Prediction of Drag Reduction Effect of Pulsating Control in Turbulent Pipe Flow by Machine Learning

Abstract

Abstract It has been widely expected that the pulsating control can reduce friction drag in various fluid systems. In order to maximize its effect, a prediction tool of drag reduction using pulsating control is required. The present study aims at the prediction of the drag reduction rate by machine learning. Multilayer perceptron (MLP) was applied as the machine learning method. Water was used as the working fluid. First, an automatic measurement system was constructed and drag reduction effect was evaluated by an experiment with various pulsation waveforms. The flow pulsation was generated by giving periodical acceleration and deceleration by a centrifugal pump in a closed circulation system. The bulk Reynolds number Reb ranges between 3400 and 3800. Next, the experiments were performed with over 5000 kinds of waveforms to make training and validation data for MLP. Within the data, the maximum drag reduction rate of 38.6% was observed. The friction coefficient Cf decreased during the acceleration period and increased during deceleration period. Finally, the drag reduction rate was predicted in three cases with different input parameters of MLP. The relationship between pulsation waveforms and the drag reduction effect was successfully predicted.