Evaluation of Artificial Neural Networks Technique for Calibration of Five-Hole Probe Measurement

Abstract

In the present study, the Artificial Neural Networks (ANN) technique was implemented to predict the flow parameters of a Five-Hole Probe (FHP). The experimental data were obtained from a subsonic open jet wind tunnel at a speed increased from 0 to 1180 rpm in increments of 200 rpm. The ANN approach is carried out in stages, starting with the method of selecting training data and validation, then increasing the number of neurons, varying the correlation between the activation function and the optimizer, and finally finding the optimal number of hidden layers. In the ANN approach, the mean absolute errors of 0.2705, 0.3326, and 1.0748 were achieved for estimating angle α which represents the angle of attack, angle β which represents the angle of sideslip, and speed, respectively. At the end of this study, the results were compared with the rational function approach. It was concluded that the ANN approach was more accurate compared to the rational function based on statistical parameters such as mean absolute error, max absolute error, and coefficient of determination (r2).