Engineering Case Studies

Numerical study

The question: Before designing a heat store, check how material assumptions and calculation resolution affect a predicted heating cycle.
The work: Model a 40 × 30 × 10 mm aluminium-silicon sample, compare calculation grids and repeat the calculation with shorter time steps.
The result: The selected 33-plane grid differs from the finer 65-plane grid by 1.4207 °C at the centre, below the declared 3 °C limit. The coarse 17-plane grid is rejected.
What it does not prove: This compares calculations, not measured alloy performance. The model excludes liquid flow, the container and imperfect thermal contact. The original screening record remains 219 of 220 checks passed.

Software benchmark

The question: A generated engineering part needs a checked calculation, not just a convincing 3D shape.
The work: Use the computer-aided design and stress workflow to calculate the first vibration frequency of a steel cantilever: a beam held at one end.
The result: The published demonstration reports a first vibration frequency of 204.0 Hz against 203.8 Hz from beam theory.
What it does not prove: One numerical benchmark is not a universal accuracy guarantee or certification of a customer part. Loads, materials, connections and calculation resolution need separate checks for each design.

Fixed-input AI replay

The question: Show what a trained AI model can reproduce, and where its predictions still differ from the reference calculation.
The work: Replay an existing trained greenhouse model on a fixed, short-step hot-weather case and compare it with the saved simulation reference.
The result: The recorded hot-case check gives a mean temperature difference of 0.277334 °C and a worst-point difference of 6.583515 °C on the native calculation grid.
What it does not prove: This is a short replay against a numerical reference, not physical greenhouse or crop validation. It does not establish arbitrary fan/shade response, long heat-wave predictions or plant-growth accuracy.