(no-name)

As vehicle electrification increases, concerns over electric vehicle (EV) charging infrastructure are rising. New charging infrastructure is being deployed to meet growing demand. To ensure this demand is continuously met, the chargers being built must be reliable. One critical component of high power, efficient EV chargers is the MOSFET. MOSFETs are electronic switches that control the flow of power in an EV charger and convert AC grid power into DC power that can be used to charge the battery. High power silicon carbide (SiC) MOSFETs are a relatively new to EV charging applications. Use of these devices in EV chargers offer substantial improvements in efficiency and size; however, the impact of these devices on charger reliability is not as well understood as older devices. To this end, accurate modeling and monitoring techniques are necessary to ensure EV chargers designed with SiC MOSFETs are both efficient and reliable.

A critical indicator of SiC MOSFET health status is the device on-resistance (Rdson). Like friction on a road or drag in the air, resistance impedes the flow of electricity in a circuit. All electronic components have some resistance. When a MOSFET is off, it blocks the flow of electricity completely. When a MOSFET is on, it allows electricity to flow freely, with a very small resistance. As a MOSFET ages, this resistance increases until the MOSFET fails. Predicting MOSFET on-resistance is also predicting MOSFET age.

Changes in temperature, humidity, and movement cause the materials inside MOSFETs to wear out making the flow of electricity more difficult. These changes in on-resistance have a strong dependency on MOSFET design, structure, and fabrication. The increase in on-resistance over time also varies between device type, manufacturer, form factor, and specification. This multi-factor dependance makes creating a simple model for MOSFET on-resistance difficult. Directly measuring on resistance is also complicated as the voltage and current of each device must be measured. This has high hardware and computational overhead.

As a solution, I am exploring data-driven models leveraging machine learning to both model MOSFET on-resistance behavior in design and monitor MOSFET on-resistance during use. The modeling developed will enable reliability-oriented EV charger design. The developed monitoring tools will facilitate rapid estimation of MOSFET on-resistance and enable predictive maintenance.

The work involves the collection of data from which to build these data driven models and the creation, training, and testing of the AI algorithms to be used. An automated testbed, in its second revision, is being used to collect the necessary data. Once collected, the data is cleaned and used to train, validate, and compare the utility of a wide array of ML methods. The highest performing models can be used in EV charger applications. A first dataset has already been collected and simple algorithms such as random forest and tree boosting show promise predicting MOSFET on-resistance with an error of less than 4% by simply measuring and recording the temperature of the MOSFET.