Generative AI-based Prompt Evolution Engineering Design Optimization With Vision-Language Model

The paper introduces a Prompt Evolutionary Design Optimization (PREDO) framework that enhances engineering design optimization by using generative AI and large language models to evolve text prompts, guiding the creation of practical 3D car designs through a vision-language model that penalizes impractical designs and a physics-based solver for performance evaluation, resulting in a significant increase in the generation of viable car designs compared to traditional methods.

Abstract

Engineering design optimization requires an efficient combination of a 3D shape representation, an optimization algorithm, and a design performance evaluation method, which is often computationally expensive. We present a prompt evolution design optimization (PEDO) framework contextualized in a vehicle design scenario that leverages a vision-language model for penalizing impractical car designs synthesized by a generative model. The backbone of our framework is an evolutionary strategy coupled with an optimization objective function that comprises a physics-based solver and a vision-language model for practical or functional guidance in the generated car designs. In the prompt evolutionary search, the optimizer iteratively generates a population of text prompts, which embed user specifications on the aerodynamic performance and visual preferences of the 3D car designs. Then, in addition to the computational fluid dynamics simulations, the pre-trained vision-language model is used to penalize impractical designs and, thus, foster the evolutionary algorithm to seek more viable designs. Our investigations on a car design optimization problem show a wide spread of potential car designs generated at the early phase of the search, which indicates a good diversity of designs in the initial populations, and an increase of over 20\% in the probability of generating practical designs compared to a baseline framework without using a vision-language model. Visual inspection of the designs against the performance results demonstrates prompt evolution as a very promising paradigm for finding novel designs with good optimization performance while providing ease of use in specifying design specifications and preferences via a natural language interface.

Method

The authors used a methodology that combines evolutionary strategies with a generative AI approach to optimize 3D car designs. They employed a vision-language model to assess and penalize impractical designs, and a physics-based solver to evaluate the aerodynamic performance of the generated designs. The core of their approach is the iterative generation of text prompts that embed user specifications, which are then used to synthesize 3D car models. The evolutionary algorithm selects and evolves these prompts based on the feedback from the vision-language model and the physics simulations, aiming to improve the practicality and performance of the generated designs.

Main Finding

The authors discovered that their Prompt Evolutionary Design Optimization (PREDO) framework, which integrates a generative AI model with a vision-language model and a physics-based solver, significantly improves the generation of practical car designs. They found that the inclusion of the vision-language model to penalize impractical designs led to an increase of over 20% in the probability of generating viable designs compared to a baseline framework without such a model. Additionally, they observed that the framework maintained a high degree of accuracy in identifying practical designs throughout the evolutionary process, and that the visual and physical guidance provided by the models was crucial in steering the evolutionary algorithm towards more feasible design solutions.

Conclusion

The conclusion of the paper is that the Prompt Evolutionary Design Optimization (PREDO) framework, which incorporates a vision-language model to guide the evolutionary process and penalize impractical designs, is a promising approach for engineering design optimization. The authors demonstrated that their method outperforms traditional methods by significantly increasing the likelihood of generating practical car designs. They also highlighted the importance of combining generative AI, large language models, and physics-based solvers to achieve efficient and user-friendly design optimization. The paper concludes by suggesting that further research is needed to explore the impact of prompt length on evolution performance, the effects of different penalty weights, and the application of the proposed method to other engineering design domains.

Keywords

Prompt Evolutionary Design Optimization (PREDO), generative AI, large language models (LLMs), vision-language model, evolutionary strategy, optimization objective function, physics-based solver, text-to-3D model, design performance evaluation, vehicle design scenario, computational fluid dynamics simulations, natural language interface, engineering design optimization, text prompts, 3D shape representation, optimization algorithm, design performance evaluation method, evolutionary design optimization, text-to-X generative models, impractical design mitigation, design optimization objective, evolutionary variation, design specifications, preferences, natural language processing (NLP), product development, geometric deep learning, 3D shape synthesis, generative models, latent representations, text-to-3D models, evolutionary design optimization process, fitness scores, convergence criteria, evolutionary selection pressure strategy, evolutionary variation, design objective, penalty score, design performance, practical designs, impractical designs, soft constraint, black-box solver, vision-language model as soft constraint, penalty weight, design optimization, projected frontal area, normalized drag coefficient, design performance objective, evolutionary design optimization framework, design accuracy, design evolution, design candidates, design optimization example, research opportunities, generative AI for evolutionary engineering design optimization.