Velocity Field Reconstruction in the Mixing Region of Swirl Sprays Using General Regression Neural Network

Abstract

A general regression neural network technique is proposed for design optimization of pressure-swirl injectors. Phase doppler anemometry measurements for velocity distributions are used to train the neural network. An overall optimized value for the width of the probability is determined. The velocity field in the extrapolation regime is reconstructed with an accuracy of 93%. Excellent agreement between the predicted values and the measurements is obtained. The results indicate that the capability of performing design- and optimization studies for pressure-swirl injectors with sufficient accuracy exists by applying modest amount of data in conjunction with an overall optimized value for the width of the probability.