Identification of airfoil stall using airborne acoustic signature under room conditions by machine learning

Detecting airfoil stall via acoustic measurements is significantly challenging in realistic and noise-rich environments. This study explores the applicability of machine learning methods for detecting airfoil stall using far-field microphone recordings in room conditions, where background noise, reverberation, and sound reflections from surrounding surfaces present significant challenges for accurate stall identification. Based on experimental measurements in a blowdown aeroacoustic wind tunnel, three machine learning techniques are employed to identify flow state in pre- and post-stall at different angles of attack. The extreme cases of attached flow at 0 ° angle of attack and deep stall at 40 ° , for which the acoustic signal is distinguishable from noise are taken as references. These cases provide baselines because in noisy environments even such distinct states can overlap with background signals. Then, the methods are extended to intermediate angles of attack, where stall onset occurs. In the first method, data preprocessing is performed using Fourier transform. However, the method performance deteriorates as the temporal gap between training and test datasets increases. The second method mitigates this issue by applying a custom filter to smoothed spectral data, which improves accuracy but still falls short of achieving consistent classification across the entire dataset. The third method integrates both statistical and frequency-based features, enabling robust classification of pre-stall, light stall, and deep stall conditions throughout the dataset. The results highlight the potential of combining machine learning with tailored data preprocessing for reliable stall detection from acoustic airborne signals in noisy environments. • This study applies Machine Learning to detect airfoil stall via airborne acoustics in noisy, reverberant room conditions. • Three Machine Learning techniques identify flow states pre/post-stall at various angles using a blowdown aeroacoustic wind tunnel. • Results show Machine Learning with tailored preprocessing enables reliable stall detection from airborne acoustics in noisy environments.