Session: 06-11-01: Applications of Artificial Intelligence/Machine Learning in Aerospace Engineering

Paper Number: 163165

Democratize Composites Simulation Using Large Language Models 

Composite materials, due to their inherently heterogeneous microstructures and anisotropic behavior, present significant challenges for accurate modeling and simulation. Bridging the gap between microstructural characteristics and macroscopic performance requires sophisticated computational tools, many of which have been extensively validated in academic and research environments. Despite these advancements, industrial adoption of such tools remains limited. The underlying reasons include both the high level of expertise required to produce reliable results and the lack of seamless integration into broader digital engineering workflows, such as Digital Twins or real-time process control.

To address these challenges, we present CompositesAI, an LLM-powered expert system developed at Purdue University's Composites Manufacturing and Simulation Center (CMSC). This system leverages recent advances in generative AI—particularly large language models (LLMs)—to democratize access to composite modeling tools and lower the technical barriers traditionally associated with them. CompositesAI is built upon six decades of composites knowledge and refined through contributions from global experts. It integrates conversational AI with backend simulation tools, allowing users to receive real-time, expert-verified responses and initiate simulation workflows through natural language prompts.

The system addresses two core limitations in current industrial practice:

Expertise Barrier: Composite simulation tools typically require in-depth domain knowledge, often at the PhD level. CompositesAI translates complex modeling tasks into natural language, enabling broader accessibility without sacrificing technical rigor.

Integration Barrier: Most modeling tools exist in silos, disconnected from product design, manufacturing, and lifecycle management platforms. CompositesAI overcomes this by converting traditional simulation codes into API-based services, enabling them to be controlled through function-calling by the LLM and integrated into larger digital workflows.

To demonstrate the utility of CompositesAI, four representative use cases are presented:

Shear Moduli Comparison: Investigation of the relative magnitudes of transverse and longitudinal shear moduli, showcasing the effect of anisotropy in unidirectional composites.

Woven Composite Optimization: Deployment of a neural network model to optimize thermal conductivity based on fiber volume fraction, yarn spacing, weave architecture, and other parameters.

Multiscale Modeling of Textile Composites: Automation of meshing and analysis tasks involving Gmsh, TexGen, and SwiftComp, allowing non-experts to execute high-fidelity multiscale simulations.

Integrated Simulation Workflow: End-to-end execution of a modeling task via API-based automation, illustrating how CompositesAI orchestrates conventional simulation tools without user interaction with source code or graphical user interfaces.

A key enabler of this capability is the reformulation of existing user manuals and tutorials into AI-friendly formats. These restructured resources are used in retrieval-augmented generation (RAG) to improve the model’s reasoning ability and to provide traceable, source-backed guidance. Function calling is employed to trigger simulations and validate AI-generated insights with deterministic outputs, closing the loop between reasoning and computation.

In conclusion, CompositesAI exemplifies a new paradigm in engineering simulation: a tightly integrated, AI-augmented expert system that reduces reliance on domain-specific knowledge and streamlines the modeling process. It has the potential to significantly accelerate the adoption of advanced composites simulation in industry by making powerful tools accessible, interpretable, and actionable for a broader range of users.

Presenting Author: Wenbin Yu Purdue

Presenting Author Biography: Dr. Wenbin Yu is the Milton Clauser Professor of Aeronautics and Astronautics at Purdue University. He also serves as the Director for the Composites Design and Manufacturing HUB (cdmHUB.org) and the Chief Technology Officer of AnalySwift LLC. His areas of expertise include micromechanics and structural mechanics, with a focus on anisotropic and heterogeneous materials and structures.
Dr. Yu has an extensive research portfolio, having authored one book and over 130 journal papers. He has also developed ten computer codes, widely used by tens of thousands across government labs (ARL, NASA, NREL, Sandia National Lab, etc.), universities (Georgia Tech, Texas A&M, UCLA, Penn State, Concordia, Seoul National Univ., Tsinghua Univ., etc.), research institutes (Korea Aerospace Research Institute, etc.), and industry leaders (Boeing, Lockheed Martin, United Technology, Spirit AeroSystems, Siemens, Vestas, Samsung, etc.). Dr. Yu has mentored 21 PhD students and 8 MS students who completed theses under his supervision. His research has been funded by federal agencies (NSF, DoE, US Army, ARO, AFRL, NASA, AFOSR, etc.) as well as private industry.
Dr. Yu is a Fellow of both ASME and ASC, and an Associate Fellow of AIAA. Additionally, he serves on the editorial boards of eight international journals, including the AIAA Journal, Thin-Walled Structures, and Mechanics of Advanced Materials and Structures Journal. He founded the ASME Aerospace Structures, Structural Dynamics, and Materials (SSDM) Conference in 2023 and served as the general chair for SSDM 2023 and 2024.

Authors: