Mechanism and Machine Theory 185 (2023) 105290

Contents lists available at ScienceDirect

Mechanism and Machine Theory

journal homepage: www.elsevier.com/locate/mechmt

Research paper

A general nonlinear order-reduction method based on the

referenced nodal coordinate formulation for a flexible
multibody system
Tengfei Yuan a, Wei Fan b, Hui Ren b, *
a
School of Astronautics, Harbin Institute of Technology, Harbin, 150001, PR. China
b
Institute of Aerospace Vehicle Dynamics and Control, School of Astronautics, Harbin Institute of Technology, Harbin, 150001, PR. China

A R T I C L E I N F O A B S T R A C T

Keywords: An accurate and efficient formulation is favorable for dynamic analysis and control in the field of
Nonlinear order-reduction method the flexible multibody system. This paper proposes a general nonlinear order-reduction method
Referenced nodal coordinate formulation that can tackle overall motions and large deformations with a significant decrease in degrees of
Modal derivatives
freedom, by incorporating the modal derivative techniques into the referenced nodal coordinate
Flexible multibody system
formulation (RNCF) developed earlier. To compute the modal derivatives straightforwardly, the
High-speed rotation
closed-form expression of the tangent stiffness matrix’s derivative is obtained. The elastic forces
are expressed as cubic polynomials of modal coordinates, such that the geometric nonlinear
deformations are explicitly expressed. The effectiveness of the proposed method is validated by
several numerical examples. The presented geometrically nonlinear order-reduction method can
achieve a great accuracy with a much fewer number of generalized coordinates, and it also in­
herits the large time step-sizes of the RNCF in dealing with large deformations and high-speed
rotations.

1. Introduction

Flexible multibody dynamics have recently been successfully implemented in engineering fields such as vehicle and aerospace
engineering, where high-speed rotations and large deformations frequently occur. Much more computational resources are required to
simulate the multibody system while the system becomes more and more complex. Those formulations are not suitable for control
design either. An accurate and efficient low-order formulation of a flexible system undergoing large overall motions and deflections is
highly required in multibody system dynamics.
The absolute nodal coordinate formulation (ANCF) has been successfully applied to the large deformation analysis of multibody
systems over the past two decades. The ANCF was developed using the global positions and position gradients as the generalized
coordinates [1]. Accordingly, it has a constant mass matrix, and the centrifugal and Coriolis forces are not exhibited explicitly, which
has an advantage in terms of numerical integration. Hence, the ANCF was deeply studied and widely used in many applications [2–5].
Although this formulation provides a general modeling framework for flexible multibody systems, the element discretization generally
involves a large number of degrees of freedom (DOFs), which makes the computational cost very expensive. Therefore, an
order-reduction method based on the ANCF is still needed to further reduce the DOFs and increase efficiency.

* Corresponding author.
E-mail address: [email protected] (H. Ren).

https://doi.org/10.1016/j.mechmachtheory.2023.105290
Received 2 January 2023; Accepted 8 February 2023
Available online 14 March 2023
0094-114X/© 2023 Elsevier Ltd. All rights reserved.
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Nomenclature

A The rotational matrix of the reference frame o-xyz.

