Nonlinear Model Updating of Aerospace Structures via Taylor-Series Reduced-Order Models

This paper presents a nonlinear model updating framework for aerospace structures that integrates Taylor-series-based reduced-order modeling with a complex unitary Cayley transform to adapt projection bases, thereby enabling accurate recovery of stiffness parameters and amplitude-dependent dynamic characteristics that linear methods cannot capture.

The Gist

Imagine you are trying to build a perfect digital twin of a real airplane wing. You have a computer model (the "digital twin") and a physical wing in a lab (the "real thing"). Your goal is to tweak the computer model so that it vibrates and moves exactly like the real wing.

This is called Model Updating. For simple, linear structures (like a stiff ruler), engineers have been doing this for decades. But airplanes aren't always simple. When they vibrate hard, parts bend, joints rub, and the structure gets stiffer or looser depending on how hard it's shaking. This is nonlinearity.

The problem? If you try to fix a computer model of a wiggly, nonlinear wing using old-school "linear" tools, the computer gets confused. It tries to fix the errors by changing the wrong things (like making the metal "stiffer" in the wrong places) just to force the numbers to match, rather than understanding the actual physics.

This paper introduces a new, smarter way to fix these models. Here is the breakdown using simple analogies:

1. The Problem: The "One-Size-Fits-All" Mistake

Imagine you are trying to tune a guitar.

• The Linear Approach: You assume the strings are perfectly stiff no matter how hard you pluck them. If the guitar sounds out of tune when you strum hard, you might try to change the wood of the guitar neck to "fix" the pitch. You are forcing a square peg into a round hole.
• The Reality: When you strum hard, the strings stretch and the pitch goes up (this is amplitude-dependent stiffness). The guitar isn't broken; it's just behaving non-linearly.

The old method (Hollins et al., 2026) was great at tuning the "wood" (stiffness) but assumed the strings never changed tension. This paper says: "Let's teach the computer that the strings do change tension."

2. The Solution: The "Taylor Series" Recipe

To handle the complexity without crashing the computer, the authors use a technique called Taylor-Series Reduced-Order Modeling (ROM).

Think of a complex, wiggly dance routine (the real wing's movement) as a song.

• The Full Model: Trying to record every single pixel of the dancer's movement. It's huge, slow, and takes forever to process.
• The Reduced Model (ROM): Instead of recording every pixel, you identify the main "moves" (the most important vibrations) and ignore the tiny, insignificant wiggles.
• The Taylor Series: This is like a recipe. You take the complex, non-linear dance and break it down into simple ingredients:
  1. The basic move (Linear).
  2. How the move changes if you do it twice as hard (Quadratic).
  3. How it changes if you do it three times as hard (Cubic).

By keeping only the most important "ingredients" (the first few terms of the recipe), they create a tiny, fast computer model that still knows how to dance like the big, heavy one.

3. The Magic Tool: The "Cayley Transform" Compass

Once they have this simplified model, they need to adjust it to match the real wing. They use a mathematical tool called the Cayley Transform.

Imagine you are navigating a ship.

• The Old Way: You try to steer by pushing the rudder in random directions. Sometimes you spin in circles or get stuck.
• The New Way: The Cayley Transform is like a magical compass that guarantees you always stay on the "ocean" of valid shapes. It ensures that as you tweak the model to match the real wing, the model never breaks its own rules (mathematically, it stays "orthogonal" or "unitary").

In this paper, they upgraded this compass. The old compass only worked for straight lines (real numbers). The new compass works for spirals and circles (complex numbers), which is necessary because vibrating wings with damping (friction) move in spirals, not just straight lines.

4. The Result: A Model That "Gets" the Vibration

The authors tested this on a model of a wingbox (the skeleton of a wing).

• The Test: They shook the wing gently, then shook it hard.
• The Old Method: When the wing was shaken hard, the old model failed. It couldn't explain why the pitch changed, so it gave up and gave wrong answers.
• The New Method: The new model said, "Ah, you're shaking hard! I know that makes the structure stiffer." It adjusted its internal math to match the real-world vibration perfectly.

The Analogy of the "Smart Thermostat":

• Linear Model: A thermostat that thinks the room is always 70°F. If you open a window, it just cranks the heat to maximum, burning energy and never getting the room right.
• Nonlinear Model: A smart thermostat that knows, "Oh, the window is open and the draft is strong. I need to adjust my strategy dynamically to keep the room comfortable."

Why Does This Matter?

1. Safety: Aerospace engineers need to know exactly how a plane will behave in extreme turbulence, not just gentle breezes.
2. Accuracy: This method recovers the true physical properties of the structure (like how stiff the metal really is) instead of faking it with wrong numbers.
3. Speed: By using the "Reduced Order" trick, they can run these complex simulations on a laptop in seconds instead of waiting days on a supercomputer.

In a nutshell: This paper teaches computers to stop treating vibrating airplane parts like stiff, boring rulers and start treating them like flexible, living things that change their behavior when pushed hard. It combines a smart recipe (Taylor Series) with a better compass (Cayley Transform) to build digital twins that actually work in the real world.

Technical Summary

1. Problem Statement

Finite Element (FE) model updating is a mature discipline for linear structures, used to reconcile computational predictions with experimental data by tuning parameters like stiffness. However, extending this to nonlinear regimes remains a significant challenge. Aerospace structures often exhibit nonlinear behaviors (e.g., geometric nonlinearity in thin panels, joint friction, free-play) that manifest as amplitude-dependent natural frequencies and mode shapes.

