Airfoil optimization using a machine learning-based optimization algorithm

Abstract

For the design of wind turbines, airfoil optimization is widely required as the operation efficiency of wind turbines is closely dependent on the airfoil aerodynamic performance, where the accuracy of the optimization method is of great importance. In this paper, a machine learning-based optimization algorithm is proposed to improve the airfoil performance. A low-speed airfoil of NACA0012 is selected as the original airfoil for the optimization. The class-shape-transformation (CST) method is used to construct the geometry for the airfoil, and the aerodynamic performance is calculated using the panel code XFOIL. The validation of the simulation is carried out by comparing the predicted aerodynamic performance with the experimental data. In order to improve aerodynamic performance of the airfoil, maximizing the lift-to-drag ratio is selected as the main objective of the optimization problem while maintaining the lift coefficient not smaller than the original values. The results show that the present machine learning-based algorithm has fairly good convergence, and can obtain much higher lift-to-drag ratio for the optimized airfoil compared with the original one. Compared with the traditional genetic algorithm method, the machine learning based method can achieve much better aerodynamic performance and much shorter simulation time for the same airfoil optimization problem. In the future, the machine learning-based method is very promising for fluid machinery design optimization.

A Review of Intelligent Airfoil Aerodynamic Optimization Methods Based on Data-Driven Advanced Models

“…Compared with genetic algorithm, the proposed method reduced the number of function evaluations needed to reach the global optimum, and thereby reduced time-todesign by 80%. Song et al [129] presented a machine learning-based algorithm that can achieve much better aerodynamic performance and much shorter simulation time for the same airfoil optimization problem compared with the traditional genetic algorithm method.…”

A review of Deep Learning Privacy, Security and Defenses

“…1, an adversarial attack on a DL system is illustrated, which is proposed to detect and classify Alzheimer's brain disease into categories such as Nondemented, Mild-demented, Very Mild-demented, and Moderate-demented. Attacks for classification • Box-constrained LBFGS L-BFGS [69] stands for Limited-memory Broyden-Fletcher-Goldfarb-Shanno algorithm. It's an optimization algorithm in the family of quasi-Newton methods that approximates the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm using a limited amount of computer memory.…”

“…Fig.1: Adversarial attack example 1.Attacks for classification • Box-constrained LBFGS L-BFGS[69] stands for Limited-memory Broyden-Fletcher-Goldfarb-Shanno algorithm. It's an optimization algorithm in the family of quasi-Newton methods that approximates the Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm using a limited amount of computer memory.…”

Machine Learning Assisted Prediction of Airfoil Lift-to-Drag Characteristics for Mars Helicopter

“…Due to the high costs of experiments and the immense computational demands of simulations, machine learning methods are increasingly used in airfoil optimization to enhance aerodynamic performance. Song et al [22] introduced a machine learning optimization algorithm to improve airfoil performance, validating the predicted aerodynamic performance with experimental data. The findings demonstrate that machine learning is more computationally efficient than conventional genetic algorithms for optimizing lift-todrag characteristics.…”