aα The αth acceleration vector of the flexible body, which is described in the reference frame o-xyz.
C The constraint equations of the system.
ez The unit vector of the Z axis, i.e., ez = [ 0 0 1 ]T .
F The external force with respect to the position coordinates r in the RNCF, as defined in Eq. (12).
f The distributed force applied to the flexible body, which is described in the frame o-xyz.
Fαe The generalized elastic force of the flexible body with respect to the αth elastic coordinates vector.
Fei The ith modal elastic force in the NR-RNCF, as defined in Eq. (23).
fi The ith component of modal elastic force in the equilibrium configuration.
H The set of time instants during the simulation.
I3 The 3 × 3 identity matrix.
Iij The (i, j) entry of the inertia matrix in the NR-RNCF.
i, j, k, l Indexes of modal coordinates.
J The current rotatory inertia tensor of the flexible body in the NR-RNCF.
̂
J The constant rotatory inertia tensor of the flexible body in the equilibrium configuration.
K|eq The tangent stiffness matrix in the equilibrium configuration, as defined in Eq. (27).
Kij The constant second-order tensor in the modal elastic force.
Km The total number of modal coordinates.
Lijk The constant third-order tensor in the modal elastic force.
M(Mαβ ) The total mass matrix of the flexible body assembled from Mαβ I3 ; and its (α, β) entry.
m The total mass of the flexible body.
Nijkl The constant fourth-order tensor in the modal elastic force.
n The total number of the elastic coordinates vectors.
p (pi ) The vector of modal coordinates; and its ith component.
Qα The external force of the flexible body with respect to the αth elastic coordinates vector in the RNCF.
Qi The ith modal force in the NR-RNCF.
Qμν The second-order tensors related to the state variables.
qα The αth elastic coordinates vector of the flexible body in the reference frame o-xyz.
̂α
q The αth vector of the undeformed elastic coordinates.
qoα The αth vector of the elastic coordinates in the equilibrium configuration.
Δq The increments of the elastic coordinates around the equilibrium configuration.
r The position vector of the reference frame o-xyz.
rp The position vector of a material particle p, defined in the global inertia frame O-XYZ.
roc The centroid position vector of the flexible body at the equilibrium configuration.
Sα The αth shape function with respect to the αth elastic coordinates vector.
t The time variable.
u The position vector of a material particle p, which is described in the reference frame o-xyz.
V The volume of the flexible body.
Wij The weight associated with the modal derivative θij, as defined in Eq. (39).
w The vector of generalized displacements from the RNCF.
w̄ The vector of generalized displacements from the NR-RNCF.
x The material coordinates of the particle
X, Y, Z Three axes of the global inertia frame O-XYZ.
x, y, z Three axes of the non-inertial reference frame o-xyz.
α, β, μ, ν Indexes of elastic coordinates vectors of the flexible body.
γi , ηi Two constant 3 × 1 inertia vectors in the NR-RNCF, defined in Eq. (20).
ηj The modal amplitude with respect to the jth linear vibration mode.
Θ The matrix formed by the modal derivatives.
θij The modal derivatives, as defined in Eq. (31).
καβμν The constant fourth coefficient tensor in the generalized elastic force.
λ The Lagrange multipliers vector related to the constraint equations.
ρ The density of the flexible body.
ϱα The node mass parameter with respect to the αth elastic coordinates vector, as defined in Eq. (7).
φi The mode shape matrix in the NR-RNCF, as defined in Eq. (17).
Φ The matrix formed by the selected vibration modes.
ΦIB The constraint mode matrix, as defined in Eq. (47).
ϕi The ith vibration mode of the flexible body in the equilibrium, configuration, which is calculated at Eq. (28).

2
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

χ rel The global relative error, as defined in Eq. (40).

Ωi , Yij Two constant 3 × 3 inertia tensors in the NR-RNCF, as defined in Eq. (20).
ω, ω̇ The angular velocity and angular acceleration of the reference frame o-xyz, respectively.
ωi The ith natural frequency of the flexible body in the equilibrium configuration, which is calculated at Eq. (28).
ωs The steady rotational speed of the angle driver function
Ψ The reduced order basis matrix.
ψiα The ith column of Ψ with respect to the αth vector.

Abbreviation
ANCF The absolute nodal coordinate formulation
DOF The degree of freedom
MD The modal derivative
MMI The maximum modal interaction
RNCF The referenced nodal coordinate formulation
SVD The singular value decomposition
VM The vibration mode