When a purely linear updating scheme is applied to such structures:

• It fails to capture amplitude-dependent frequency shifts.
• It may converge to biased stiffness parameters that attempt to "absorb" the nonlinear discrepancy rather than reflecting the true physical state.
• Existing linear updating frameworks (e.g., Hollins et al., 2026) rely on real orthogonal bases and linear eigenvalue problems, which are insufficient for damped, nonlinear systems with complex eigenvectors.

2. Methodology

The paper proposes a unified framework that integrates Nonlinear Model Order Reduction (NMOR) with a projection-basis adaptation scheme. The methodology consists of four main stages:

A. Structural Modeling and Nonlinear Formulation

• Base Model: The authors utilize a high-fidelity FE model of an aircraft wingbox panel (70,890 DOFs), partitioned into 11 substructures.
• Nonlinearity: The system is augmented with proportional (Rayleigh) damping and a cubic stiffness nonlinearity ($f_{NL} = K_3 x^{\circ 3}$), representing hardening-spring behavior.
• State-Space Conversion: The second-order equations of motion are recast into a first-order autonomous system: $\dot{w} = Aw + F_{NL}(w)$.
• Jacobian Analysis: The Jacobian matrix $A$ of the linearized system is decomposed into right ($\Phi$) and left ($\Psi$) eigenvectors, satisfying a biorthogonality condition ($\Psi^H \Phi = I$).

B. Taylor-Series Reduced-Order Modeling (NMOR)

• Expansion: The nonlinear force term is expanded using a Taylor series about the equilibrium point. For cubic stiffness, the second-order term vanishes, leaving a third-order operator $C$.
• Projection: The nonlinear equations are projected onto the reduced basis formed by the dominant eigenvectors of the damped Jacobian.
• Result: This yields a low-dimensional Nonlinear Reduced-Order Model (ROM) that retains the amplitude-dependent dynamics (e.g., frequency shifts) while reducing computational cost.

C. Complex Basis Adaptation via Unitary Cayley Transform

• Challenge: The linear updating scheme by Hollins et al. uses the Cayley transform to map a skew-symmetric matrix to an orthogonal matrix (real Stiefel manifold). This fails for the complex eigenvectors of damped nonlinear systems.
• Innovation: The authors generalize the Cayley transform from the Real Orthogonal Group $O(r)$ to the Unitary Group $U(r)$.
• Mechanism: By replacing skew-symmetry with skew-Hermiticity, the transform parametrizes the adaptation of the complex projection basis. This ensures the basis remains unitary (preserving orthogonality in the complex domain) during optimization.

D. Nonlinear Model Updating Optimization

• Objective: Minimize the discrepancy between experimental and analytical mode shapes using the Modal Assurance Criterion (MAC).
• Variables: The optimization vector includes stiffness modifiers ($\mu$) and Cayley transform parameters ($\tau$).
• Constraint: Frequency matching constraints are applied.
• Strategy: A two-stage approach is used: first optimizing stiffness with a fixed basis, then refining the basis adaptation.

3. Key Contributions

1. Unified Framework: Successfully bridges the gap between Tantaroudas's NMOR technique (Taylor-series expansion) and Hollins's basis adaptation scheme.
2. Unitary Cayley Transform: Generalizes the basis adaptation method to handle complex eigenvectors, enabling the updating of damped, nonlinear systems.
3. Amplitude-Dependent MAC: Introduces an objective function that accounts for the fact that mode shapes and frequencies change with vibration amplitude, a factor ignored by linear schemes.
4. Computational Efficiency: Demonstrates that the NMOR approach captures nonlinear physics with significantly fewer degrees of freedom (e.g., 15 modes vs. 30 full states in the study) while maintaining high accuracy.

4. Numerical Results

The framework was validated using synthetic data from the Hollins et al. wingbox experiment.

• Nonlinear Behavior Characterization:
  • As cubic stiffness ($K_{nl}$) increases, the system exhibits hardening behavior, with dominant frequencies shifting from ~22 Hz (linear) to ~58 Hz at high amplitudes.
  • Low damping allows these nonlinear effects to persist, while high damping suppresses them.
• ROM Accuracy:
  • Linear ROM: Failed to track the full-order model (FOM) in nonlinear regimes, showing errors >150% in time-domain predictions.
  • NMOR: Accurately tracked the FOM with errors <1% for moderate nonlinearity and <10% for extreme nonlinearity.
  • Convergence: NMOR error decreased monotonically with the number of retained modes, whereas linear ROM error remained flat (unable to compensate for missing physics).
• Updating Performance:
  • Linear Updating: Converged to biased stiffness parameters. At high vibration amplitudes, the Mean MAC dropped to 0.928.
  • NMOR Updating: Successfully recovered the true stiffness parameters. At high amplitudes, the Mean MAC remained above 0.999.
  • Computational Cost: The NMOR integration was 2.0–2.2× faster than the full-order model, despite including nonlinear projections.

5. Significance

This paper addresses a critical gap in aerospace structural dynamics: the inability of standard linear model updating to handle amplitude-dependent nonlinearities.

• Physical Fidelity: By incorporating the describing-function approximation and Taylor-series operators, the method provides a physically consistent parametrization of nonlinear stiffness.
• Robustness: The unitary Cayley transform ensures numerical stability when adapting complex bases, a necessity for damped and aeroelastic systems.
• Practical Application: The ability to recover true stiffness parameters even when experimental data is collected at high amplitudes (where linear models fail) makes this approach vital for the certification and health monitoring of modern, lightweight aerospace structures that operate in nonlinear regimes.

Future Work: The authors plan to apply this framework to experimental data from physical wingboxes with introduced nonlinear elements and extend it to aeroelastic systems where the Jacobian is inherently non-symmetric.