This study proposes an ML-based interactive framework for early-stage design, addressing the challenge where physical prototypes are accurate but costly, and virtual prototypes are affordable but less reliable. The NN-based human-in-the-loop framework integrates pre-training and fine-tuning techniques to reduce reliance on extensive physical prototyping while maintaining model accuracy. Using projectile motion as an example, the framework demonstrates its ability to guide design by iteratively updating models based on limited experimental data and human expertise. The results highlight the framework’s effectiveness in achieving performance comparable to models trained on larger datasets, offering a cost-effective solution for creating accurate design models.
