The automotive industry could be facing a potential design revolution with the introduction of DrivAerNet++, a large dataset containing over 8,000 automotive designs. The massive project, carried out by engineers at MIT, it promises to accelerate the development of more efficient cars from an energy point of view and with lower environmental impact.
The DrivAerNet++ dataset was created to provide engineers and designers with an extensive library of automotive designs, each with detailed aerodynamic specifications. These designs were generated taking inspiration from the most common models on the roads today, ensuring the relevance and applicability of the data. Using generative artificial intelligence toolsengineers can quickly analyze massive amounts of data, identify patterns, and generate new designs efficiently. This process not only reduces development time, but also improves the accuracy and effectiveness of design solutions, potentially reducing research and development costs.
MIT engineers simulate the aerodynamics for a given shape of automobile, which they represent in various modes, including “camsurface pi” (left) and “flow lines” (right). Credits: Mohamed Elrefaie:
Each design in the DrivAerNet++ dataset includes detailed simulated fluid dynamics data, allowing developers to evaluate how air moves around vehicles and optimize designs for maximum aerodynamic efficiency. This feature is crucial for increasing fuel efficiency in internal combustion engines and the range of electric vehicles. DrivAerNet++’s ultimate goal is to support the transition to a more sustainable automotive future.
“This dataset lays the foundation for the next generation of AI applications in engineering, promoting efficient design processes, reducing R&D costs and driving advancements towards a more sustainable automotive future“, says Mohamed Elrefaie, a doctoral candidate in mechanical engineering at MIT.
Elrefaie and his colleagues will present a paper detailing the new dataset and applicable AI methods at the NeurIPS conference in December. Co-authors are Faez Ahmed, assistant professor of mechanical engineering at MIT, Angela Dai, associate professor of computer science at the Technical University of Munich, and Florin Marar of BETA CAE Systems. “Often, in car design, the preliminary process is so expensive that manufacturers can only make small changes from one version to another“, says Ahmed. “But if you have larger datasets where you know how each design performs, you can now train machine learning models to iterate quickly, thus increasing the likelihood of getting a better design.”
“This is the best time to accelerate innovations in the automotive world, as cars are among the biggest polluters in the world, and the faster we can reduce that contribution, the more we can help the climate,” says Elrefaie.
