GPU Parallel Computing With Application To Aircraft Design

Modern aircraft design is a typical multidisciplinary design optimization process. There exist complex and time-consuming simulation analysis models, such as the structural analysis and aerodynamic analysis based on the finite element method, and the coupling between disciplines, such as the structure and aerodynamic coupling. The large amount of computation is a notable problem in the optimization design of modern aircraft. It is well known that the efficiency of the whole optimization design is greatly determined by the efficiency of the optimization algorithm. As a global optimization algorithm with good performances, particle swarm optimization(PSO) has gain wide attention in the aircraft design due to its simple structure and global search ability. The global optimal solution is obatined by PSO based on the random search of the whole design space. It costs huge computational time, thus its application in modern aircraft design is greatly limited. It is necessary to improve the computational speed of PSO algorithm.At presnet, it is hardly to improve the comutational speed of PSO from the algorithm itself. In recent years, Graphics Processing Unit(GPU) has been widely used in the field of general computing, such as General-Purpose Computing on Graphics Processing Units(GPGPU), due to its good floating-point computing capability, high concurrency degree and relatively cheap features. It has great potential in scientific research and engineering. In 2007, NVIDIA launched a parallel computing platform, named as Compute Unified Device Architecture(CUDA), which greatly promoted the application of GPU. Now, GPU is widely used in fluid mechanics, finite element simulation, molecular dynamics and other fields. To address the intensive computational cost of PSO, a parallel PSO algorithm based on GPU using the CUDA platform is developed in this thesis. Make the best of the basic architecture of parallel computing of PSO, fine-grained parallelization is implimented with GPU. The velocity and position initializing, fitness evalutation and updating of speed and position of each particle are implimented parallely to acheive full speed-up of the PSO algorithm and greatly reduce the computational time.As an important discipline in the overall optimization design of the aircraft, trajectory optimization has important influence on the overall design of the aircraft. To improve the efficiency of trajectory optimization, the widely used direct shooting method is employed to solve the trajectory optimization problem in this thesis. Firstly, the optimal control problem is transcried into a nonlinear programming problem(NLP), and then the parallel PSO algorithm developed above is applied to solve the NLP, during which the flight distance of the trajectory simulation subroutine is used as the fitness function. This method greatly reduces the computational time, and provides a new trajectory optimization method for engineering and scientific research.Based on the finite element method, the structure analysis(Finite Element Analysis, FEA) and aerodynamic analysis(Computational Fluid Dynamics, CFD) are widely used in the modern aircraft design. However, the highly intensive computatioal cost is always a notable problem. Surrogate model is often used to replace the FEA and CFD. But surrogate model can induce errors. Meanwhil, in order to guarantee the accuracy of the surrogate model, a large number of sample points are required, which still results in high computational cost. Therefore, from the parallel computing point of view, GPU is propsoed to accelerate the FEA and CFD process. The comutational efficiency is improved greatly, and the error of the surrogate model is avoided as well.Through simulation analysis of many examples and applications to aircraft design, the proposed acceleration strategies based on GPU parallel computing are demenstrated to be effective in this work. Compared to the traditional CPU(Processing Unit Central) serial computing, considerable speedup is achieved by using GPU. Overall, the GPU parallel computing has great potential and wide applicability in aircraft design.