Batteries are complex electrochemical devices whose short-term and long-term operational lives are highly dependent on their construction, materials, and use cases. Models have been developed and used to improve the performance of batteries by providing different material design, model based control and model based battery management systems¹.

There is a recent trend in data-science-based approaches for modeling different devices and processes^2,3. It has been demonstrated that using machine learning techniques can retain 99.9% of the accuracy of a complex model while improving solve time by four orders of magnitude, which can allow for the use of high accuracy models in a real-time control scenario³.

When data science techniques, and machine learning in particular, are implemented in battery systems, they are typically used as black boxes to predict battery behavior⁴. In this talk, we will highlight why this approach is ill suited for use in physics-based battery models, which are highly nonlinear, and how an understanding of the system can improve the result of the machine learning effort for use in optimization and control, while still allowing for the prediction of important internal states. Additionally, we will highlight ways to employ the flexibility of machine learning algorithms to allow for more accurate state estimation and voltage prediction.

Real-time model-based control has demonstrated significant improvements in time remaining, state of charge, and state of health, even using simple equivalent circuit models⁵. By reducing the order and solve time of higher complexity models, we believe that better predictability is possible.

Acknowledgments

The authors thank the United States Department of Energy (DOE) for the financial support for this work though the Advanced Research Projects Agency – Energy (ARPA-E) award #DEAR0000275, as well as the Clean Energy Institute (CEI) through the DIRECT program at the University of Washington.

References

1. Y. Dai, and V. Srinivasan,. J. Electrochem. Soc., 163, A406 (2016).
2. D. A. C. Beck, J. M. Carothers, V. R. Subramanian, and J. Pfaendtner, “Data science: Accelerating innovation and discovery in chemical engineering,” AIChE J., vol. 62, no. 5, pp. 1402–1416, May 2016.
3. Blake R. Hough, author. Jim Pfaendtner, degree supervisor. 2016 Thesis (Ph. D.)--University of Washington,2016.
4. J. Liu, A. Saxena, K. Goebel, B. Saha, and W. Wang, “An Adaptive Recurrent Neural Network for Remaining Useful Life Prediction of Lithium-ion Batteries,” NATIONAL AERONAUTICS AND SPACE ADMINISTRATION MOFFETT FIELD CA AMES RESEARCH CENTER, NATIONAL AERONAUTICS AND SPACE ADMINISTRATION MOFFETT FIELD CA AMES RESEARCH CENTER, Oct. 2010.
5. Williard, Nick, Wei He, and Michael Pecht. "Model Based Battery Management System for Condition Based Maintenance." Technical Program for MFPT 2012, The Prognostics and Health Management Solutions Conference - PHM: Driving Efficient Operations and Maintenance, 2012