Al-Based Self-Learning Synthesis for Electrified Drives

Authors

A. Sturm, R. Henze, C. Wolgast, P. Eilts, Braunschweig University of Technology; F. Küçükay, Innovationsgesellschaft Technische Universität Braunschweig mbH, Braunschweig:

Year

2024

Print Info

Production/Publication ÖVK

Summary

The publication presents a tool chain for the synthesis of electrified drives with an AI-based self-learning control and optimization algorithm that differs significantly from the state of the art. The algorithm and the tool chain are the results of a research project funded by the DFG and successfully completed in 2023. The drive synthesis consists of a coupled combustion engine and transmission synthesis as well as synthesis-like optimizations for electric motors and batteries. The publication first presents the entire computer-aided tool chain, including the individual modules (synthesis, requirements verification, simulation and evaluation) and the implemented models and methods. A special focus is then placed on the control/optimization algorithm, which was newly developed as part of the project. This is necessary because the large number of variable synthesis parameters, such as the number of mechanical gears, stroke/bore ratio or number of cylinders, means that the possible solutions can be in the millions. A new type of algorithm based on artificial intelligence was developed to cope with the large number of possible combinations. In addition to the AI elements, the algorithm is supported by stored expert knowledge. The application of the algorithm to an optimization problem, which was previously described by a full factorial simulation, shows that the algorithm finds the global optimum in 1 % to 3 % (depending on the setting of the algorithm) of all theoretically possible simulations and has a very good coverage rate for the top 100 possible solutions. Finally, the tool chain is applied to a D-segment vehicle for the synthesis of hybrid drives and the results, including the functionality of the algorithm, are presented.

ISBN

978-3-9504969-3-2

DOI

https://doi.org/10.62626/u74k-c4nx