The development of a distributed methodlogy for process simulation with applications in pipe flow /
Cohen, Steven, 1973- · eScholarship@McGill (McGill) · 1990
The increases in complexity of industrial process simulations necessitate faster methods and more accurate models for analysis. In many cases automation of industrial plants using real-time control cannot be solved using present computer methodologies. This thesis introduces a parallel methodology that outlines the steps to design and implement a distributed continuous simulation. Implementing the system in parallel reduces the simulation time to comply with predetermined time constraints. The methodology determines the structure of the simulator, organization of the distributed tasks and management of the variable memory allocation. A dynamic model for transient, incompressible pipe flow is used to test the parallel methodology. The generalized flow equations are developed for two applications: a tank-pipe and pipe network model. The models are translated into computer algorithms using parallel techniques developed in the simulation methodology. The individual subroutines of the simulation are directly implemented on a parallel processing and a multi-tasking computer. The flow algorithms are tested on both computers to demonstrate the benefits obtained by using the parallel methodology on a multi-processor computer. Model sizes are varied to provide a performance scale comparing multi-processor results with that of a single processor.