Parallel dynamic simulation of industrial chemical processes on distributed-memory computers

Abstract

We describe here a parallel-modular approach to solve plantwide, as well as multiple plant problems, by running simulations in parallel on a distributed network of computers. Two implementations of the parallel-modular approach are presented. In the first, SPEEDUP (Aspen Technology, Inc.) serves as the master process and converges the overall simulation using simultaneous calls to SPEEDUP slave simulations for function and/or derivative evaluations, for both steady-state and dynamic simulations. In the second implementation, MUSIC (MUlti-SImuIation Configuration) combines any number of independent dynamic simulations together with a graphical-user interface to create a single client-server environment for parallel dynamic simulation. In this paper, we use MUSIC to integrate SPEEDUP and OTISS dynamic simulations, running them in parallel in a networked computer environment (MUSIC and OTISS from Aspen Technology, Inc.). The parallel-modular approach is tested using large industrial simulation case studies. For the parallel SPEEDUP environment, we present results for the steady-state simulation of three integrated sections of a petrochemical facility (>175,000 equations). Good speedup factors are achieved by exploiting parallel execution and problem decomposition. One of the plant sections is also used for dynamic simulation by decomposing it into two subsections. Numerical results are presented showing a significant improvement by using two processors. Using the MUSIC environment, we demonstrate performance on the dynamic simulation of a gas injection and processing plant (>80,000 equations). Parallel execution on two processors provides real-time performance for this integrated OTISS and SPEEDUP dynamic simulation.