Jinxin Chen Builds AI System to Design Concrete Structures
Reinforced concrete structures are physically heavy — and designing them can be just as weighty, burdened by a complex, code-heavy, time-consuming process that’s still prone to human error.
But Jinxin Chen ’27, a civil engineering Ph.D. candidate, and his advisor, Yi Bao, associate professor in the Department of Civil, Environmental and Ocean Engineering, have automated the process with artificial intelligence. The result? A system that quickly delivers accurate, code-compliant, easily interpreted designs.
The project tackles long-standing challenges in structural engineering design, including slow and tedious manual methods, traditional software that’s difficult to learn, and a lack of transparency with machine learning models. Even promising new artificial intelligence tools such as large language models (LLMs) can deliver misleading, unreliable answers.
Published in Automation in Construction, one of the field’s top journals, the research paper introduces a framework powered by AI LLMs to automate the design of reinforced concrete beams. Unlike traditional machine learning models that accept data and deliver “black-box” answers without showing the process behind the solution, this new system enables engineers to use plain language and receive step-by-step explanations throughout the process. The LLM acts as the brain that orchestrates the workflow to efficiently deliver accurate, code-compliant solutions.
“My inspiration was to democratize access to AI for structural engineers,” Chen explained. “Most engineers don’t trust black-box models. They want to see why the AI made a decision and check it against design codes. This project aims to bridge that gap.”
Engineering with expert code — and everyday conversation
After exploring the limitations of existing software, Chen began using LLMs to build an architecture that breaks the work into smaller steps, then calls upon a team of specialized processing agents to handle individual tasks throughout the procedure.
“It’s the best of both worlds,” Chen explained, “It offers the rigorous accuracy of computational tools and the LLM's ability to present a clear, auditable narrative of the design process. Rather than replacing engineers, it acts as a powerful computational partner.”
Pre-approved tools handle all the engineering calculations to avoid the risk of LLM computational hallucinations that can present fictional output as apparent fact.
“The LLM never guesses,” Chen added. “It simply coordinates the right tools and translates the results into clear, verifiable language.”
An intuitive graphical user interface (GUI) allows the engineers to submit design problems, upload design plans and ask questions with everyday phrasing, no coding expertise required.
“The GUI makes the AI feel like a trusted colleague,” Chen said. “You can ask it to change the beam width or explain which code it followed — and it will respond in plain language.”
To validate the system, Chen designed and evaluated 30 reinforced concrete beams, then compared the results to those from SAP2000, an industry-standard software for structural analysis. The impressive results included a 97% match with SAP2000, a 90% time reduction compared to manual design and clear, understandable reasoning for every design choice.
“The most surprising finding was just how reliable our system proved to be,” he said. “Its ability to consistently match the SAP2000 output was a powerful validation of our approach.”
This success demonstrates that the system is reliable, relatable and ready to do more.
“Right now, it’s like a workshop stocked with tools for concrete beam design,” Chen said. “Next, we’ll add tools for steel columns, timber frames and more to scale it for more complex structural design challenges. As we add more tools, and as the LLM gets smarter, the range and complexity of projects we can tackle will grow exponentially.”