Rapid Worst-Case Gust Identification for Very Flexible Aircraft Using Reduced-Order Models

Nomenclature

• The listed symbols cover aerofoil geometry, inertia, stiffness, structural response, gust parameters, and reduced-order model quantities.Examples include gust gradient distance Hg, response metric J, and retained ROM modes m.Source: Nomenclature · pp. 1–2
• Greek symbols and state variables denote quantities including angle of attack, flap deflection, plunge displacement, eigenvalues, eigenvectors, and air density.The glossary also defines natural frequencies and reduced velocity.Source: Nomenclature · p. 2
• The acronym list expands terms for model reduction, aircraft classes, aerodynamic methods, and analysis tools.It includes ROM, NMOR, UVLM, UAV, HALE, and VFA.Source: Nomenclature · p. 2

1 Introduction

• Nonlinear full-order simulations can make searches over O(10^2) to O(10^3) gust combinations take weeks to months, despite their importance to certification.A single simulation with hundreds to thousands of DOF typically takes hours on one processor; the Helios investigation highlighted the importance of gust response.Source: §1 · p. 2
• Existing MOR approaches have limitations: POD needs excitation-specific training, harmonic balance targets periodic regimes, and linear identification misses geometric nonlinearities.The paper's NMOR projects onto a Jacobian eigenvector basis and retains nonlinear terms through Taylor expansion, yielding a ROM applicable to arbitrary gust profiles without regeneration [Da Ronch et al., 2013a,b, Tantaroudas and Da Ronch, 2017].Source: §1 · p. 3
• Subsequent work developed nonlinear modal aircraft models, control and estimation methods, adaptive ROM sampling, and assessments of flexible-wing modeling and worst-case gust loads.The passage cites Wang et al. and, Artola et al., Goizueta et al., Riso and Cesnik, and Balatti et al..Source: §1 · p. 3
• Earlier NMOR work covered gust loads, nonlinear control, coupled flight dynamics, wind-tunnel validation, and aerodynamic-model fidelity.The companion paper presents a detailed derivation with third-order Taylor terms and systematic eigenvector selection [Da Ronch et al., 2013a] [Da Ronch et al., 2013b] [Tantaroudas et al., 2015, 2014] [Da Ronch et al., 2014a, Papatheou et al., 2013, Fichera et al., 2014] [Da Ronch et al., 2014b] [Tantaroudas et al., 2026].Source: §1 · p. 3
• The paper demonstrates ROM-based worst-case gust searches across three test cases, reports speedups up to 600×, and distinguishes linear and nonlinear ROM validity domains.It also offers guidance for integrating the method into certification workflows.Source: §1 · p. 3

2.1 Discrete Gust: 1-Minus-Cosine Profile

• The 1-minus-cosine discrete gust is parameterized by peak velocity w0 and gradient distance Hg, and certification searches Hg to maximize the structural response.The gust velocity is zero outside [0, Hg]; its frequency content is centered at fg = U∞/Hg, with short gusts exciting structural modes and long gusts exciting rigid-body flight dynamics.Source: §2.1 · p. 3, Figure 1 · p. 4

2.2 Continuous Turbulence: Von Kármán Spectrum

• Continuous-turbulence analysis uses the Von Kármán PSD of vertical gust velocity, with turbulence intensity σw, scale length Lw, and spatial frequency Ω.Time-domain realizations are generated by filtering white noise through a rational approximation matched to the spectrum.Source: §2.2 · p. 4

2.3 Gust Penetration Effects

• For long-span aircraft, gust penetration creates spanwise loading phase variations through the time delay between gust arrival at the wing root and tip.The straight-wing configurations here have Λ = 0, so gust encounter is uniform along the span.Source: §2.3 · p. 4

3.1 Coupled System Formulation

• The coupled model combines unsteady strip-theory aerodynamics, geometrically exact nonlinear beam structures, and quaternion-based flight dynamics.Its state vector partitions into aerodynamic, structural, and rigid-body states, while gust input accounts for stripwise gust velocity and penetration delays; the Küssner function models unsteady lift build-up using two augmented aerodynamic states per strip.Source: §3.1 · pp. 4–5

