Designing reliable aircraft can be a challenging and time-consuming process. However, researchers are constantly striving to find ways to speed up this process and make it more efficient. One potential solution comes in the form of deep learning models, which have the ability to train machine learning algorithms to help designers identify the most promising aircraft designs or potential flaws.
To train these models, researchers require comprehensive datasets that contain a wide range of aerial vehicle designs. However, compiling these datasets can be difficult as many designs are protected by proprietary contracts or are simply difficult to source.
In a recent development, researchers at SRI International, the Southwest Research Institute, and Vanderbilt University have created a large-scale dataset called AircraftVerse. This dataset consists of thousands of aircraft designs of varying complexities and could be used to train machine learning algorithms to assist designers in their work.
AircraftVerse contains 27,714 diverse air vehicle designs — the largest corpus of engineering designs with this level of complexity, explain Adam D. Cobb, Anirban Roy, and their colleagues in their research paper.
Unlike existing datasets used to train machine learning algorithms for computer-assisted design (CAD), such as SketchGraphs, DeepCAD, and ABC datasets, AircraftVerse contains fully-fledged aircraft designs that combine multiple components like propellers, wings, motors, and batteries.
Each design in the dataset comprises various artifacts, such as a symbolic design tree describing topology and subsystems, a Standard for the Exchange of Product (STEP) model data, a 3D CAD design using a stereolithography (STL) file format, a 3D point cloud representing the design’s shape, and evaluation results from high-fidelity physics models that capture performance metrics like maximum flight distance and hover-time.
The aircraft designs included in the dataset were created using a deep learning-based approach, guided by general rules provided by expert aircraft designers. The final versions of these designs were then put through engineering models that generated metadata summarizing their specific characteristics and performance.
As part of their dataset release, the researchers also present baseline surrogate models that utilize different design representation modalities to predict design performance metrics. This additional information can provide valuable insights and aid in the use of learning in aircraft design.
The good news for designers and developers worldwide is that this newly created dataset, along with its baseline models and underlying code, is now publicly available online. With accessibility to this dataset, designers and developers can benefit from its wealth of information to assist them in the design and evaluation of new aerial vehicles.
In conclusion, the creation of the AircraftVerse dataset by a team of researchers represents a significant step forward in the field of aircraft design. By providing a comprehensive dataset of diverse aircraft designs, this research has the potential to greatly enhance the use of machine learning in the design and development of aerial vehicles. Designers and developers worldwide now have access to this valuable resource, which could revolutionize the way aircraft are designed and evaluated in the future.
