Electric vehicles (EVs) are all about reducing carbon emissions. Artificial intelligence (AI) can help. This FAQ looks at a few examples like how AI and machine learning (ML) can be used to make context-aware operational recommendations to EV drivers for extending EV battery lifetime and how AI can work together with wide bandgap semiconductors (WBGs) like gallium nitride (GaN) to produce intelligent power electronics, increasing efficiency, reducing drivetrain weight, and supporting preventative maintenance for increased availability, reliability, and safety.
Driving an EV is a new experience for most drivers, and they may not be fully aware of how best to use the vehicle to maximize battery lifetime. An AI/ML system has been developed to assist EV drivers make optimal use of the battery, thereby extending battery life.
The algorithm is designed to make long-term and short-term operational context forecasts using parameters like time of day, driving patterns, and weather conditions. It also develops predictions of EV operating parameters using various prediction inputs and contextual forecasts. The output of the algorithm is a series of recommendations to the driver on how best to manage battery use based on the parameters that have the largest impact on battery life like the ambient temperature, the C-rate of charging and discharging, the state of charge (SoC) and the time intervals between full cycles. (Figure 1). Two keys to this approach are determining context and predicting parameters.
The context begins with anticipated long-term activities over the next day, like driving to and from work, based on a historical database of driving activity. So, the long-term context needs to distinguish between workdays and non-workdays and is used to provide a baseline expectation of needed activities. For example, it can be combined with the SoC of the battery and the temperature in the ML algorithm to predict the optimal time and optimal C-rate for charging the battery. The ML algorithm also considers short-term deviations from the anticipated driving context to provide adjustments in the recommendations. Short-term deviations can include additional trips, changes in routing to run errands, unexpected traffic due to an accident, weather changes, and so on. The ML algorithm continuously updates the anticipated EV context over the next 24 hours for every quarter-hour. The context is combined with EV parameters to produce the driver recommendations.
Predicting parameters and adding advice
In addition to EV usage context, EV operating parameters need to be forecast for the same 24-hour period and time increments. The ML algorithm predicts battery degradation parameters like SoC, and discharge C-rate based on the anticipated driving context. A significant challenge here is that it will take a large database of EV context predictions and correlated operating parameters to train the ML sufficiently for it to consistently make useful recommendations. The algorithm will initially be limited in its abilities and would be expected to get better over time as the database grows. As a result, any advice needs to be accompanied by an explanation of its intent and how/why it was derived from the context. AI will be used to develop the natural language explanation needed by the driver.
AI and WBGs
GaN can be used to make high-efficiency and compact power modules for use in EV traction inverters, DC/DC converters, and battery chargers. In one case, GaN is being used to design a compact motor driver with integrated AI failure prediction capabilities. Predicting failures to enable preventative maintenance is especially important in applications like EVs where failures can result in serious malfunctioning of the vehicle and possible harm to the occupants. Enabling preventative maintenance can also increase the operational life of the power converter.
The project aims to develop AI-based failure prediction algorithms that can run efficiently on a microcontroller (MCU). The MCU will be integrated into the motor driver along with recently developed vertical GaN devices to make a cognitive power module. The co-design of the power semiconductor devices, microcontroller, and AI prediction algorithms is expected to result in a high-efficiency solution that can detect potential power converter faults before they result in failure. In later designs, the AI element will be expanded to also detect potential failure in the traction motor (Figure 2).
AI has the potential to significantly improve EV efficiency, reliability, and safety. Its potential applications include providing drivers with recommendations of actions to take to extend battery life and monitoring the power electronics in the motor drive and the traction motor to provide forecasts of potential failures to make EVs more efficient and EV occupants safer.
Artificial Intelligence for Power Electronics in Electric Vehicles: Challenges and Opportunities, Journal of Electronic Packaging
Artificial Neural Networks-Based Multi-Objective Design Methodology for Wide-Bandgap Power Electronics Converters, IEEE Open Journal of Power Electronics
Context-aware recommendations for extended electric vehicle battery lifetime, Sustainable Computing
Leveraging AI and Machine Learning to Optimize EV Energy Efficiency, V-HOLA Labs
PowerCare, Fraunhofer Institute
Prediction of EV Charging Behavior Using Machine Learning, IEEE Access