AI-generated EV designs from MIT may influence appearance of automobiles in future
According to the engineers, this open-source database, called "DrivAerNet++," has designs based on the most popular automobile models now on the market.
News, 23 December 2024
Future automobiles can be swiftly constructed by combining the more than 8,000 electric vehicle (EV) ideas developed by MIT engineers with artificial intelligence (AI).
According to the engineers, this open-source database, called “DrivAerNet++,” has designs based on the most popular automobile models now on the market. These designs are displayed as 3D models that include details like the design’s aerodynamics.
Despite being around for almost a century, electric automobiles have recently seen a sharp increase in popularity. Companies must invest years, resources, iterations, and revisions in the design of these cars before they can create a tangible prototype.
The parameters and outcomes of these tests (as well as the prototypes’ aerodynamics) are confidential due to their unique nature. According to the scientists, this implies that notable improvements in EV range or fuel efficiency may be gradual.
However, the goal of the new database is to exponentially accelerate the search for improved automotive designs.
This digital collection of automobile designs contains comprehensive information on aerodynamics and specifications. According to the researchers, this digital library might be used in the future to create new designs for electric cars when paired with AI models.
According to the engineers, manufacturers can create EV designs more quickly than ever before by simplifying a time-consuming procedure.
AI-powered car design creation in a matter of seconds
The MIT SuperCloud, a superpowerful cluster of computers used for scientific research that can be accessed remotely, was utilized to construct the dataset, which yielded 39 terabytes of data while using 3 million central processing unit hours.
For every baseline automobile type, the researchers used an algorithm that methodically adjusted 26 parameters, including as vehicle length, underbody characteristics, tread and wheel designs, and windshield slope. Additionally, they used an algorithm to identify whether a newly created design was truly original or a clone of an existing design.
After that, each 3D design was transformed into a readable format, such as a mesh, a point cloud, or just a list of measurements and specifications. Lastly, they computed the air flow surrounding each created design using intricate fluid dynamics models.
