Combined Acceleration Methods for Solid Rocket Motor Grain Burnback Simulation Based on the Level Set Method

Abstract

A detailed study of a set of combined acceleration methods is presented with the objective of accelerating the solid rocket motor grain burnback simulation based on the level set method. Relevant methods were improved by making use of unique characteristics of the grains, and graphical processing unit (GPU) parallelization is utilized to perform the computationally intensive operations. The presented flow traced the expansion of burning surfaces, and then Boolean operations were applied on the resulting surfaces to extract various geometric metrics. The initial signed distance field was built by an improved distance field generating method, and a highly optimized GPU kernel was used for estimating the gradient required by the level set method. An innovative Boolean operation method, thousands of times faster than ordinary ones, was ultimately proposed. Performance tests show that the overall speedup was close to 15 on desktop-class hardware, simulation results were proven to converge to analytical results, and the error boundary was 0.25%.

Precise Design of Solid Rocket Motor Heat Insulation Layer Thickness under Nonuniform Dynamic Burning Rate

“…Both Havoc and DiFi rely on the Voronoi diagram [19] to query the minimum distance. In this article, only the points on H are of our concern, so there are fewer points to process compared to full burnback simulations [10,12]. Our tests show that the AABB tree method outperforms the Voronoi-type methods when the number of points to be queried is relatively small.…”

“…where D x , B is the length of the shortest line segment connecting B to x . There have been some highperformance minimum distance algorithms, including the Havoc algorithm [16] employed in [10], the DiFi algorithm [17] employed in [12], and the Axis-Aligned Bounding Box (AABB) tree method [18]. Both Havoc and DiFi rely on the Voronoi diagram [19] to query the minimum distance.…”

“…Details about the level set evolving operations can be found in [12] and are not repeated here. However, the level set evolution process implemented by [12] is performed with a fixed propagation distance step.…”

“…Earlier studies about the burnback model can be divided into two categories: (a) the uniform models [9,10], which assume the transient burning rate to be consistent across the entire grain, and (b) the nonuniform models [11][12][13][14], which take the spatial variation of the burning rate (caused by erosive burning or multiple types of propellants) into consideration. Generally, uniform burnback models handle most SRMs at enough accuracy, while the nonuniform ones provide better versatility at the cost of much higher computation consumption.…”

“…Generally, uniform burnback models handle most SRMs at enough accuracy, while the nonuniform ones provide better versatility at the cost of much higher computation consumption. In this article, we have developed a high-performance implementation of the uniform burnback model, while the level set method [12] is used as the nonuniform burnback. The work presented by [12] only produces burning-area data with respect to flame propagation distance, so we have enhanced the implementation to predict the exposure duration data as well.…”