Predicting Battery Aging in Ride Sharing and Connected Electric Vehicle Applications

PI: BJ Yurkovich

 

Predicting both natural and unnatural death of a battery pack is a difficult endeavour given the complexities and variations in the environment that an automotive battery pack experiences. In this project, we explore developing data models leveraging machine learning and large amounts of data to develop real time algorithms that are capable of predicting when a battery becomes unusable in an automotive application. We focus on training computer models (neural networks) with large amounts of data that need only a small set of inputs during actual vehicle operations allowing for rapid, low computation methods for prediction to be employed in a connected, resource constrained vehicle application. Through this work, we have established a baseline methodology for achieving such prediction goals, as well as validated our findings on some rapidly aged Li-Ion battery cells. Future work includes expanding the methodology to include more automotive battery data from real vehicles and establishing more sophisticated machine learned models.