Aerodynamic Coupling Model (ACM)

• ACM is a modeling framework that explicitly links aerodynamic phenomena with structural and flight dynamics through reduced-order or high-fidelity approaches.
• It ranges from operator-level state-space mappings to full two-way fluid-structure interaction systems, tailoring fidelity to specific regimes.
• ACM techniques improve computational efficiency by replacing full CFD solves with learned surrogates while accurately capturing aeroelastic effects.

Aerodynamic Coupling Model (ACM) denotes a class of models that explicitly represent the channel by which aerodynamic states, flow-relative kinematics, structural motion, or actuation generate loads, pressures, or wrenches that feed other subsystems. The term is not used uniformly across the literature. In transonic aeroelastic reduced-order modeling it denotes a learned aerodynamic operator mapping modal motion histories to surface pressure distributions and then to structural forcing (Fonzi et al., 2023). In winged-blimp dynamics it denotes a fixed-wing-style aerodynamic submodel defined in the velocity frame and blended with a drag-based regime model (Wang et al., 25 Feb 2026). In free-flying flexible-aircraft analysis it appears as an unsteady strip-aerodynamic subsystem with Wagner and Küssner memory states embedded in a coupled aeroelastic-flight-dynamics state space (Tantaroudas et al., 23 Mar 2026). Across these usages, the common element is explicit representation of aerodynamic coupling rather than implicit reliance on full CFD or purely static coefficient lookup.

1. Terminology and conceptual scope

Across the cited literature, ACM is best understood as a modeling role rather than a single canonical equation set. It may denote a reduced aerodynamic surrogate, a fixed-wing coefficient model, a two-way fluid-structure interaction framework, a vibro-acoustic loading pathway, or a minimal inflow-sensitive actuator model. This suggests that the essential ACM question is not “which single formulation is correct,” but “which aerodynamic operator couples which states or inputs to which loads, in which regime, and at which fidelity.”

Context	ACM meaning	Reference
Transonic aeroelastic ROM	Learned state-space map from modal motion to surface pressure and structural loads	(Fonzi et al., 2023)
Winged blimp hybrid model	Fixed-wing-style aerodynamic wrench model in the velocity frame	(Wang et al., 25 Feb 2026)
Flexible aircraft with gusts	Strip-theory unsteady aerodynamic subsystem with augmented memory states	(Tantaroudas et al., 23 Mar 2026)
Flapping-wing FSI	Two-way aerodynamic-structural coupling framework in ANSYS	(Qiang et al., 2014)
Flexible frame structures	Co-rotational quasi-steady distributed aerodynamic load model	(Vanzulli et al., 2022)
Dual-rotor redundant actuation	Minimal thrust-inflow coupling model with trim-defined damping	(Franchi, 8 May 2026)

A second recurring distinction is between operator-level ACMs and architecture-level ACMs. Operator-level ACMs directly map motion or inflow to aerodynamic outputs, as in state-space pressure ROMs or velocity-frame force models. Architecture-level ACMs specify how aerodynamics, structure, acoustics, and flight dynamics are numerically linked, as in two-way FSI or class-3/class-4 aeroacoustic workflows.

2. State-space and aeroelastic ACMs

In transonic aeroelasticity, ACM commonly denotes a reduced aerodynamic dynamical system that replaces repeated high-fidelity flow solves. A representative formulation learns a discrete-time surrogate from CFD with Dynamic Mode Decomposition with control: $$\mathbf{x}^{k+1}=\mathbf{A}\mathbf{x}^k+\mathbf{B}\mathbf{u}^k.$$ Here the aerodynamic state is not the full flow field but the surface pressure-coefficient vector,

$$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$

and the input vector contains modal amplitudes and their first and second derivatives. After POD truncation with basis $\mathbf{U}_r$, the reduced model becomes

$$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$

This formulation is ACM in a direct sense: structural generalized coordinates drive a reduced aerodynamic state, the reconstructed surface pressure field is converted to aerodynamic forces through surface areas and normals, and those loads are projected back into structural modal equations. For the Benchmark Super Critical Wing at Mach $0.74$ and zero angle of attack, the reported best case uses 30 aerodynamic modes, reduces runtime from about 180 hours on a 40-core cluster node to minutes on a single core, and preserves flutter identification while reproducing spanwise pressure distributions at 20% and 80% span (Fonzi et al., 2023).

A frequency-domain variant appears in transonic typical-section modeling. There the ACM is the modal aerodynamic influence matrix

$$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$

which maps generalized coordinates to generalized aerodynamic forces through

$$\{\bar Q_a(\kappa)\}=\frac{(U^*)^2}{\pi\mu}[A(\kappa)]\{\eta(\kappa)\}.$$

The diagonal terms represent self-aerodynamic effects; the off-diagonal terms are the aerodynamic cross-couplings between plunge and pitch. The identified transfer functions are then fitted with an Eversman-Tewari rational-function approximation suitable for augmented state-space flutter analysis (Carloni et al., 2023).