To reduce the computational cost, several approaches, including modal and non-modal based techniques, were proposed for model
reduction. In recent years, component mode synthesis techniques have been extensively studied to reduce the dimensionality of models
in the field of multibody system dynamics [6–13], as reviewed by Sonneville et al. [6] and Géradin et al. [7]. Ellenbroek and Schilder
[8,9] incorporated the Craig-Bampton method with the absolute interface coordinates floating frame of reference formulation to
improve the convergence and efficiency of the simulation. Zwölfer and Gerstmayr [10] applied the component mode synthesis to the
node-based floating frame of reference formulation to reduce the model order and simplify the calculation process. Vidoni et al. [11,
12] extended the equivalent rigid-link system with the component mode synthesis to develop a novel dynamic model of spatial flexible
mechanisms. Kobayashi et al. [13] adopted the Craig-Bampton method to achieve the model order reduction for the planar ANCF beam
in the case of small axial deformations. Similarly, Otsuka and Makihara [14] presented a reduced-order ANCF thin plate model based
on the assumption of small in-plane deformations. However, the small deformation assumption restricts the application of those
modal-based methods in ANCF.
To extend the modal reduction technique to large deformations, modal reduction algorithms were proposed by Tang et al. [15] and
Tian et al. [16] based on successively local linearization for the flexible multibody system described by ANCF. However, continuous
monitoring of relative errors was required to ensure accuracy, and the reduction basis was updated successively at each linearization
point during the dynamic simulation. The main limitation of the approach is that the reduction basis changes as the linearization point
changes, bearing the drawback of frequently recalculating the modal basis.
Furthermore, non-modal order reduction methods, such as the proper orthogonal decomposition, were exploited to achieve model
order reduction for ANCF models [17–20]. The proper orthogonal decomposition method and its variants were extended from the
planer ANCF element [17,18] to the spatial element [19,20]. Luo et al. [19] adopted this method dramatically decreasing the
calculation time while maintaining accuracy. However, as a statistical method, the entire simulations of the full model need to be
conducted in advance to obtain sufficient data for reduction. In addition, the proper orthogonal decomposition method is generally
sensitive to the parameters and simulation scenarios of the full-order model.
The projection of the elastic coordinates of a structure into a set of modal coordinates is a standard technique for linear structural
dynamics. However, this standard modal reduction technique cannot be straightforwardly applied to the ANCF due to the rigid
motions and the nonlinearity of its elastic forces [21]. Ren and Yang [22] developed a referenced nodal coordinate formulation (RNCF)
by presenting the ANCF in a non-inertial reference coordinate system. The dynamic configuration of a flexible body is described as the
superposition of the gross motion of the reference frame and the deformation defined in the reference frame. Since the deformation is
separated, then the projection techniques can be adopted. The RNCF approach is highly effective in simulations of flexible body
systems undergoing large-distance travel, high-speed rotation, and large deformation [22,23]. Furthermore, the RNCF can have large
time step-sizes by the adoption of the Lie group technique, resulting in higher efficiency than the ANCF, especially for high-speed
rotations [22].
As a state-of-the-art reduction method in recent years, modal derivatives (MDs) have received much attention. The MDs are
computed through a perturbation analysis of the eigenvalue problem, which is a simulation-free approach [24], and they have been
successfully applied to solve nonlinear problems. Idehlson and Cardona [25,26] first introduced MDs to enrich the modal basis to solve
nonlinear structural dynamics problems, and Weeger et al. [27] investigated their approximation properties. Wu and Tiso[28,29]
enriched the classical linear basis, such as the Craig-Bampton basis and the standard Rubin basis, with MDs to study the dynamics of
flexible multibody systems with large rotations based on the floating frame of reference formulation. Later, this approach was
incorporated into the multiple-scale method to study temperature-dependent nonlinear systems [30], and it was applied to the dy­
namics of geometrically nonlinear structures subjected to contact interactions [31]. Furthermore, the MDs were also used to achieve
model order reduction for shape imperfections and defective structures [32,33]. To the best of the authors’ knowledge, the MDs are not
yet used in combination with ANCF or RNCF to achieve the modal order reduction.
In this paper, a nonlinear order-reduction method based on the RNCF, which is referred to as the NR-RNCF, is developed for flexible
multibody systems. The vibration modes (VMs) appended with the MDs are used as the extended modal basis, and it is the first time to

