Form Representation and Derivation for Vehicle Body based on Neural Networks

The absence of form representation system hinders the development of parametric design method on vehicle body design. Based on the PointNet autoencoder deep network, a method for representing the form of car bodies is proposed. The model training is completed on the ShapeNet dataset to finish the pre-trained representation model CodeCar, and the representational effect and optimal representational space of CodeCar are explored. CodeCar is capable of representing the form of car bodies in three-dimensional space within a high-dimensional space. Based on CodeCar, the AutoStylist intelligent form exploration method proposed can be applied to the forward-looking design stage of automotive body styling and tested in actual automotive styling design projects. This research innovatively applies artificial intelligence technology in the field of automotive body design, trains the automotive body form representation model CodeCar, and proposes new-generation computational design methods such as AutoStylist, promoting the integrated development of parametric design and engineering design for car bodies, and further transitioning towards intelligent design methods.