An approach that links nonlinear model reduction techniques with control vector parametrization-based schemes is presented, to efficiently solve dynamic constraint optimization problems arising in the context of spatially-distributed processes governed by highly-dissipative nonlinear partial-differential equations (PDEs), utilizing standard nonlinear programming techniques. The method of weighted residuals with empirical eigen-functions (obtained via Karhunen-Loève expansion) as basis functions is employed for spatial discretization together with control vector parametrization formulation for temporal discretization. The stimulus for the earlier approach is provided by the presence of low order dominant dynamics in the case of highly dissipative parabolic PDEs. Spatial discretization based on these few dominant modes (which are elegantly captured by empirical eigenfunctions) takes into account the actual spatiotemporal behavior of the PDE which cannot be captured using finite difference or finite element techniques with a small number of discretization points/elements. The proposed approach is used to compute the optimal operating profile of a metallorganic vapor-phase epitaxy process for the production of GaN thin films, with the objective to minimize the spatial nonuniformity of the deposited film across the substrate surface by adequately manipulating the spatio-temporal concentration profiles of Ga and N precursors at the reactor inlet. It is demonstrated that the reduced order optimization problem thus formulated using the proposed approach for nonlinear order reduction results in considerable savings of computational resources and is simultaneously accurate. It is demonstrated that by optimally changing the precursor concentration across the reactor inlet it is possible to reduce the thickness nonuniformity of the deposited film from a nominal 33% to 3.1%.
All Science Journal Classification (ASJC) codes
- Environmental Engineering
- Chemical Engineering(all)