Generating High-Resolution Flight Parameters in Structural Digital Twins Using Deep Learning-based Upsampling

The structural digital twin is a virtual representation of physical entities that accurately predicts the evolution of structural damage through multidisciplinary and multi-level probabilistic simulations. It provides crucial support for prognostic and health management. Flight parameters are important input data for airframe digital twin to support aerodynamic and structural simulations. However, many small aircraft or UAVs often suffer from insufficient sampling rates of flight parameters due to cost limitation or premature service. In this study, we propose a deep learning-based flight data upsampling method that effvbectively enhances the resolution of flight data. The method constructs an upsampling model using a one-dimensional super-resolution convolutional residual network, defines multiple loss functions associated with the flight data, and uses a highly sampled test aircraft dataset for training. The proposed method is validated using real UAV flight test data and several criteria, achieving good results with different upsampling factors. This approach is expected to facilitate the construction of structural digital twins in the future.