3
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

combine the RNCF with the MDs for model reduction. The nonlinear strain-displacement relationship is considered, and the geometric
nonlinearity is retained in the NR-RNCF. The nonlinear elastic modal force is expressed as cubic polynomials of modal coordinates, and
corresponding constant coefficient matrices can be calculated in advance to increase efficiency. Moreover, the procedure of calculating
MDs is simple and effective. The generalized mass matrix is constant, and the derivative of the stiffness matrix with respect to the
modal amplitudes is analytically expressed. A simple selection criterion is used to select the most significant MDs as the basis without
computing their actual shapes. The NR-RNCF inherits the advantages of the RNCF, and it also can achieve the sufficient accuracy with
fewer degrees of freedom and has higher efficiency than the RNCF, which makes the reduced-order model favorable for dynamic
analysis and control of flexible multibody systems.
The organization of this work is as follows. The NR-RNCF is described in Section 2. The procedure for forming the reduction basis is
discussed in Section 3. The effectiveness of the NR-RNCF is validated by numerical examples in Section 4. Then some comments and
conclusions are presented in Section 5. Some computational details are given in the appendix to make this work self-sustained.

2. Description of the NR-RNCF

This part aims to describe the detailed procedure of the NR-RNCF. Thus, a brief description of the RNCF is presented in Section 2.1,
and the NR-RNCF is then derived in Section 2.2.

2.1. A brief introduction of the RNCF

In the RNCF description [22], the absolute position of a material particle can be described as the superposition of the motion of the
non-inertial reference coordinate system o-xyz and the position of the particle within the frame o-xyz, as shown in Fig. 1.
The translational and rotational motions of the frame o-xyz can be depicted by the position vector r(t) and rotational matrix A(t) in
the global inertia frame O-XYZ, respectively. Supposed that the position vector of a material particle with respect to the frame o-xyz is
u(x, t), where x denotes the material coordinates of the particle p. Hence, the corresponding position vector in the global frame O-XYZ
can be expressed as
rp (x, t) = r(t) + A(t)u(x, t) (1)

In the reference frame o-xyz, the spatial discretization of the flexible body can be executed as that in the ANCF, which can be written
as
u(x, t) = Sα (x)qα (t), α = 1, 2, ⋯, n (2)

where the Einstein summation convention is adopted here, i.e., (·)α and (·)α represent summation; qα ∈ R3×1 is the nodal position or
slope vector; Sα is the associated shape function; n is the number of generalized coordinate vectors in the ANCF elements. Substituting
Eq. (2) into Eq. (1) yields
rp (x, t) = r(t) + Sα (x)A(t)qα (t), α = 1, 2, ⋯, n (3)

Therefore, the generalized coordinates of the flexible body in the RNCF are composed of r(t), A(t), and qα (t). Suppose that the
distributed force applied to the flexible body is f(x, t), which is defined in the frame o-xyz. Then, the governing equations of the flexible
body can be derived from the Lagrange equations [22], which consist of three parts: equations for translational motions
( )T
∂C
mr̈ + ϱα Aaα + λ=F (4)
∂r

Fig. 1. Descriptions of position vectors in the RNCF.

4
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

equations for rotational motions

( )T
( ) ∂C
ϱα qα × AT r̈ + M αβ qα × aβ + λ = qα × Qα (5)
∂π

and equations for deformations

( )T
∂C
ϱα AT r̈ + M αβ aβ + Fαe + λ = Qα , α = 1, 2, ⋯, n (6)
∂qα
The inertia parameters in Eq. (4) ~ Eq. (6) are defined as follows:
∫ ∫ ∫
m= ρ(x)dv, ϱα = ρ(x)Sα (x)dv, M αβ = ρ(x)Sα (x)Sβ (x)dv (7)
V V V

in which V is the volume of the flexible body and ρ is the density. The acceleration vector aα is
aα = q̈α + 2ω × q̇α + ω̇ × qα + ω × (ω × qα ) (8)

⋅
where ω(t) and ω(t) is the local angular velocity and angular acceleration of the non-inertial frame o-xyz, respectively. C and λ denote
the constraint equations of the system and the corresponding Lagrange multipliers, respectively. Note that the expression for Fαe is the
same as that in the ANCF, and detailed derivations for elastic forces of beams and plates can be found in Refs. [22] and [34],
respectively. The generalized elastic force Fαe is expressed by