A physics-based state-space ACM is developed for free-flying flexible aircraft with gusts. The complete first-order state is

$$\mathbf{w}= \begin{Bmatrix} \mathbf{w}_f\ \mathbf{w}_s\ \dot{\mathbf{w}}_s\ \mathbf{w}_r \end{Bmatrix},$$

where $\mathbf{w}_f$ are aerodynamic augmented states, $\mathbf{w}_s$ structural DOFs, and $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$0 rigid-body states including quaternion attitude. Each strip’s local effective velocity is

$$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$1

so the aerodynamic state depends simultaneously on rigid-body translation, rigid-body rotation, elastic deformation, structural velocity, and gust velocity. Unsteady memory is introduced through Wagner and Küssner states, making the ACM explicitly non-quasi-steady while remaining low-order and directly linearizable (Tantaroudas et al., 23 Mar 2026).

3. ACMs in rigid-body and hybrid vehicle dynamics

In winged-blimp modeling, ACM denotes the fixed-wing-style aerodynamic submodel valid in the attached-flow, moderate-to-high-speed, small-angle-of-attack regime. The rigid-body state is

$$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$2

with translational body-frame velocities $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$3 defining

$$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$4

The aerodynamic wrench is written in the velocity frame and rotated to the body frame: $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$5 with force and moment components parameterized as dynamic-pressure-scaled coefficient maps in $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$6 plus linear rotational damping terms. The identified ACM region is

$$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$7

obtained from experimentally determined thresholds and a $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$8 transition band around $$\mathbf{x}=\begin{bmatrix} Cp_1 & Cp_2 & \dots & Cp_m \end{bmatrix}^T,$$9 and $\mathbf{U}_r$0. Outside that regime the ACM is blended with a Generalized Drag Model through

$\mathbf{U}_r$1

where $\mathbf{U}_r$2 is produced by a neural mixer. Region-wise RMSE values show ACM-only is best in the ACM region and poor elsewhere, while the learned hybrid preserves ACM performance where its assumptions hold and suppresses spurious lift in low-speed, high-$\mathbf{U}_r$3 descent (Wang et al., 25 Feb 2026).

A related ACM interpretation appears in aggressive quadrotor modeling, where the aerodynamic coupling is decomposed into a nominal blade-element-momentum model and a learned residual: $\mathbf{U}_r$4 The nominal rotor model already contains cross-axis coupling through inflow, in-plane force, flapping, and translational airflow, while the residual network captures body/rotor interaction, rotor/rotor interaction, and short-memory effects from recent motion and motor histories. The resulting BEM+NN model achieves force and torque RMSE $\mathbf{U}_r$5 N and $\mathbf{U}_r$6 Nm on the full training set and is evaluated with motion-capture flight data up to 18 m/s (Bauersfeld et al., 2021).

An even more compact ACM appears in redundant dual-rotor actuation. There the single-rotor thrust law

$\mathbf{U}_r$7

makes net force

$\mathbf{U}_r$8

explicitly sensitive to air-relative velocity. The incremental damping coefficient is defined as

$\mathbf{U}_r$9

and for the affine-in-inflow model becomes proportional to $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$0. The ACM role here is not lift generation but trim-defined aero-mechanical damping modulation via input redundancy (Franchi, 8 May 2026).

4. High-fidelity, FE, and multiphysics ACMs

Some ACMs are not reduced surrogates but full two-way coupling frameworks. A flexible cicada-wing model links ANSYS Fluent, Transient Structural, and System Coupling in a partitioned staggered FSI loop with subiterations inside each time step. The interface conditions are

$$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$1

and the structural dynamics use

$$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$2

With $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$3 and 250 time steps, the coupled flexible-wing simulation yields $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$4 and $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$5, compared with $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$6 and $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$7 for the rigid wing under the same kinematics. In this usage, ACM is effectively synonymous with a two-way aerodynamic-structural coupling framework (Qiang et al., 2014).

A lower-order FE ACM for flexible frame structures embeds quasi-steady aerodynamics directly in a co-rotational beam residual: $$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$8 The local relative velocity is

$$\tilde{\mathbf{x}}_{k+1}=\tilde{\mathbf{A}}\tilde{\mathbf{x}}_k+\tilde{\mathbf{B}}\mathbf{u}_k,\qquad \mathbf{x}=\mathbf{U}_r\tilde{\mathbf{x}}.$$9

projected onto the current section plane to generate distributed drag, lift, and torsional moment. Equivalent nodal aerodynamic forces are obtained by virtual work, so load direction follows the current deformed section orientation. The formulation is explicitly quasi-steady and neglects the aerodynamic tangent matrix in Newton iterations (Vanzulli et al., 2022).

A specialized vibro-acoustic ACM appears in turbulent-boundary-layer transmission analysis. There the aerodynamic input is a wall-pressure cross-PSD,

$0.74$0

mapped to nodal force PSD and then to structural, cavity, and radiation responses through an FE-RRM chain. This is not a general CFD aeroelastic ACM, but it is a precise aerodynamic coupling model for TBL excitation of panel-cavity-panel systems (Adhikary et al., 2022). In aeroacoustics more broadly, numerically decoupled class-3 models represent one-way forward coupling from flow to acoustics, whereas class-4 models solve the full fluid-structure-acoustic interaction in coupled form (Schoder, 2024).

