PSI - Issue 80

Tafara E. Makuni et al. / Procedia Structural Integrity 80 (2026) 105–116 Tafara E. Makuni / Structural Integrity Procedia 00 (2019) 000 – 000

107

3

to gain knowledge on desired outputs (Jones, et al., 2020). The development of a DT model for aerodynamic load reconstruction would be invaluable for structural usage monitoring, SUM, as a more realistic aerodynamic loading profile could better predict the service life of an AV through a more accurate prediction of the structural response of the wing in real time (Stieger, 1932) (Balageas, 2002). The development of a DT model for aerodynamic load reconstruction would be invaluable as inputting aerodynamic loading to obtain a structural response is complex (Shkarayev, et al., 2002) (Trivailo, et al., 2006) (Maneth, et al., 2016). At present, the main challenge in better understanding the structural response of an aeroplane wing due to aerodynamic loading are mainly associated with, (i) the high computational costs involved with multiple and detailed simulations, and (ii), the expense of experimental testing (Makuni, et al., 2014) (Makuni, et al., 2015) (Kalsi, et al., 2018), in both the fields of aerodynamics and structural mechanics. This paper details the low fidelity simulations conducted for developing a DT platform for aerodynamic load reconstruction. The results are verified through simple experiments of force measurements on an aerofoil section. The most comprehensive study that combined Aerodynamics and Structures in terms of digital twinning, has been conducted by Professor Willcox’s laboratory based at the University of Texas. Here the physical asset is a 12-ft wingspan unmanned aerial vehicle, UAV. The approach combined a library of component-based reduced-order models to create data-driven physics-based DTs. The reduced-order modelling produced physics-based computational models that are fast and accurate enough for predictive DTs (Kapteyn, et al., 2022). The DT can be deployed and updated using interpretable machine learning, ML, specifically optimal trees (Kapteyn, et al., 2020). This data is further augmented further using experimental or historical databases (Kapteyn, et al., 2004). The methods are demonstrated through the development of the UAV (Kapteyn, et al., 2004) (Kapteyn, et al., 2020) (Kapteyn, et al., 2022). In terms of load reconstruction, proper orthogonal decomposition, POD, has been used to reconstruct the aerodynamic flow field around an aerofoil using limited data (Bui-Thanh, et al., 2004). Reconstructing the aerodynamic loading from measured values on an AV in real time is an important but challenging task for the next generation of AVs (Shkarayev, et al., 2002). This has been difficult to achieve using traditional methods (Trivailo, et al., 2006), but digital twinning offers a new method to address this challenging situation. The DT platform presented in this paper aims to reconstruct the aerodynamic loading of an aeroplane wing based on a limited number of sensor readings for realistic flight conditions. Aeroplane wings are usually designed using 2D aerofoil sections (Stinton, 2001) (Raymer, 2012) (Abbott, et al., 2012). When designing 2D aerofoil profiles, a popular software to use is XFOIL which can calculate the lift, drag, moment and pressure coefficient; C D , C L , C M , and C p , respectively; using a vortex panel-based method (Drela, 2013); and uses Karman-Tsien compressibility correction at high subsonic free-stream speeds. XFOIL focuses on the steady-state aerodynamics of that aerofoil profile; and can perform both a viscous and inviscid analysis (Gudmundsson, 2014). JavaFoil can also be used to calculate C D , C L , C M , and C p to for different M ∞ , α ∞ and h (hence Re) . JavaFoil software uses established aerodynamic principles such as potential flow analysis using a higher-order panel method and boundary layer analysis using an integral method to analyse aerofoil performance in low-medium speed subsonic flows (Hepperle, 2017). XFOIL has a Re dependency when calculating the drag due to the way in which drag is calculated. JavaFoil does not have this Re dependency meaning that it can be used to assess the effects of Re . In 2020, Conlan-Smith et al., optimised the shape of an aerofoil profile using XFOIL using a C L -based criteria (Conlan-Smith, et al., 2020). In 2002, Shkarayev et al. developed an inverse-based interpolation methodology that parametrically approximates the loading then uses a least-squares based minimisation of calculated strain values to the recorded strain values. In this study, the DT model was validated using both computational and experimental data; with strain gauges, SGs, applied on both the upper and lower surfaces of the aircraft wing (Shkarayev, et al., 2002). Later in 2004, Tessler et al. applied this method to an aircraft wing box structure and achieved limited success in reconstructing the loading profile using traditional methods. Shkarayev et al. also used the inverse method to reconstruct the loading on a cantilevered rectangular plate under a traverse loading provided by a polynomial (Shkarayev, et al., 2004). The inverse method can also involve Neural Networks, NN. In 2006, Trivailo et al. used trained NNs to build a model capable of estimating the aerodynamic loading inversely from the strain gauge, SG, data (Trivailo, et al., 2006). The results followed the general trends well, however contained significant error in predicting the exact load. This method however, achieved better accuracy than traditional methods that have used the inverse method to reconstruct the loading on an aircraft wing box (Shkarayev, et al., 2004).