3.2 Nonlinear Model Order Reduction

• NMOR expands the nonlinear residual about trimmed equilibrium through second order and projects the system onto biorthonormal Jacobian eigenvectors.The bilinear interaction coefficients are computed by matrix-free finite differences; basis selection retains near-origin real eigenvalues for rigid-body and gust coupling dynamics and lightly damped complex eigenvalues for structural bending and torsion.Source: §3.2 · p. 5
• The ROM is excitation-independent: its coefficients are computed once and it can then be driven by different gust profiles without regeneration.This property makes parametric sweeps over gust parameters computationally trivial.Source: §3.2 · p. 5

4 Worst-Case Gust Search Methodology

• The constrained search maximizes the structural response metric over gust gradient distance, using the certification-prescribed design gust velocity.The metric is typically maximum wing-tip vertical displacement or maximum root bending moment.Source: §4 · p. 5
• The workflow constructs a ROM once, sweeps candidate gust gradients, identifies the worst case, and validates it with one full-order nonlinear simulation.For each candidate Hg, the m-dimensional ROM is time-marched and its response metric recorded.Source: §4 · p. 5
• The linear ROM retains only the Jacobian term, while the nonlinear ROM also retains second-order Taylor terms to capture geometric stiffening and large-deformation coupling.The linear ROM is valid for small perturbations with tip deflections below approximately 10% of wingspan.Source: §4 · p. 5

5.1 Three-Degree-of-Freedom Aerofoil

• The test case is a two-dimensional aerofoil with pitch, plunge, and trailing-edge flap degrees of freedom, represented by 14 full-order states and a four-mode ROM.The states comprise eight aerodynamic and six structural states.Source: §5.1 · p. 5, Figure 3 · p. 6
• Cubic hardening nonlinearities are included in pitch and plunge stiffness, with Kα3 = 3 and Kξ3 = 1.0.The search uses maximum plunge displacement as its response metric and sets peak gust velocity to 14% of freestream.Source: §5.1 · p. 6
• At tg ≈55 semichords, the pitch response has its largest oscillation, near the plunge natural frequency; the worst case is not the longest gust.The intermediate-length gust produces frequency-content resonance that amplifies the dynamic response beyond the quasi-static longest-gust case.Source: §5.1 · p. 6
• The nonlinear ROM reproduces full-order responses across the parameter space and evaluates all 1,000 gust design sites at a fraction of full-order computational cost.The sweep covers 1,000 gust gradient distances from 0.1 to 100 semichords at reduced velocity U*=4.5, below the linear flutter speed.Source: §5.1 · p. 6

5.2 Global Hawk-Like UAV

• The 540-DOF UAV model is reduced to an eight-mode ROM, and 80 gust cases spanning 0 to 776 MAC are swept to identify the worst-case condition.The model uses a 17.75-m semi-span, with 20 wing beam elements and beam-stick fuselage and empennage representations.Source: §5.2 · p. 6
• At the worst case, the approximately 9 m wing-tip deflection reaches 50% of semi-span, and the nonlinear ROM captures saturation while the linear ROM overpredicts the peak.Geometric stiffening rotates the local lift vector inward during upward bending, reducing net vertical force.Source: §5.2 · p. 6

5.2.1 Effect of gust gradient distance

• Increasing gust intensity substantially increases pitch-oscillation amplitude, while peak response varies non-monotonically with gust parameters.The reported dependence supports exhaustive parametric searches rather than simplified analytical estimates.Source: §5.2.1 · p. 7
• The ROM makes the UAV parametric sweeps computationally affordable, as illustrated by the gust-intensity response study.The passage connects the practical value of ROM sweeps to prior work [Da Ronch et al., 2013a, 2014b].Source: §5.2.1 · p. 7

