Multiobjective Optimization of Rocket Engine Pumps Using Evolutionary Algorithm

A design optimization method for turbopumps of cryogenic rocket engines has been developed. Multiobjective Evolutionary Algorithm (MOEA) is used for multiobjective pump design optimizations. Performances of design candidates are evaluated by using the meanline pump flow modeling method based on the Euler turbine equation coupled with empirical correlations for rotor efficiency. To demonstrate the feasibility of the present approach, a single stage centrifugal pump design and multistage pump design optimizations are presented. In both cases, the present method obtains very reasonable Pareto-optimal solutions that include some designs outperforming the original design in total head while reducing input power by 1%. Detailed observation of the design results also reveals some important design criteria for turbopumps in cryogenic rocket engines. These results demonstrate the feasibility of the EAbased design optimization method in this field. INTRODUCTION While the budget for space development programs has drastically shrunk in most countries, recent and future space missions increasingly demand high performance and reliable rocket engine systems and components, such as turbopumps. Progress in computational fluid dynamics (CFD) methods and development of powerful computational facilities have contributed to the reduction in required cost and time to develop advanced turbopump designs. The design process still largely depends on experienced designers. Therefore, numerical design methods coupled with CFD, which are capable of efficiently developing advanced turbopump designs, can greatly reduce such dependency. Among numerical optimization algorithms, gradientbased methods are long-standing and most widely used approaches. These methods use the gradient of an objective function with respect to changes in design variables to calculate a search direction using steepest descent, conjugate gradient, quasi Newton techniques, or adjoint formulations. The solution obtained by these methods will be a global optimum, only if the objective and constraints are differentiable and convex. Unfortunately, the distribution of an objective function of real-world design problems is usually multimodal and one could only hope for a local optimum neighboring the initial design point. Therefore, to determine the global optimum, one must optimize from a number of initial points and check for consistency in the optima obtained. In this sense, the gradient-based methods are not robust. Evolutionary Algorithms (EAs, for example, see [5]) are emerging design optimization algorithms modeled on the mechanism of natural evolution. EAs search from multiple points, instead of moving from a single point. In addition, they require no derivatives or gradients of the objective function. These features lead to robustness and simplicity in coupling with any evaluation codes. Parallel efficiency also becomes very high by using a simple master-slave concept for function evaluations, if such evaluations consume most of CPU time. Design optimization using CFD is a typical case. Application of EAs to multiobjective design problems is also straightforward because EAs maintain a population of design candidates in parallel. Due to these advantages, EAs are unique and attractive approach to real-world design optimization problems. Recently, EAs have been successfully applied to aerospace design optimization problems.