Can AI Be Trusted with Structural Analysis? Lessons from Beam Calculations
Artificial intelligence can produce an engineering answer in seconds. It can explain a method, write equations, and present a calculation that looks convincing. For a structural engineer, however, the decisive question is whether that answer accurately represents how the structure behaves.
A June 2026 paper by Mohammad Mamon Fayiz Hamdan, An Assessment of Artificial Intelligence Capabilities in Structural Analysis: A Comparative Study of Determinate and Indeterminate Beams, examines this question through beam analysis.
What the study tested
The study evaluated ChatGPT, DeepSeek, Gemini, and Perplexity using beam problems with different levels of static indeterminacy and complexity. Their answers were compared with classical analytical solutions using the force and slope-deflection methods, alongside SAP2000 results.
The reported pattern was clear: the tested AI systems produced reasonably accurate answers for determinate beams and some systems with low indeterminacy, while discrepancies increased for more complex indeterminate systems. The author concluded that the evaluated systems were insufficiently reliable for complex structural analysis.[1]
Why beam complexity matters
A statically determinate beam can be analysed using equilibrium equations alone. An indeterminate beam requires additional conditions describing how its deformations fit together.
Consider a beam resting on several supports. Calculating reactions requires more than balancing the applied loads. The solution must also respect the displacement conditions at the supports and the continuity of the beam.
This distinction explains why checking only the total reactions is insufficient. A set of reactions might balance the external load yet still fail to represent the beam's deformation correctly.
The engineering lesson is straightforward: a plausible calculation must satisfy the physical conditions of the entire structural model.
A polished answer still needs checking
Clear formatting, technical vocabulary, and a long derivation can make an answer feel authoritative. None of these features demonstrates that its assumptions are correct.
When reviewing any calculation, engineers should ask:
Are the supports, releases, dimensions, and loads represented correctly?
Do the reactions satisfy force and moment equilibrium?
Does the solution satisfy the required displacement and rotation conditions?
Are units and sign conventions consistent?
Do the results agree with an independent calculation?
These questions apply equally to hand calculations, software models, and AI-generated answers. Structural software also depends on correct modelling and thoughtful interpretation.
What this means for using AI
My practical interpretation is to give AI a supporting role in an engineering workflow. It can be asked to explain terminology, organise a calculation, or propose steps for review. Any numerical output should remain provisional until independently checked.
For example, asking AI to explain why a continuous beam requires compatibility conditions can support learning. Accepting its support moments as design values without verification is a much stronger decision.
A useful workflow therefore starts with an engineer-defined model, proceeds through an established analysis method, and ends with independent checks. AI assistance should fit within that process.
Keep the conclusion within the evidence
The paper's findings concern the systems and beam problems evaluated. They should not be treated as proof that every AI application, specialised engineering tool, or future model will perform identically.
Likewise, good performance on a simple example does not establish reliability across a wider range of structures. Confidence should come from validation against the intended task.
For engineers exploring AI, this study offers a valuable reminder: speed and fluency are useful, but structural analysis requires demonstrable correctness. The strongest use of a new tool begins with understanding its limits and verifying its outputs.
Reference
[1] Hamdan, M. M. F. (2026). An Assessment of Artificial Intelligence Capabilities in Structural Analysis: A Comparative Study of Determinate and Indeterminate Beams. International Research Journal of Engineering and Technology (IRJET), 13(6).
Comments
Post a Comment