Neural Network-Based Correction of Transducer Errors in Wind Tunnel Experiments Using Data-Driven Modeling

This paper aims to use neural networks to correct transducer errors in wind tunnel experiments. It focuses on three essential measurements: forces in X and Z directions (Fx, Fz), and moment in Y direction (My). By training a Multilayer Perceptron (MLP) using flow-on sensor outputs and associated aerodynamic signals obtained by time averaging. The model learns to extract accurate force and moment values from distorted input. The resulting model has RMSE < 0.1 and R² > 0.98, indicating high generalization. Plotted visuals show strong alignment of predicted and real values, demonstrating the method’s efficiency for sensor correction. This approach has the potential to improve real-time aerodynamic testing by providing robust, adaptive calibration where conventional methods fall short