A physics-informed AI framework could help engineers rapidly design solid rocket grain structures that deliver targeted combustion behavior, although experimental validation is still needed.
Paper: Generative AI empowered inverse design of 3D energetic material structures for customizable combustion. Image credit: AI-generated image created using ChatGPT/OpenAI
A recent study appearing as an article in press in the journal Communications Engineering introduces a generative artificial intelligence (AI) framework for the inverse design of three-dimensional energetic material structures with customizable combustion behavior. The researchers combined a Deep Eikonal Auto-Decoder (DEAD), a Denoising Diffusion Probabilistic Model (DDPM), and gradient-based optimization to generate structures that match target combustion profiles. The findings demonstrate a fast and effective simulation-based approach for designing energetic materials with tailored combustion profiles.
Combining Generative AI for Customizable Combustion
Energetic materials are important for a wide range of aerospace, defense, and propulsion systems as they release energy through controlled combustion. In solid rocket motors, the geometry of the energetic material determines how the burning surface evolves. This behavior directly influences pressure generation, thrust, and burn duration. Designing grain structures that produce specific combustion profiles is therefore essential for achieving targeted propulsion performance.
Conventional design methods rely on repeated simulations, numerical optimization, and engineering expertise to identify suitable grain geometries. Although these approaches can produce effective designs, they require significant computational resources and explore only a small fraction of the available design space. Recent advances in AI have accelerated material design, but most AI models predict combustion performance from an existing geometry rather than generating a new structure that meets a predefined performance target.
In this study, researchers developed a hybrid AI framework for inverse design. The framework combines a Deep Eikonal Auto-Decoder (DEAD), a Denoising Diffusion Probabilistic Model (DDPM), and gradient-based optimization. Together, these components generate simulation-feasible three-dimensional grain structures that closely match user-defined pressure-time combustion profiles.
Training a Physics-Informed AI Model for Inverse Design
The researchers built a multi-stage workflow that combined physics-based geometry modeling with generative artificial intelligence. They first constructed a large training dataset containing 47,800 three-dimensional energetic material structures derived from 42 representative grain geometries. These structures produced 4,763,200 valid structure-performance data pairs after filtering nonphysical chamber-pressure cases, providing a comprehensive foundation for model training and validation.
The first stage employed a DEAD to encode and reconstruct three-dimensional grain geometries. The model described each geometry using a signed distance function constrained by the eikonal equation. This representation preserved smooth geometric boundaries while accurately capturing changes in the burning surface during combustion.
The second stage used a Denoising Diffusion Probabilistic Model (DDPM) to generate latent structural representations from user-defined pressure-time curves. A gradient-based optimization algorithm then refined the generated geometries to improve agreement with the target combustion profile. Finally, the team evaluated the framework across multiple design scenarios, including dual-thrust rocket grains and other complex combustion profiles.
Generative AI Enables Customized Combustion Design
The proposed framework successfully generated three-dimensional energetic material structures that closely reproduced user-defined combustion profiles in computational tests. The generative model directly produced candidate structures from the desired pressure-time curve. Most designs were generated and refined in approximately 15 minutes, substantially reducing computational time while maintaining high simulated performance matching.
The Deep Eikonal Auto-Decoder played a key role in preserving geometric realism throughout the design process. By enforcing eikonal constraints, the model reconstructed smooth burning surfaces that accurately represented combustion evolution. This physics-informed representation enabled the diffusion model to generate diverse grain geometries while avoiding unrealistic shapes.
The framework also demonstrated its ability to solve challenging inverse design problems. It generated dual-thrust grain structures capable of delivering different thrust levels during successive combustion stages and accurately reproduced complex pressure-time combustion profiles. Gradient-based optimization further refined the generated structures, bringing their predicted combustion behavior closer to the target profiles.
The hybrid AI framework reduced the time and computational effort required to design energetic materials. It rapidly generated multiple grain structures that met predefined combustion requirements, reducing reliance on repeated conventional optimization cycles. By combining physics-based modeling with generative AI, the framework provides a promising, scalable computational approach for designing customized energetic materials at the conceptual design stage.
Accelerating the Design of Next-Generation Energetic Materials
The study demonstrates the potential of generative AI to advance the inverse design of energetic materials. This integration of physics-informed modeling and generative AI enables rapid design while maintaining the geometric accuracy required for simulated combustion prediction.
The framework gives engineers greater flexibility during the design process. It has the potential to accelerate the development of customized energetic materials for solid rocket propulsion and other applications that demand precise combustion control. The results highlight the framework's potential, but additional development and validation are needed before it can be deployed in practical engineering applications.
Future studies should expand the training dataset, incorporate higher-fidelity combustion simulations, and experimentally validate the AI-generated grain structures. The authors noted that future work should account for erosive burning, nozzle throat erosion, thermal radiation, pressure-induced deformation, and structural integrity. Applying the framework to more complex energetic material systems could further improve its accuracy and versatility. As generative AI gains traction in engineering design, this physics-informed approach could accelerate the development of next-generation propulsion materials while reducing design time and computational effort.
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.