AI system accelerates aircraft concept design using language models

Written by Copernical Team Monday, 14 July 2025 10:48

font size decrease font size increase font size
Print
Email

Tokyo, Japan (SPX) Jul 11, 2025

A research team at Beihang University has unveiled a new artificial intelligence-based method that can quickly generate conceptual aircraft designs-potentially transforming how early-stage design is done in aerospace engineering. By using large language models (LLMs) like GPT and Gemini, the team created a prototype system that delivers viable design solutions with little to no prior training ex

AI system accelerates aircraft concept design using language models
by Riko Seibo
Tokyo, Japan (SPX) Jul 08, 2025

A research team at Beihang University has unveiled a new artificial intelligence-based method that can quickly generate conceptual aircraft designs-potentially transforming how early-stage design is done in aerospace engineering. By using large language models (LLMs) like GPT and Gemini, the team created a prototype system that delivers viable design solutions with little to no prior training examples.

Traditionally, aircraft conceptual design has required expert engineers to manually iterate ideas based on experience, complex calculations, and established theory. This new approach, developed under the leadership of Professor Mingqiang Luo, aims to offload routine and repetitive design tasks to generative AI tools-freeing human designers to focus on innovation and decision-making.

The research explored how general-purpose LLMs could be adapted for engineering use. The team developed three types of design prompts, tested them in zero-shot and few-shot modes, and applied models from the GPT, Gemini, Llama, and Qwen families. Their system successfully generated aircraft concepts that were later evaluated by experts for feasibility, creativity, and usefulness.

"Across different use cases-civil or military, rapid or creative design-the language models showed strong adaptability," said Luo. "Some AI-generated concepts scored on par with human designers in expert assessments."

Model size turned out to be a major factor. Versions of the Qwen model with fewer parameters (under 7 billion) performed poorly in generating valid designs. But with larger versions-57 billion parameters and above-the AI-generated outputs closely matched those of professional engineers. Interestingly, adding even a single example to the prompt improved the performance of smaller models significantly, despite no clear correlation between the input and output designs.

Crucially, Luo clarified that the goal is not to replace engineers but to assist them. "Generative AI can handle low-value, high-repetition tasks like parameter estimation and intent parsing," he said. "This allows human experts to concentrate on innovation and core technical challenges."

Future plans include expanding the system to later stages of aircraft development, including requirements analysis and detailed design. The team envisions integrating LLMs with standard engineering platforms like SysML, CATIA, and Matlab to support full life-cycle design through AI-human collaboration.

"Our vision is an intelligent design co-pilot-an AI partner that supports engineers from start to finish," Luo concluded.

The team also includes Yao Tong, Shangqing Ren, and Zheng Zhang from Beihang University, along with Chenguang Xing and Ziliang Du from the Chinese Aeronautical Establishment.

Research Report:A rapidly structured aircraft concept design method based on generative artificial intelligence