Craig-Bampton-based Quadratic Manifold for Nonlinear Substructuring

This paper proposes a Nonlinear Craig-Bampton (NL-CB) method that extends classical component mode synthesis to geometrically nonlinear structures by constructing a quadratic reduction manifold, enabling efficient and stable reduced-order modeling through Galerkin projection.

The Gist

Imagine you are trying to simulate how a massive, complex skyscraper will sway during an earthquake. If you tried to model every single atom and every single bolt in that building using a supercomputer, the calculation would take years. It’s simply too much data.

To solve this, engineers use a trick called "Substructuring." Instead of looking at the whole building at once, they break it into smaller pieces—like floors or wings—and create a "mini-model" (a Reduced Order Model) for each piece. Then, they just snap those mini-models together.

The problem? Most of these "mini-model" tricks only work for simple, linear movements (like a spring stretching). But real buildings, airplanes, and tiny microscopic sensors (MEMS) are nonlinear. They bend, twist, and stretch in ways that aren't simple. If you use a linear mini-model on a nonlinear structure, your simulation will be as useless as a paper umbrella in a hurricane.

This paper introduces a new way to build these mini-models called the NL-CB method. Here is how it works, explained through a few analogies.

---

1. The "High-Frequency" Filter (The Background Noise Analogy)

In any complex structure, there are two types of movements:

• Low-frequency movements: The big, slow, sweeping motions (like the whole building swaying).
• High-frequency movements: The tiny, rapid vibrations (like the rattling of a windowpane).

The researchers realized that if you are trying to understand the big sway, you don't need to track every tiny rattle individually. It’s like listening to a symphony: you want to hear the melody (the low frequencies), not the microscopic sound of the violinist's bow rubbing against the string (the high frequencies).

The NL-CB method uses a mathematical "filter." It takes those tiny, high-frequency vibrations and "condenses" them. Instead of treating them as separate moving parts, it mathematically links them to the big, slow movements. It says: "If the building sways this much, we know the windows will rattle exactly this way." This keeps the model tiny and fast without losing the important details.

2. The "Quadratic Manifold" (The Rollercoaster Track Analogy)

Usually, when engineers try to add nonlinearity to these mini-models, the math becomes a nightmare. The "map" of possible movements becomes so complicated that the computer gets lost.

The authors use something called a Quadratic Manifold.
Imagine you are driving a car. A linear model assumes the road is perfectly flat. A nonlinear model tries to account for every single pebble and crack, which is exhausting to calculate.

The Quadratic Manifold is like designing a smooth, curved rollercoaster track. It’s not flat, so it captures the "curves" (the nonlinearities) of the real world, but it’s a smooth, predictable mathematical shape (a parabola). Because the track is a smooth curve rather than a jagged mess of pebbles, the computer can "drive" through the simulation at lightning speed.

3. Why does this matter? (The "Lego" Advantage)

The most important part of this paper is Modularity.

Imagine you are building a complex Lego castle. If you decide to change the shape of one tower, you don't want to have to rebuild the entire castle from scratch. In traditional nonlinear modeling, if you change one tiny part of a structure, you often have to re-run the entire massive simulation.

Because this method uses "substructures," it’s like having Lego bricks. If you change the design of one wing of a building, you only have to re-calculate the "mini-model" for that specific wing. The rest of the building stays exactly as it was. This makes designing new airplanes, safer bridges, or more sensitive medical sensors much, much faster.

Summary

In short, the authors have created a way to:

1. Break big problems into small pieces (Substructuring).
2. Ignore the "noise" while keeping the "melody" (High-frequency condensation).
3. Use smooth, curved math to handle complex twisting and bending (The Quadratic Manifold).
4. Keep everything modular so engineers can swap parts in and out without starting over.

The result? A simulation that is incredibly fast (up to 66,000 times faster in some cases!) but still accurate enough to predict how a microscopic sensor or a massive beam will actually behave in the real world.

