement is proposed and achieved via multimodal fusion networks and graph convolution networks.