5. Identification, stabilization, and numerical realization

A large fraction of ACM research concerns how the coupling operator is identified and made numerically usable. In the transonic surface-pressure ROM, CFD snapshots are assembled into

$0.74$1

and the DMDc consistency relation

$0.74$2

is solved by SVD-based pseudoinversion. Because direct DMDc fitting can produce spurious unstable poles, the model is stabilized by extracting quasi-steady slopes $0.74$3, subtracting the quasi-steady part, identifying unsteady-only dynamics, and then reconstructing

$0.74$4

This removes lift drift and drives reduced-system eigenvalues back inside or on the unit circle (Fonzi et al., 2023).

Frequency-domain ACM identification imposes different requirements. In transonic typical-section ROMs, simultaneous multi-mode excitation with Walsh functions is efficient only if both the input signals and their derivatives are sufficiently orthogonal. Transfer functions are estimated through PSD and cross-PSD processing, and Hanning windows are reported to reduce spectral leakage substantially before rational-function approximation. The resulting ACM is therefore as much a signal-processing construction as an aerodynamic one (Carloni et al., 2023).

For descriptor-like aeroelastic systems, standard DMDc is modified to include next-step inputs: $0.74$5 The regression augments the input block with

$0.74$6

so that algebraic dependence on $0.74$7 is retained. Multiple local models are trained at different flight speeds and then interpolated not by direct matrix interpolation but by reconstructing the high-dimensional state and spline-interpolating in physical space over $0.74$8 to $0.74$9 (Fonzi et al., 2020).

In regime-partitioned rigid-body modeling, ACM identification can be separated from transition modeling. The winged-blimp ACM-GDM framework first fits ACM parameters on ACM-region data with $$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$0, then GDM parameters on GDM-region data with $$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$1, and only then trains a feedforward neural mixer with 3 layers, hidden sizes 32 and 16, ReLU activations, and sigmoid output. The mixer loss combines model error with anchor-point, monotonicity, and smoothness regularization. The paper also notes that the printed monotonicity penalty appears sign-inconsistent with the accompanying prose, making code inspection necessary for implementation fidelity (Wang et al., 25 Feb 2026).

6. Assumptions, limitations, and interpretive issues

A central limitation of many ACMs is local validity. Linear state-space transonic ROMs are trained on nonlinear CFD trajectories but remain local best-fit operators around the selected operating condition; when the response enters shock or separation regimes absent from training, the model diverges and additional local surrogates or LPV-style interpolation are required (Fonzi et al., 2023). The winged-blimp ACM is explicitly restricted to the attached-flow, high-$$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$2, small-$$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$3 regime and can generate spurious lift in low-speed, high-$$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$4, separated-flow motion; its own failure example predicts $$[A(\kappa)] = \begin{bmatrix} -\dfrac{C_{l_h}(\kappa)}{2} & -C_{l_\alpha}(\kappa) \ C_{m_h}(\kappa) & 2C_{m_\alpha}(\kappa) \end{bmatrix},$$5 during a near-vertical descent, producing erroneous forward drift (Wang et al., 25 Feb 2026). The strip-theory ACM for free-flying flexible aircraft assumes independent 2D strips, thin-airfoil theory, attached flow, and approximate drag, so it is intended for HALE-type, high-aspect-ratio, pre-stall configurations rather than strong 3D or separated-flow regimes (Tantaroudas et al., 23 Mar 2026).

A second issue is the distinction between quasi-steady, unsteady-memory, and fully coupled ACMs. Co-rotational beam ACMs compute loads from instantaneous projected relative velocity and empirical coefficients, omitting wake memory, dynamic stall, vortex shedding, and the aerodynamic tangent matrix (Vanzulli et al., 2022). By contrast, Wagner/Küssner strip ACMs retain circulatory memory in first-order aerodynamic states, while class-3 aeroacoustic couplings typically remain one-way and neglect acoustic back-coupling to the flow (Schoder, 2024). Thus “ACM” does not, by itself, specify whether the model is static, dynamic, weakly coupled, or bidirectionally coupled.

A common misconception is that ACM always means “lift model.” The literature does not support that reduction. In some cases ACM outputs surface pressure distributions on the wetted surface for subsequent nodal or modal force recovery; in others it outputs a 6D aerodynamic wrench; in others it maps TBL wall-pressure statistics into structural force PSDs; and in the VADA formulation it characterizes the local sensitivity of net force to air-relative velocity rather than lift generation. Another misconception is that ACM always refers to a standalone aerodynamic subsystem. Several papers use it at the level of a numerical coupling framework linking aerodynamics to structures, acoustics, or flight dynamics.

In this broader sense, ACM design is defined by three coupled choices: the aerodynamic variables retained (pressure field, coefficient surfaces, sectional states, doublet strengths, wall-pressure spectra, or thrust-inflow laws), the subsystems coupled (structure, rigid-body motion, acoustics, gusts, or control allocation), and the validity regime under which the resulting operator is expected to remain faithful.