5.3 Very Flexible Flying-Wing

• The test case is a 32-m-span flying wing with 1,616 full-order degrees of freedom, reduced to a 9-mode ROM and evaluated over 37 gust durations.The sweep spans 0.5–20 s gust durations, corresponding to Hg = 12.5–500 m.Source: §5.3 · p. 7

5.3.1 Worst-case identification

• At approximately 5.0 s (Hg = 125 m), the worst-case gust produces wing-tip deformation of approximately 10% of the wingspan and a 7.7-degree maximum bending slope.The peak structural response metrics are plotted against gust duration in Figure 8.Source: Figure 8 · p. 9, §5.3.1 · p. 9

5.3.2 Computational speedup

• 600× speedup comes from integrating a 9-dimensional ROM instead of a 1,616-dimensional model and avoiding per-step nonlinear-equilibrium iterations.Table 2 covers the computational cost comparison for 37 gust cases.Source: §5.3.2 · p. 10, Table 2 · p. 10

5.3.3 Linear versus nonlinear ROM accuracy

• The nonlinear ROM with second-order Taylor terms matches the full-order response through the worst-case time history, while the linear ROM overpredicts peak displacement as geometric stiffening grows.Figure 9 compares the full-order, nonlinear-ROM, and linear-ROM time histories.Source: §5.3.3 · p. 10, Figure 9 · p. 10

5.3.4 ROM convergence with number of modes

• Nine retained modes achieve convergence: rigid-body modes capture mean displacement, the first bending mode dominates oscillations, and higher modes refine high-frequency detail.Figure 10 shows convergence as modes are added.Source: §5.3.4 · p. 10, Figure 10 · p. 11

5.3.5 Effect of structural flexibility on flight dynamics

• The flexible aircraft and equivalent rigid aircraft have measurably different vertical-displacement trajectories at the worst-case gust, so flexibility affects flight-dynamic response.Figure 11 compares the flexible trajectory (solid) with the rigid trajectory (dashed).Source: §5.3.5 · p. 11

6.1 Practical Guidelines for ROM Selection

• Below approximately 10% wingspan tip deformation, the linear ROM is adequate for screening; above that level, second-order nonlinear terms are essential to capture geometric stiffening.In the larger-deformation regime, the linear ROM can overpredict loads and potentially misidentify the worst case.Source: §6.1 · p. 11

6.2 Integration into Certification Workflows

• The workflow uses ROMs to sweep the flight-condition and gust-length space, then validates candidate worst cases with a small number of full-order simulations.The reported speedups range from 30× for the UAV to 600× for the flying wing while maintaining final design-load fidelity.Source: §6.2 · pp. 11–12

6.3 Limitations and Extensions

• The current framework uses strip-theory aerodynamics, adequate for preliminary design and high-aspect-ratio wings at moderate angles of attack; detailed certification may require UVLM or panel methods.The NMOR framework is described as compatible with such aerodynamic upgrades.Source: §6.3 · p. 12
• The ROM framework can extend to lateral gusts, gust gradients, and combined gust-manoeuvre cases by adding disturbance input channels.Source: §6.3 · p. 12

• Up to 600× speedup reduces the 1,616-DOF flying-wing search from 222 hours to 22 minutes.Source: §7 · p. 12
• The linear ROM is sufficient below 10% wingspan tip deformation; larger deformations require a second-order nonlinear ROM to capture geometric stiffening.Source: §7 · p. 12
• Intermediate gust-gradient distances can produce the largest structural loads through resonant excitation, so the worst case is not a monotonic function of gust gradient distance.The conclusions state that this dependence necessitates exhaustive parametric searches.Source: §7 · p. 12
• The ROM can screen for candidate worst cases and use a small number of full-order simulations for validation in certification workflows.Source: §7 · p. 13
• The reduction procedure can support higher-fidelity aerodynamic models, including CFD and UVLM, because it operates on the coupled system's Jacobian eigenspectrum.The passage also cites [Da Ronch et al., 2012, Tantaroudas and Da Ronch, 2017].Source: §7 · p. 13