A decomposition algorithm for large scale surrogate models

In engineering designs, it is necessary to deal with large numbers of parameters and constrains. However, finite computer resources do not allow all the parameters to be taken into consideration. In recent years, surrogate models were employed to increase the computational efficiency. This approach not only increases the computational speed, but it also provides the means for global optimization methods. Unfortunately, the surrogate models have practical limitations. It is very difficult, if not-impossible, to construct a surrogate model with more than a very limited number of parameters. Since the construction of surrogate models requires increasing number of samples with increasing number of parameters, this makes the application of the method infeasible. To overcome these limitations, a decomposition approach is proposed for structural designs having a simple geometrical pattern with repetition. The application of the algorithm to a stiffened panel structure is presented.