machine learning for feature identification and characterization of fluid flow around a cylinder using airborne acoustic signature

This study employs machine learning (random forest, adaptive boosting, and multilayer perceptron) to identify flow features around a cylinder using airborne acoustic signatures. acoustic data, including sound pressure levels, are derived from numerical simulations. the studied machine learning models effectively distinguish between different flow states, classified based on the values of reynolds number. furthermore, this study investigates the impact of observer position on the accuracy of machine learning models for flow differentiation. the results show that random forest detects 7.5° rotations of observation point with 66.63% accuracy at re = 30000, outperforming visual methods. notably, the detection performance of the models remains consistent regardless of the observer's distance from the sound source, in both the near and far fields. it is worth noting that this study integrates numerical simulations with practical applications, such as wind turbine noise monitoring, where deviations in acoustic sensors can impact the performance of machine learning classification.