Joint Research Conference
June 24-26, 2014
In most engineering design problems constraints tend to drive optimal designs towards limits on performance and geometry. Therefore, the global optimum tends to be a tight solution, lying on the boundary of many active constraints. When using surrogates, locating a tight solution relies on an accurate representation of the constraint boundaries. This can be difficult to achieve especially in the presence of complex, multiple constraints or when the constraints are more complex or harder to model than the objective. In these cases it becomes sensible to only construct surrogates for active constraints (if possible, all inactive ones are ignored) and attempt to reduce the predicted error explicitly along the constraint boundaries. Here we investigate several formulations to enhance improvement criteria, encourag-ing updates to be selected along the constraint boundaries whilst efficiently converging towards the feasible global optimum.