Optimising Aerodynamics of Car using Dynamic Flow Lines Adaptation
DOI:
https://doi.org/10.21467/proceedings.7.6.57Keywords:
3D Convolutional Neural Networks, Genetic Algorithms, Computational Fluid DynamicsAbstract
Aerodynamic optimization plays a vital role in achieving sustainability, enhanced vehicle performance, and fuel efficiency in modern automotive design. Conventional methods, such as wind tunnel testing and computational fluid dynamics (CFD), are often time-intensive and resource-demanding, which can slow the design process. This study proposes a machine learning-based framework for dynamic flow line adaptation, enabling efficient real-time aerodynamic optimization for vehicles. By leveraging deep learning techniques, the approach mitigates the challenges associated with traditional methods, achieving notable improvements in computational efficiency and design flexibility. Our findings demonstrate substantial reductions in aerodynamic drag (12%) and computational overhead (15% compared to traditional CFD methods), while maintaining structural integrity within 95% confidence levels. The system achieves 92% prediction accuracy for drag coefficients with an average processing time of just 3.2 seconds per design iteration, facilitating more streamlined and cost-effective vehicle design workflows.
References
[1] Z. Yang, W. Gu, and Q. Li, "Aerodynamic design optimization of race car rear wing," in Proc. 2011 IEEE Int. Conf. Comput. Sci. Autom. Eng., Shanghai, China, 2011, pp. 642–646. doi: 10.1109/CSAE.2011.5952758.
[2] X. Wu, P. Wang, J. Li, S. Sun, and Y. Din, "Adjoint-based optimization for Blended Wing Body Underwater Gliders’ shape design," in Proc. 2018 OCEANS - MTS/IEEE Kobe Techno-Oceans (OTO), Kobe, Japan, 2018, pp. 1–5. doi: 10.1109/OCEANSKOBE.2018.8559363.
[3] T. Rios, B. Van Stein, T. Bäck, B. Sendhoff, and S. Menzel, "Point2FFD: Learning shape representations of simulation-ready 3D models for engineering design optimization," in Proc. 2021 Int. Conf. 3D Vision (3DV), London, United Kingdom, 2021, pp. 1024–1033. doi: 10.1109/3DV53792.2021.00110.
[4] S. J. Jacob, M. Mrosek, C. Othmer, and H. Köstler, "Deep learning for real-time aerodynamic evaluations of arbitrary vehicle shapes," Friedrich-Alexander-Univ. Erlangen-Nürnberg & Volkswagen AG, Erlangen, Germany, Tech. Rep., 2023. doi: 10.48550/arXiv.2108.05798.
[5] E. Sharifi, A. A. T. Borojeni, and M. H. Hekmat, "Investigation of the Adjoint Method in Aerodynamic Optimization Using Various Shape Parameterization Techniques," J. Braz. Soc. Mech. Sci. & Eng., vol. 32, no. 2, pp. 176–184, Jun. 2010. doi: 10.1590/S1678-58782010000200012.
[6] A. Usman, M. Rafiq, M. Saeed, A. Nauman, A. Almqvist, and M. Liwicki, "Machine Learning Computational Fluid Dynamics," in Proc. 33rd Swedish Artif. Intell. Soc. (SAIS) Workshop, Luleå, Sweden, Jun. 2021. doi: 10.1109/SAIS53221.2021.9483997.
[7] Z. Yang, J. Tan, X. Wang, Z. Yao, and B. Liang, "Reinforcement Learning-Based Robust Tracking Control Application to Morphing Aircraft," in Proc. 2023 Am. Control Conf. (ACC), San Diego, CA, USA, 2023, pp. 2481–2486. doi: 10.23919/ACC55779.2023.10156320.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Shreyas Reddy, Siddharth Rajendran , Rashid T Tahasildar, Surobhi Halder, Umme Haani (Author)
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
