An improved genetic algorithm and its applications to the optimisation design of an aspirated compressor profile

Abstract

SUMMARYGenetic algorithm (GA) is a widely used method for numerical optimisation owing to their good global search ability; however, their local search ability has an obvious shortcoming. To improve local search ability, this paper introduces a simplex method and combines it with a GA to form an improved genetic algorithm (IGA). In the IGA, at each generation of the original GA, high-fitness individuals are selected as vertices of a simplex, and then a one-dimensional search within the simplex is conducted to obtain the most-fit individuals while replacing the inferior ones. Typical test functions show that the IGA can effectively improve the optimisation effect over that of the original GA. To further verify the IGA's practicability, an aspirated compressor profile is optimised with profile, suction flow rate and suction flow location as coupled design parameters. The results again show that the IGA has a better optimising effect than the GA. In addition, it is also verified that coupling the profile and suction flow parameters results in a design that outperforms the uncoupled design; therefore, designing an aspirated compressor blade by arranging suction flow on a conventional blade without considering suction flow is not a good method.

Controlling the flow loss of a supersonic compressor rotor using the blade slotting method

“…The optimal design is carried out on several compressor/fan rotors, stators and stages, and the results are satisfactory. [26][27][28] The process of optimization is shown in Figure 24. The numerical optimization uses a GA; the initial blade superposition modification is used, and the modifications are obtained using the parameterization method of the Bezier curve 29 (seen in Figure 25).…”

A reinforcement learning approach to airfoil shape optimization

“…Gradient-free methods address some of these issues: they are better at finding the global optimum 2 and are well suited for complex optimization tasks (non-linear and non-convex functions) 2,7 . Within gradient-free methods, Genetic Algorithms (GA) modified for aerodynamic optimization have been developed to tackle the limitations of gradient-based methods 22,23 . The benefits of gradient-free methods come at the cost of more complex methods having higher computational costs (compared to gradient-based approaches), poor constraint handling abilities and limitations on the number of design variables handled 2 , resulting in low convergence speeds when coupling gradient-free methods with high-fidelity solvers 6,8,9 .…”