MSCT: A Multiscale Convolutional Transformer Model for Load Prediction of Aircraft Landing Gear

Aircraft landing gear load monitoring helps detect structural problems early and prevents potential accidents. Load prediction methods are the most important part of load monitoring, directly determining the accuracy of landing gear load assessment. However, current approaches exhibit limitations such as insufficient modeling of nonlinearities, reliance on simulation-generated data, and failure to consider spatial dependencies among landing gear sensors. In this article, we propose a multiscale convolutional transformer (MSCT) model to address these issues and enhance load prediction performance. Specifically, we conducted ground calibration test on the right landing gear of a real aircraft, employing fiber Bragg grating (FBG) sensors to collect strain data corresponding to heading (X), longitudinal (Y), and axial (Z) load axes. The MSCT model integrates multiscale convolutional layers, positional encoding (PE), and cross-scale attention mechanisms to effectively capture spatial correlations and local-global dependencies among sensors. Comparative experiment demonstrate that MSCT achieves superior prediction accuracy and generalization capability, with mean absolute percentage errors (MAPE) of 2.3821%, 2.8064%, 0.6286%, 1.7606%, and 2.7387% for X, -X, Y, Z, and -Z directions, respectively. We also conducted an ablation study to show the benefits of each component in MSCT.