Highlights

Objective and constraints typed as maths expressions over bounded design variables
Local and global search: multistart, differential evolution, particle swarm and Nelder-Mead
Design of experiments with Sobol and correlation sensitivity, tornado chart and scatter matrix
Polynomial, RBF and Kriging metamodels ranked by cross-validation, and surrogate-assisted search
Pareto fronts by NSGA-II, weighted sum or ε-constraint, with hypervolume

A question it answers

Which of my twelve design variables actually drive the result, and what is the lightest design that still satisfies the stress and deflection limits?

What you put in

Design variables with lower and upper bounds; the objective written as a mathematical expression; equality and inequality constraints as expressions; choice of local or global search method; sampling plan and sample count for exploration; a second objective when a trade-off study is wanted.

What you get

Optimum design vector with constraint feasibility and a convergence diagnosis; variance-based and correlation-based sensitivity indices with a tornado chart and scatter matrix; surrogate models ranked by cross-validated predictive quality with an interactive response-surface map; surrogate-assisted optimisation history reported in true model evaluations; Pareto front in objective space with a hypervolume convergence trace.