The Algorithm Study And System Implementation Of The Hybrid Optimization Approach For Blade Aerodynamic Design

The study on the design methodology of turbomachinery blade with low loss, high load and good performance under full condition is one of the frontiers in the field of turbomachinery development.In this dissertation, a hybrid optimization approach for automatic design of 3D turbine blades was proposed and well implemented. The applications to the stacking design optimization and full 3D aerodynamic design optimization for an annular cascade demonstrated the validity, robustness and efficiency of this hybrid approach. The achievements include:1. The hybrid optimization approach for blade aerodynamic design optimization consists of four key steps: Design of Experiments (DOE) study for design variables, Genetic Algorithm (GA) for global exploration, Response Surface Model (RSM) for approximation, iteration of Sequential Quadratic Programming (SQP) and RSM for local refinement.2. DOE study offers a systematic and efficient means of analyzing the design space. It provides essential information for screening a large set of inputs to determine the most influential ones to be used as design variables for optimization. It also can estimate a rough optimal design based on DOE results.3. Blade design optimization problems are multimodal and discontinuous problems, the use of gradient-based numerical optimization algorithms perform inefficiently and drop into local minimum prematurely. Hence, the exploratory algorithms such as GAs are required for global exploration.4. The RSM approximation models can be efficiently used to reduce the number of exact CFD analysis, create smooth response functions and improve convergence of optimization algorithms.At the stage of RSM and SQP iterations, once the design cannot be improved based on RSM, exact analyses will be performed by 3D Navier-Stokes CFD5. simulation to update the RSM model until convergence. In this way, the accuracy and performance of the exact CFD simulations are well achieved.6. The 3D blade surface is parameterized by Non-Uniform Rational B-Splines (NURBS) and the CFD simulation is accelerated by JAVA RMI parallelization.