Fαe = καβμν Qμν qβ , α, β, μ, ν = 1, 2, ⋯, n (9)

where καβμν is a constant four-order tensor, and

1( )
Qμν (t) = q (t) ⋅ qν (t) − ̂ ̂ν
qμ ⋅ q (10)
2 μ

in which q
̂ α are the undeformed elastic coordinates. The Jacobian matrix corresponds to Eq. (9) is

∂Fαe
= καβμν Qμν I3 + καμβν qμ qTν (11)
∂qβ
In addition, the generalized external forces corresponding to the generalized coordinates r(t) and qα (t) are
∫ ∫
F(t) = A(t) f(x, t)dv, Qα (t) = Sα (x)f(x, t)dv (12)
V V

2.2. Governing equations of the NR-RNCF

To improve computational speeds, the model order reduction methodology is naturally introduced to reduce the degrees of
freedom. In this section, the reduced equations are obtained via a classical Galerkin projection by projecting the incremental of the
elastic coordinates q around the equilibrium configuration on a time-independent reduced order basis Ψ, as
q(t) = qo + Ψp(t) (13)

where qo are the elastic coordinates at the equilibrium configuration and p(t) is the vector of modal coordinates. The component form
of Eq. (13) can be written as

qα = qoα + ψiα pi , i = 1, 2, ⋯, Km (14)

where Km is the number of the modal coordinates, ψiα is the ith column of the reduced order basis Ψ with respect to the αth vector, and
pi is the corresponding modal coordinate. Substituting Eq. (14) into Eq. (8) yields
⋅ ( ) [ ( )]
aα = ψiα p̈i + 2ω × ψiα ṗi + ω × qoα + ψiα pi + ω × ω × qoα + ψiα pi (15)

Substituting Eqs. (14) and (15) into the Eq. (4) yields
( )T
⋅ [ ( )] ∂C
mr̈ + A(ω × c̄) + Aφi p̈i + A ω × 2φi ṗi + ω × c̄ + λ=F (16)
∂r

where

c̄ = mroc + φi pi , mroc = ϱα qoα , φi = ϱα ψiα (17)

5
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

and roc is the centroid position vector of the flexible body at the equilibrium configuration. Substituting Eqs. (14) and (15) into Eq. (5)
yields
( )T
( ) ( ) ( ) ∂C ( )
(18)
⋅
c̄ × AT r̈ + Jω + γi − ηij pj p̈i + ω × (Jω) + 2ṗi Ωi + Yij pj ω + λ = qoα + ψiα pi × Qα
∂π

where the current rotatory inertia tensor

( )
J=̂ J + Ωi + ΩiT pi + Yij pi pj (19)

and corresponding constants are

T
̂ T
J = M αβ q̃oα q̃oβ ,
T ̃i
Ωi = M αβ q̃oα ψ ̃j ψ
Yij = M αβ ψ ̃i , γi = M αβ qo × ψi , ηij = M αβ ψiα × ψjβ (20)
β, α β α β

where the “~” denotes the skew operator. Given a three-dimensional vector a = ( a1 , a2 , a3 )T , one has
⎡ ⎤
0 − a3 a2
̃ ⎣
a = a3 0 − a1 ⎦ (21)
− a2 a1 0

Substituting Eqs. (14) and (15) into Eq. (6), and using the Galerkin projection, i.e., making a dot product with respect to ψiα , yield
( )T
( )T ( ) ∂C
φiT AT r̈ + γi − ηij pj ω̇ + I ij p̈j − 2ωT ηij ṗj − ωT Ωi + Yij pj ω + Fei + λ = Qi , i, j = 1, 2, ⋯, Km (22)
∂pi

j
where the mass matrix Iij = Mαβ ψiα ⋅ ψβ . The nonlinear elastic modal force can be expressed as

