Subsonic Wind Tunnel Wall Interference Correction Method Using Data Mining and Multi-Layer Perceptron

This study proposes a novel method for correcting subsonic wind tunnel wall interference based on data-driven approaches, which achieves parameter efficiency and can be applied without any additional measurements. The proposed method involves selecting important parameters and correcting wall interference using a multi-layer perceptron (MLP). The prediction concepts are simplified from the conventional correction method into two methods: the Hybrid method, which predicts flights in wind tunnel state, and the Direct method, which predicts flights in free air state. The Hybrid method predicts the increment of the angle of attack ( \(\Delta \alpha \) $\mathrm{\Delta }\alpha$ ) and corrected aerodynamic coefficient ( \({C}_{{{\text{L}}}_{{\text{c}}}}\) $C_{{\text{L}}_{\text{c}}}$ ), while the Direct method predicts the free air aerodynamic coefficient ( \({C}_{{{\text{L}}}_{{\text{FA}}}}\) $C_{{\text{L}}_{\text{FA}}}$ ). Three-dimensional wind tunnel state models and three-dimensional free air state models are built using a panel method to construct the database for the MLP. Data mining methods such as the Pearson correlation coefficient (PCC), a self-organizing map (SOM), and a parallel coordinate plot (PCP) are used for parameter sensitivity analysis. As a result, 13 model parameters for the Hybrid method and 11 model parameters for the Direct method are selected. The MLP-based correction results show that the proposed wall interference correction method achieves results much closer to the free air state and true value of \({C}_{{{\text{L}}}_{{\text{c}}}}\) $C_{{\text{L}}_{\text{c}}}$ compared to classical methods.