Model-driven software development and verification solutions for safety critical battery management systems : a quantitative evaluation of probabilistic inference & artificial intelligence methods

Fleischer, Christian; Sauer, Dirk Uwe (Thesis advisor); Schaltz, Erik (Thesis advisor)

Aachen : ISEA (2020, 2021)
Book, Dissertation / PhD Thesis

In: Aachener Beiträge des ISEA 142
Page(s)/Article-Nr.: 1 Online-Ressource (x, 356 Seiten) : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2020


The automotive industry is reaching an unprecedented level of turning point. The new challenges are wide-ranging, the future of combustion engine has been signed and sealed, as electrical vehicles with their high-voltage (HV) batteries are predestined as replacing propulsion systems. Today, SW methods and algorithms are not mature enough consider-ing that the HV traction battery is a complex, electro-chemical device in a new application, with the prerequisite in knowledge and understanding of the working principles including aging mechanism. These cause complex multi-time scale dynamics by superimposed interactions of chemical reactions which are complex to model. In recent years the need of battery systems and its advanced SW has increased significantly in order to reach performance, energy, lifetime and safety requirements. At the same time, the cost must drop to compete with conventional propulsion systems. Therefore, the question which put OEMs at stake is how to use the battery in an optimal way to ensure a competitive HV traction battery system. Therefore, this work discusses model-based battery management design, which does not mean applying visual methods to address the problems, rather we are using qualitative models to represent battery behaviour, e.g. looking at the electrical properties of the voltage response when applying current to the battery. Providing accurate power and energy estimates to the vehicle propulsion system is part of the on-line battery management primary objective. This work provides a set of adaptive algorithms for online battery state- and parameter estimation evaluating different aspects, such as their efficiency, robustness, accuracy and implementation effort on target. Bayesian optimal filters for non-linear systems, linearization methods e.g. extended Kalman filter, and other numerical approximation methods (Gaussian sum, iterative quadrature, and deterministic sampling approximation) are quantitatively benchmarked. The comprehensive discussion involves the definition of the estimation framework which is integrated for state-and parameter estimation using joint and dual estimation. Additionally, new estimation techniques from the field of machine / deep learning are introduced to predict battery lifetime. The Bayesian estimators are benchmarked in regards to various battery equivalent circuit models as they are needed for accurate voltage, state-of-charge, parameter or available power / energy estimation. Battery management systems are safety-critical software systems, containing code which failure or malfunction can trigger or contribute to a hazard jeopardizing people’s lives. The quality of released SW is subject to the effort of which level the testing is executed. Therefore, SW testing and hardware-in-the-loop BMS validation on functional and non-functional level is introduced, leading to a novel battery system simulation environment including fault insertion. Additionally, a new way of ISO26262 complaint model-driven software development for BMS is discussed, introducing a new tool-chain for SW architecture and model development on multi-core target CPUs optimizing computing power for individual cell monitoring. The introduced ecosystem fulfils the requirements during development phase with traceability, validation and verification.