Fei = ψiα ⋅ Fαe (23)

and Qi = ψiα ⋅ Qα . Note that the geometrically nonlinear elastic modal force Fei can be explicitly expressed as a cubic polynomial of
modal coordinates by substituting Eq. (14) into Eq. (23), which is

Fei = f i + K ij pj + Lijk pj pk + N ijkl pj pk pl (24)

The detailed derivation and expression of Eq. (24) is provided in Appendix A. Note that f i , Kij , Lijk , and Nijkl are constants and can be
computed in advance. The corresponding Jacobian matrix can be written as

∂Fei ( ) ( )
= K ij + Lijk + Likj pk + N ijkl + N ikjl + N ilkj pk pl (25)
∂pj

3. Reduction basis

The key to a promising reduction method is to find a suitable basis Ψ that can reproduce the full solution, i.e., the RNCF model, with
good accuracy. In this section, the construction of the reduction basis will be discussed in detail.
The vibration mode superposition method was widely used in linear dynamics problems. However, in geometrically nonlinear
systems, linear vibration modes usually fail to accurately capture the nonlinear response of the system with the increasing deviation
from the linearization point [28]. To overcome this difficulty, the MDs stemming from vibration modes can be appended to the linear
reduction basis, as shown in [25,26]. Therefore, the reduction basis is constructed by vibration modes calculated around a given
equilibrium position and the corresponding MDs in this paper.

3.1. Linearization and eigenvalue analysis

Before conducting modal reduction, eigenvalue analysis is required to obtain the linear modal basis used in the subsequent pro­
cedure, which is performed around an equilibrium configuration. In the following, the linearization around a rigid-reference
configuration will be discussed. A rigid-reference configuration is obtained when all the increments of the generalized elastic co­
ordinates, as well as all velocities and the external forces applied to the system, are zero [35]. The linearization of Eq. (6) around the
rigid-reference configuration leads to
MΔq̈ + KΔq = 0 (26)

where M is the mass matrix, K is the tangent stiffness matrix, and Δq is the increments of the generalized elastic coordinates around the
equilibrium configuration. According to Eq. (6), the mass matrix M and stiffness matrix K can be expressed as

6
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

∂Fαe ( o )
M = M αβ I3 , K|eq = q (27)
∂qβ β
Note that the mass matrices M must be constant in the RNCF, and the stiffness matrices K|eq is a symmetric positive definite matrix
at the equilibrium configuration.
As reported in [36], the linearized modal equations of motion of a flexible multibody system can be derived from the homogeneous
form of Eq. (26),
( ⃒ )
K⃒eq − ω2i M ϕ i = 0 (28)

where ωi is ith eigenfrequency, and ϕi is the corresponding eigenvector.

3.2. Derivation of modal derivatives

In geometrically nonlinear systems, a reduction basis formed by linear vibration modes only usually fails to capture the large
deflection. To cope with this problem, modal derivatives (MDs) can be used to enrich the existing linear reduction basis to represent the
effects of nonlinearity [25–29].
As discussed in Refs [27–29], it is reasonable to assume that the displacement vector q is C2-differentiable with respect to the modal
parameter η in geometrically nonlinear systems. Thus, the generalized elastic coordinate vector q can be expanded into a Taylor series
around the equilibrium configuration (i.e. η = 0),
∑k
∂q ⃒⃒ 1∑ k ∑ k
∂2 q ⃒⃒ ( )
q = qo + eq ηi + eq ηi ηj + O ‖ η ‖
3
(29)
i=1
∂ηi 2 i=1 j=1
∂η i ∂ηj

and the derivative of the generalized elastic coordinate vector q with respect to the modal amplitude ηi at the equilibrium configuration
is as follows
∂q ⃒⃒
=ϕi (30)
∂ηi eq

where ϕi is the ith vibration mode, as could be expected from previous results. Therefore, the second derivative, i.e., the modal de­
rivative, can be expressed as

