Parallel Modular Dynamic Process Simulation

To meet the needs of plant wide dynamic process simulation of today's complex, highly interconnected chemical production plants, parallelizable numerical methods using divide and conquer strategies are considered. The large systems of differential algebraic equations (DAE's) arising from an unit oriented modular modeling of chemical and physical processes in a chemical plant are partitioned into blocks. Using backward differentiation formulas (BDF), a partitioned system of nonlinear equations has to be solved at each discretization point of time. By formally extending these systems, block-structured Newton-type methods are applied for their solution. These methods enable a coarse grain parallelization and imply an adaptive relaxation decoupling between blocks. The resulting linear subsystems with sparse and unsymmetric coefficient matrices are solved with a Gaussian elimination method using pseudo code techniques for an efficient multiple refactorization and solution. Results from dynamic simulation runs for industrial distillation plants on parallel computers are given