GPU Accelerated Trajectory Simulation for Large Scale Monte-Carlo and Optimization

Abstract

For large systems of systems modeling and simulation studies it is often required to perform a multitude of simulations with parameter variation to evaluate the control system strategy or overall system performance. Evaluations of these parametric studies for small systems is often trivial and may be completed on a desktop computers, whereas larger systems, with run times greater than real-time, porting to a cluster is often necessary. With the improvement of General Purpose Graphics Processing Units (GPGPU) computing capabilities, it is more tractable to port models to GPU kernels for parallel execution. This paper demonstrates the use of graphics processing units for Monte Carlo trajectory analysis with the intent of optimizing flight trajectories and improving guidance laws. A variety of approaches are attempted to port a legacy simulation for a guided mortar application to a code base which may be run on a GPGPU with the objective to minimize the introduction of error and improve workflow for the systems engineer. Historically, this process is time intensive and requires deep understanding of the implementation details for the hardware. The approach is to minimize the human translation of code from one language to another (Ex: Simulink to NVidia CUDA) and utilize modern methods and libraries to perform this translation and optimization. The paper will describe the integration of high performance codes and compilers to demonstrate large scale simulation and speedup of trajectory simulations.