Training artificial neural networks to control an active grid for shaping inflow profiles

Abstract

In wind tunnel experiments, one approach to reproducing the representative flow environment around test bodies is the use of obstacles, which necessitates longer test sections. As an alternative, an active grid system, consisting of motor-driven rods with attached winglets, was introduced by Makita (1991 Fluid Dyn. Res. 8 53–64). Since then, numerous studies have investigated the effects of synchronized and independent rod motions, the influence of input parameters on the resulting flow characteristics, and the ability to generate desired profiles through parameter adjustments. The present study explores a potential method for determining the input parameters required to produce specified mean velocity and turbulence intensity profiles. To this end, a mini wind tunnel equipped with an active grid system was designed, and various combinations of input parameters were tested to generate training data for artificial neural networks (ANNs). The results were consistent with previous findings, indicating that mean velocity profiles depend primarily on the flow restriction created by the collective winglet positions, while turbulence intensity profiles are influenced by both the restricted flow areas and the rotational speeds of the rods. The ANN successfully learned to predict input parameters for target mean velocity profiles, although achieving similar performance for turbulence intensities proved more challenging. Consequently, the turbulence intensity part of the study focuses on assessing the learning potential of ANNs for turbulence intensities and provides recommendations for future research.