We just published a new preprint “CarHoods10k: An Industry-grade Data Set for Representation Learning and Design Optimization in Engineering Applications” introducing the OSU-Honda automobile hood data set. The data set comprises over 10,000 3D mesh geometries of industry-grade car hood frames, including corresponding design parameters and performance values. It was generated by an automated, industry-grade Computer Aided Design (CAD) workflow and provides realistic designs that were validated by experts with respect to realism, manufacturability, variability, and performance.
In our preprint, we demonstrate the application of such approaches to our data set to solve typical engineering tasks. These include unsupervised representation learning and design space exploration using geometric deep learning, performance prediction using machine learning, and topology optimization under manufacturability constraints based on evolutionary algorithms. The OSU-Honda automobile hood data set offers researchers and practitioners the unique opportunity to develop novel machine learning and AI methods on realistic data with relevance to real-world engineering design.
The data set is made public without any restrictions to enable the evaluation and development of state-of-the-art machine learning approaches for the automotive engineering design domain. We invite you to use the data set and look forward to seeing your research on it.