∂2 q ⃒⃒ ∂ϕ i ⃒⃒
= = θij (31)
∂ηi ∂ηj eq ∂ηj eq

i.e., the modal derivative θij reveals the variation of a certain mode ϕi when the structure is displaced according to the shape described
by the mode ϕj. According to Eq. (29), q can be approximated by the second-order Taylor expansion as
∑
k
1∑ k ∑ k
q ≈ qo + ϕ i ηi + θij ηi ηj (32)
i=1
2 i=1 j=1

As discussed in [25,26], MDs are computed by differentiating the nonlinear eigenvalue problem (see Eq. (28)) with respect to the
modal amplitude. This procedure yields a more complicated expression since the coefficient matrix is not only singular but needs to be
factorized for each eigenfrequency. Therefore, a simplified format that neglects mass-related terms was also presented [25,26]. Thus,
MDs can be calculated by
⃒ ∂ϕ i ∂K ⃒⃒
K⃒eq =− eq ϕ i (33)
∂ηj ∂ηj
Note that (∂K/∂ηj)|eq is the derivative of the tangent stiffness matrix with respect to the modal amplitude ηj, which is obtained by
giving the system a displacement in the direction of ϕj, as
( )
∂K ⃒⃒ ∂K q = qo + ϕj ηj ⃒⃒
eq = η =0 (34)
∂ηj ∂ηj j

In practice, the computation of (∂K/∂ηj)|eq can be performed using a finite difference scheme. However, the finite difference
approximation to obtain the derivatives can be very sensitive with respect to the step size [37]. Fortunately, the computational of
(∂K/∂ηj)|eq in the RNCF is simple and efficient because it can be expressed analytically as
[ ( )] [( )]
∂K ⃒⃒ αβμν 1
(35)
T
eq = κ qoμ ⋅ ϕjν + ϕjμ ⋅ qoν I3 + καμβν qoμ ϕjν + ϕjμ qoν T
∂ηj 2

The detailed derivation of Eq. (35) is provided in Appendix B. Note that the expression of Eq. (35) is similar to Eq. (11), so its
calculation is straightforward. Moreover, the generality of the RNCF makes Eq. (35) widely available for beams, plates, or more

7
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

complex structures. Therefore, the MDs can be derived from

∂ϕ i ( ⃒ )− 1 ∂K ⃒
θij = = − K⃒eq ⃒ ϕ (36)
∂ηj ∂ηj eq i

in which (K|eq)− 1 exists as rigid body motions are suppressed, and the matrix of coefficients K|eq can only be factorized once for all. It
should be noted that the MDs are symmetric, i.e., θij = θji. Therefore, a total of k(k + 1)/2 MDs can be obtained with k given modes.
To take full advantage of invariant parameters in Section 2.2, Eq. (32) will not directly apply in this study. Instead, the MDs will be
used as the augmentation of the linear basis to capture the effects of nonlinearity, that is
( )
η
q = qo + Φη + Θξ = qo + [ Φ Θ ] = qo + Ψp (37)
ξ

where
( )
η
Ψ = [ Φ Θ ], p= (38)
ξ

in which Φ is the matrix consisting of selected vibration modes, Θ is the matrix formed by the related modal derivatives, and η and ξ
are the corresponding modal coordinate vector, respectively, both of which form the generalized modal coordinate vector p.

3.3. Optimization of the reduction basis

In practice, not all MDs are necessary to capture the nonlinear response of the system. Therefore, some selection criteria [38,39] can
be used to determine the best reduced MDs set for a given problem in advance. The maximum modal interaction (MMI) [39] has proven
simple and efficient in selecting MDs. The most significant contributions can be estimated prior to the actual calculation of the MDs.
Hence, it is adopted here to pick the most significant MDs.
The MMI criterion selects the MDs featured by the maximum interaction to augment the linear basis. The modal interaction is
quantified by the product of two arbitrary modal amplitudes of these VMs during a linear simulation. Therefore, the weight defined by
the modal interaction can be written as
∫T
⃒ ⃒
Wij = ⃒η (t)η (t)⃒dt
i j (39)
0

