SHD Sim Studies
Optimisation (derivative-free)
Genetic, particle swarm, Nelder-Mead over the sweep machinery
NLopt / pagmo
In development.
This module is phase 4 on paper — but it needs no new backend. The solver behind it is chosen and its capability is known. SHD Sim is available today.
What it is
The same sweep machinery with an algorithm choosing the next point instead of a grid. Genetic algorithms, particle swarm and Nelder-Mead need no gradients, so they work on any case that produces a number, at the cost of many more runs than a gradient method would need.
What this analysis needs from you
- An objective function computed from each run
- Constraints stated explicitly, not implied
- A run budget, since these methods will use everything you give them
Typical uses
- Shape and sizing optimisation within existing parameters
- Multi-objective trade-offs, giving a Pareto front rather than one answer
- Tuning a design against a constraint set
- Problems where no adjoint exists
Industries
Sectors where this analysis is routinely asked for.
- Product development
- Turbomachinery
- Structures
- Energy
What comes out
- Optimised parameter values
- Convergence history
- Pareto fronts for competing objectives
- Constraint activity report
Backend
NLopt / pagmo
Chosen for this analysis type, with its capability verified against the module plan — not decided later.
Status
Phase 4 on paper — but it needs no new backend. Telling us you need this analysis moves it up the order — modules are sequenced by who is waiting for them.