Identification and classification of operating flow regimes and prediction of stall in a contra-rotating axial fan using machine learning

Prediction of stall before it occurs, or detection of stall is crucial for smooth and lasting operation of fans and compressors. In order to predict the stall, it is necessary to distinguish the operational and stall regions based on certain parameters. Also, it is important to observe the variation of those parameters as the fan transitions towards stall. Experiments were performed on a contra-rotating fan setup under clean inflow conditions, and unsteady pressure data were recorded using seven high-response sensors circumferentially arranged on the casing, near the first rotor leading edge. Windowed Fourier analysis was performed on the pressure data, to identify different regions, as the fan transits from the operational to stall region. Four statistical parameters were identified to characterise the pressure data and reduce the number of data points. K-means clustering was used on these four parameters to algorithmically mark different regions of operation. Results obtained from both the analyses are in agreement with each other, and three distinct regions have been identified. Between the no-activity and stall regions, there is a transition region that spans for a short duration of time characterised by intermittent variation of abstract parameters and excitations of Fourier frequencies. The results were validated with five datasets obtained from similar experiments at different times. All five experiments showed similar trends. Neural Network models were trained on the clustered data to predict the operating region of the machine. These models can be used to develop control systems that can prevent the stalling of the machine.