Prediction of Dynamical Properties of Flow Over a Three-element Airfoil via Computationally Intelligent Architectures

Abstract

In this paper we study various computationally intelligent architectures for prediction of pressure values and velocity components of flow past a three-element airfoil. Six sensor locations are selected around the airfoil and the goal is to predict the flow behavior at the rear of the airfoil using pressure readings from the remaining five sensors. To make the problem more interesting we require the predictor to estimate the flow twenty time steps ahead of current time. Data is collected from CFD simulations of the flow and predictors are built using four different computationally intelligent architectures: Multilayer Perceptron (MLP), Adaptive Neuro Fuzzy Inference System (ANFIS), Radial Basis Function Neural Network (RBFNN), and Least Squares Support Vector Machine (LS-SVM). Levenberg-Marquardt optimization technique is utilized for parameter tuning purposes. In addition, a simple linear predictor is built as a benchmark for comparing the MLP, ANFIS, RBFNN, and LS-SVM based predictors. It is observed that MLP and ANFIS based predictors achieve the best prediction, and the performace of all predictors are superior to that of the simple linear predictor.