[Paper Review] RePaint-Enhanced Conditional Diffusion Model for Parametric Engineering Designs under Performance and Parameter Constraints

Ke Wang, Hoang Nguyen Vu+3 more|arXiv (Cornell University)|Jan 30, 2026

Probabilistic and Robust Engineering Design0 citations

TL;DR

Introduces a RePaint-enhanced framework that uses a pre-trained performance-guided DDPM to generate parametrically constrained designs from partial references without retraining, enabling mask-based repainting under performance and parameter constraints.

ABSTRACT

This paper presents a RePaint-enhanced framework that integrates a pre-trained performance-guided denoising diffusion probabilistic model (DDPM) for performance- and parameter-constraint engineering design generation. The proposed method enables the generation of missing design components based on a partial reference design while satisfying performance constraints, without retraining the underlying model. By applying mask-based resampling during inference process, RePaint allows efficient and controllable repainting of partial designs under both performance and parameter constraints, which is not supported by conventional DDPM-base methods. The framework is evaluated on two representative design problems, parametric ship hull design and airfoil design, demonstrating its ability to generate novel designs with expected performance based on a partial reference design. Results show that the method achieves accuracy comparable to or better than pre-trained models while enabling controlled novelty through fixing partial designs. Overall, the proposed approach provides an efficient, training-free solution for parameter-constraint-aware generative design in engineering applications.

Motivation & Objective

• Motivate and enable performance- and parameter-constrained parametric design generation without retraining.
• Enable missing design components to be completed from partial references while honoring constraints.
• Provide an efficient inference-time mechanism to repaint partial designs under constraints.

Proposed method

• Utilizes a pre-trained performance-guided denoising diffusion probabilistic model (DDPM).
• Implements a mask-based resampling (RePaint) during inference to repaint partial designs.
• Enforces performance and parameter constraints during the repainting process without retraining the model.
• Applies the framework to parametric ship hull and airfoil design benchmarks to demonstrate feasibility.

Experimental results

Research questions

• RQ1 Can partial reference designs be completed into full designs while meeting performance and parameter constraints without retraining the DDPM?
• RQ2 How does mask-based RePaint perform in enforcing constraints and enabling controlled novelty in engineering designs?
• RQ3 What is the comparative accuracy and design quality relative to pre-trained models on parametric hull and airfoil tasks?

Key findings

• The method generates novel designs with expected performance based on a partial reference design.
• RePaint achieves accuracy comparable to or better than pre-trained models on the evaluated tasks.
• The approach enables controlled novelty by fixing partial designs during generation.
• The framework offers a training-free solution for parameter-constraint-aware generative design in engineering contexts.