Generating and testing options.
WHAT GENERATIVE DESIGN DOES
Given constraints and objectives, produces many candidate designs.
The engineer evaluates and selects.
WHERE IT WORKS
Optimisation problems with well-defined constraints: weight, material, load paths.
WHAT IT PRODUCES
Options, some of which are unbuildable or impractical.
That is expected. Selection is the engineer's job.
FOR SIMULATION
Machine learning can approximate expensive simulations, producing faster results.
The approximation is valid within the range it was trained on and unreliable outside it.
THE EXTRAPOLATION WARNING
A surrogate model asked about conditions outside its training range produces a confident wrong answer.
Know the valid range.
WHAT MUST STILL HAPPEN
Verification by proper analysis for anything that will be built.
WHAT TO DOCUMENT
Which tools were used, within what limits, and what verification was performed.
THE PRACTICAL POSITION
Useful for exploring options quickly. Not a substitute for analysis.
