AI-native generative design for structural engineering

The brief

Sector: Structural engineering. Need: replace manual topology exploration with a generative AI pipeline that proposes structurally valid component geometries within client-specified material and load constraints, validated in-loop by FEA simulation.

Workstream decomposition

• WS-1: Constraint formulation. Formalise the client's material, load, and manufacturing constraints as machine-readable specifications.
• WS-2: Generative model. Train a conditional diffusion model on the client's historical design corpus with physics-informed loss terms.
• WS-3: FEA validation loop. Integrate the generative pipeline with an automated FEA solver for in-loop structural validation.

Deliverable shape

• Generative design pipeline with automated FEA-in-the-loop validation
• Training corpus curation methodology
• Research dossier: 31 source rows

Outcomes

The pipeline reduced topology exploration time by 60% versus the client's manual process, while maintaining structural validity in 94% of generated candidates (vs. ~40% for unconstrained generation).