Virtual data-based generative optimization using domain-adaptive designable data augmentation (DADDA): Application to electric vehicle design

In the field of virtual product development, product performance is evaluated at an early stage through metamodel-based optimization using simulation, reducing development time and cost, and improving product quality. However, it takes a lot of time to construct a simulation model. The model must go through calibration and validation process to reduce errors with actual systems. This study proposes a novel engineering design process that can perform virtual data-based generative optimization by adapting the design domain of existing systems to those of new systems without requiring many simulations. A domain-adaptive designable data augmentation (DADDA) algorithm is proposed using inverse generator-implemented data augmentation, domain adaptation, and design optimization techniques. The DADDA algorithm can quickly provide optimal design variables for new systems compared to the genetic algorithm-based approximate optimization. The proposed process was applied to electric vehicle design. Driving responses and design variables related to safety performance were selected, and a small amount of training data was obtained using Modelica-based electric vehicle. As a result, it was shown that DADDA could significantly reduce the time required to derive optimal design variables by approximately 53% compared with the existing method. In addition, it is possible to derive the optimal performance for a new electric vehicle, which is approximately 21% better than that of an existing vehicle.