Parallel reactor plant simulation.

Abstract

A parallel algorithm for the simulation of a nuclear power plant based on an existing serial code (TRAN-MB) has been developed, implemented, and tested on a distributed-memory parallel processor--the NCUBE/six hypercube parallel processor at the University of Michigan. In addition to the algorithm development for a distributed-memory parallel processor, adaptability to a shared-memory processor is examined using the IBM3090/600 at the Cornell National Supercomputer Facility. The underlying idea is to partition the reactor plant spatially and assign the resultant computational modules to separate processors. The parallel algorithm is based on a modified Momentum Integral Method for the primary loop and the Extended Implicit Continuous-fluid Eulerian (EICE) method for the secondary side of the steam generator, along with a movable boundary model to track the secondary-side phase change. The one-dimensional, slip flow model is formulated in a standard way using three conservation equations and the equation of state. The primary system pressure is calculated by integrating the mass conservation equation around the primary loop. The flow distribution is obtained by integrating the momentum conservation equation with incompressible flow approximation. This approach removes the unrealistic flow oscillations generated by TRAN-MB. The integration procedure results in a set of discretized equations in a summation form, which can be parallelized. The enthalpy of each cell is calculated with the EICE method, a marching method which can be implemented in parallel. The hypercube parallel processor is very convenient for implementing the parallel summation and parallel marching solutions. However special care must be given to minimize the message passing time when developing the parallel algorithm for the hypercube. The test case for the resultant parallel algorithm is the TMI-2 accident. A speedup of close to seven is obtained on the NCUBE, which is sufficiently fast to allow the simulation of the TMI-2 transient faster than real time. Test calculation results show improved agreement with the actual plant data than with TRAN-MB.