Technical Summary

Technical Summary: Craig-Bampton-based Quadratic Manifold for Nonlinear Substructuring

Authors: Alexander Saccani and Paolo Tiso (ETH Zürich)

1. Problem Statement

The primary challenge addressed in this paper is the extension of Component Mode Synthesis (CMS)—specifically the widely used Craig-Bampton (CB) method—from linear structural dynamics to geometrically nonlinear systems.

While CMS is highly effective for modular, large-scale linear assemblies, its nonlinear counterparts are currently limited. Existing nonlinear model reduction techniques (like ICE or Spectral Submanifolds) are typically "monolithic," meaning they require the entire structure to be assembled before reduction. This lack of modularity makes them difficult to integrate into industrial workflows where different components are designed independently, and it prevents efficient model updating (where changing one component would require re-computing the entire global model).

2. Methodology

The authors propose a new Nonlinear Craig-Bampton (NL-CB) Reduced-Order Model (ROM). The core innovation is the construction of a quadratic reduction manifold derived via perturbation analysis.

Key methodological steps include:

• Manifold Derivation: Instead of a linear subspace, the internal degrees of freedom (DoFs) are approximated using a nonlinear manifold. The authors assume that high-frequency Fixed-Interface Modes (FIMs) can be statically enslaved to low-frequency FIMs and interface coordinates (Static Modes). This is justified by the assumption that the inertia of high-frequency modes is negligible during the dynamic response of the lower-frequency modes.
• Taylor Expansion: The manifold is expressed as a second-order Taylor series in terms of the reduced coordinates (low-frequency modes and interface modes).
• Galerkin Projection: The substructure equations of motion are projected onto the tangent space of this quadratic manifold.
• Simplifying Assumptions for Efficiency: To ensure the ROM is computationally efficient for time integration, the authors:
  1. Truncate the resulting high-order polynomial forces (which could reach 7th order) to third order.
  2. Neglect "convective" inertial and damping terms arising from the quadratic manifold.
  3. Neglect the projection of external loads onto the quadratic part of the manifold.
• Element-Level Projection: To avoid the massive memory requirements of assembling full-order nonlinear tensors, the authors implement an element-level projection strategy, computing nonlinear stiffness tensors directly from individual finite elements.

3. Key Contributions

• NL-CB Framework: A novel, modular nonlinear substructuring method that preserves the "divide and conquer" philosophy of classical CMS.
• Structure Preservation: The authors mathematically prove that the NL-CB ROM preserves the Lagrangian structure of the original Finite Element (FE) model. This ensures that the reduced mass, damping, and stiffness matrices are positive definite and that the system possesses a consistent potential energy, leading to superior numerical stability during time integration.
• Computational Efficiency: The method provides a polynomial structure that allows for massive speed-ups in time integration compared to full FE models.
• Mathematical Equivalence: The authors demonstrate that if the quadratic manifold is truncated to its linear component, the method perfectly retrieves the classical linear Craig-Bampton method.

4. Results and Validation

The method was validated through four increasingly complex numerical experiments:

1. Flat von Kármán Beam: Demonstrated that the quadratic manifold is essential for capturing "bending-stretching" coupling. Without the quadratic terms, the ROM was overly stiff.
2. Curved von Kármán Beam: Successfully captured the strong nonlinear coupling between low-frequency modes inherent in curved geometries.
3. Nonlinear Flat Panel: Tested under acoustic pressure (white noise). The NL-CB ROM accurately predicted power spectral densities (PSDs) and showed that the method remains robust even when the interface reduction is modified (demonstrating modularity).
4. MEMS Gyroscope: A high-dimensional case (over 260,000 DoFs). The NL-CB ROM achieved a speed-up of approximately $6.65 \times 10^4$ in time integration compared to the full FE model, while maintaining high accuracy in capturing both in-plane and out-of-plane nonlinear responses.

5. Significance

This work bridges the gap between modular substructuring and nonlinear dynamics. It provides a mathematically rigorous and computationally efficient framework for the nonlinear analysis of large-scale assemblies. Its significance lies in its potential for industrial application: it allows engineers to perform rapid, nonlinear design optimizations and parametric studies on complex systems (like aerospace or MEMS devices) by updating only the specific components that are modified, rather than re-simulating the entire assembly.