Robust and Universal Battery Modeling Algorithm for ImpedanceSpectroscopy based on Standard MCU
Andreas Mangler,
Technische Universität Chemnitz
Impedance spectroscopy is a recognized non-invasive measuring method, especially for analyzing batteries. At present, impedance spectroscopy is used successfully in laboratory conditions; widespread industrial use in embedded systems has yet to be adopted. One of the challenges being faced is finding scalable modeling algorithms that can be used universally while still being suitable for autonomous embedded real-time systems with less memory requirements. A robust and universally applied methodology for non-parametric representation is substantiated here for describing the non-linear impedance model.
Adopting very stringent, monitored, and controlled processes is an additional future requirement when handling Li-ion batteries, as they are classified as hazardous goods. When selling millions of batteries, it is essential that all processes can be fully traced at any time. Modeling data that accumulates when selling batteries are stored and managed long term in an enterprise resource planning system. As a result, each battery or each battery stack has a simplified impedance model, thus acting as the battery’s virtual DNA upon its delivery. A key requirement for modeling and data storage in any enterprise resource planning system is to eliminate the use of differential equations. Establishing impedance spectroscopy in an array of application fields demands algorithms functioning, wherever viable, in accordance with a series of identical mathematical algorithms and methods, and robust and simplified modeling being possible. A three-dimensional representation is typically used to visualize the impedance of batteries. The algorithm should support any kind of parametric representation, e.g. depending on Z real, Z imaginary, frequency, time, settling time, relaxation time, temperature, state-of-charge, charging and discharging cycles, voltage or current profile. The research project emerged from a so-called big data approach that seeks to describe millions of different non-linear models with just one algorithm. To do so, a non-parametric impedance representation was chosen based on a model library consisting of standardized basic mathematical functions using line-search successive approximation, which was then combined with a defined methodology. By combining various approaches from recognized methods, such as frequency band segmentation, frequency band extension, determination of extrema, weight function, substitution, K-Nearest Neighborhood functions, and interpolation, it was possible to develop a new optimized algorithm that can be used for industrial, automotive or medical applications in the future.