Agentic LLM-enhanced optimisation and verification frame-work for structural design (September 2026).

This paper was presented by Yinghao Cheng at the 42nd International Manufacturing Conference (IMC42), co-organized by the Irish Manufacturing Council and Queen’s University Belfast, which was held on 10 and 11 September 2026 in Belfast, Northern Ireland. The plenary sessions of the conference occurred on the 10th September at the Advanced Manufacturing Innovation Centre (AMIC), Global Point, Newtownabbey, followed by technical sessions and a tour of AMIC facilities in the afternoon.

Abstract
Structural design is a highly skilled and challenging process due to the trade-offs among performance, cost, safety, user preferences and manufacturability. Although evolutionary algorithms can generate high-quality candidate solution sets, they rely on well-defined fitness functions and constraints, and the following design verification remains labour-intensive. To address this gap, an LLM-enhanced optimisation and verification frame-work for structural design is proposed. Multiple LLM-based agents are coordinated to translate natural-language design intentions into fitness and constraint evaluation measures for evolutionary optimisation, and the framework also supports automated ver-ification of Pareto solutions against design codes. A 2D truss bridge case study demon-strates its effectiveness in reducing skill requirements, development effort and thus im-proving design efficiency.

The link to the full paper will be provided in due course.