Courtyard School
A canopy of 212 perforated panels, generated for the hours when children are outside.
I’m Noor Haddad, a computational architect. I use generative design and climate models to shape shaded courtyards and cooler buildings, and then I go back in August with a thermal camera to check that they work.
For each site I generate thousands of canopy and massing options, score every one with a surrogate model, and keep the few that shade the most hours with the least material.
The surrogate is a neural network trained on 52,000 full climate simulations. A full simulation of sun, radiation and airflow takes about three hours; the surrogate gives its comfort estimate in 40 milliseconds, so the search can be wide. I write the search in Python over Rhino geometry, and the finalists always get the full simulation again.
I choose from the shortlist with the client; the model only ranks. And the surrogate knows Dubai summers: on a humid coastal site it is recalibrated with new simulations before anyone trusts it.
A canopy of 212 perforated panels, generated for the hours when children are outside.
Thirty-six homes along shaded lanes. The massing was chosen from 4,000 generated options for the way the wind comes down the valley.
An open map of shade, hour by hour, at 1,200 bus stops, so the worst ones get canopies first.
Writing and talks
One site, four weeks. You get three massing and shade options with modelled comfort hours, cooling load and a cost range, drawn and explained in one review at this table.