Grid Smoothing Research Articles

On the structure and extinction of interacting lean methane/air premixed flames

Lean methane/air premixed flames are studied numerically, using a detailed chemical model of 74 reaction steps with 28 species, to investigate the flame interaction between two stretched premixed flames with unequal intensities in a counterflow. The finite difference method, time integration and modified Newton iteration are used, and adaptive grid technique and grid smoothing have been employed to adjust the grid system according to the spatial steepness of the solution profiles. The flame structure of the methane premixed flames has four layers, consisting of the radical cutoff layer, the main heat release layer of the methane decomposition, the radical branching layer and the hydrogen and carbon monoxide oxidation layer. Two modes of flame interaction, the strong and weak interactions depending on the differences in the flame intensities, and their respective behaviors are identified. It has been found that heat loss and incomplete reaction are responsible for the flame extinctions in the weak and strong interaction regimes, respectively. The calculated extinction boundaries are in good agreement with the experimental results.

Parallel unstructured grid generation

A parallel unstructured grid generation algorithm is presented and implemented on the INTEL hypercube. Different processor hierarchies are discussed, and the appropriate hierarchies for mesh generation and mesh smoothing are selected. A domain-splitting algorithm for unstructured grids, which tries to minimize the surface-to-volume ratio of each subdomain, is described. This splitting algorithm is employed both for grid generation and grid smoothing. Results obtained on the INTEL hypercube demonstrate the effectiveness of the algorithms developed.

Partitioning and preconditioning of finite element matrices on the DAP

A method for solving linear finite element equations is proposed for highly parallel SIMD architectures with limited local memory. The method is based on a fine domain decomposition that can be conveniently mapped onto the processor array. Then the equations are structured using a form of one-way dissection and the equations solved by an iteration that combines conjugate gradient relaxation with coarse grid smoothing. Numerical experiments using the 64 × 64 DAP are presented.