where Wij represents the weight associated with the modal derivative θij, and ηi is the response of the ith linear vibration mode for a
given external load over a time span [0, T]. Since this procedure involves only the linear modal superposition, the MMI matrix is
extremely cheap to obtain [39]. The most relevant MDs will be the ones with the highest values of the corresponding Wij coefficients in
the MMI matrix.
Therefore, the reduction basis Ψ can be further condensed, which can effectively reduce the nonlinear system and establish an
accurate reduced model. Moreover, it should be noted that the MDs are insufficient to guarantee orthogonality and linear indepen­
dence with respect to the linear basis. Thence, the condition number of the reduction basis may be bad. Hence, the singular value
decomposition (SVD) is necessary to deflate the reduction basis to avoid bad conditioning or even singularities in the reduced model,
and a numerically stable reduction basis is obtained [37]. The detailed procedure is shown in Algorithm 1, where s is the number of
columns of the matrix Ψ, i.e., the number of critical VMs and MDs initially selected.
Accordingly, the flowchart of the whole procedure of constructing the reduction basis is presented in Fig. 2.

4. Numerical examples

In this section, the dynamic simulation of flexible body systems with high-speed rotation and large deformation will be presented to

Algorithm 1
The deflation technique via Singular Value Decomposition.
[ ]
Input: the reduction basis: Ψ = ϕ 1 , ⋯, ϕ k , ⋯, θij , ⋯ ∈ R3n×s , and the small tolerance ϑ = 1 × 10− 6
in this work.
Output: the numerically stable reduction basis Ψ.
1. svd(Ψ) = UΣVT , where U ∈ R3n×s , Σ ∈ Rs×s
2. Σ = diag[ σ1 , σ2 , ⋯, σs ], with σ1 > σ2 > ⋯ > σs > 0
3. for i → s to 1 do
4. if σi > ϑ ⋅ σ1
5. break
6. end if
7. end for
8. Km ← i
9. Ψ ← U(:, 1 : Km )

8
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 2. The flowchart of the construction of the reduction basis.

validate the applicability and reliability of the proposed NR-RNCF. The overall computation flow of the NR-RNCF is shown in Fig. 3,
and all the simulations are performed with GNU Octave 6.2.0 on a PC having a Core i7–9700 K processor and 16GB RAM.
As illustrated in Fig. 3, the invariant parameters can be calculated in the pre-processor part, and the detailed procedure can be
found in Section 2.2. Note that the pre-processor part is excluded from the complexity evaluation since these constant coefficient
matrices are calculated only once before the start of the dynamic simulation. The governing equations are solved by the adaptive-
stepsize generalized-α integrator [40,41] and the computational time is averaged from five repeated calculations in the following

Fig. 3. The overall computation flow of the proposed method.

9
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

numerical examples.
To take a comprehensive comparison, the solution of RNCF [22] is taken as a reference solution. Moreover, to assess the
displacement error of the NR-RNCF compared with the RNCF, the global relative error χ rel of the displacement is defined as following
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑H ̅
T
i=1 [w(ti ) − w̄(ti )] [w(ti ) − w̄(ti )]
χ rel = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑H × 100% (40)
T
i=1 w(ti ) w(ti )

where w(t) is the vector of generalized displacements at the time t obtained from the RNCF model, w̄(t) is the vector of generalized
displacements based on the NR-RNCF, and H is the set of time instants during the simulation.

4.1. High-speed rotating systems

For the rotating system, the position vector and the local coordinate frame at the rotating shaft are usually selected as the reference
coordinate system. Suppose the position vector of the reference coordinate system is r(t) = 0 and the rotation motion of the frame is
prescribed. Hence, the governing Eq. (6) of the RNCF become
[ ]
( )
M q̈β + 2ω × q̇β + ω̇ × qβ + ω × ω × qβ + Fαe = Qα
αβ
(41)

and the corresponding Eq. (22) of the NR-RNCF become

( )T ( )
I ij p̈j − 2ωT ηij ṗj + γi − ηij pj ω̇ − ωT Ωi + Yij pj ω + Fei = Qi (42)

the coefficient matrices in Eq. (42) is constant, as shown in Section 2.2. Suppose the rotation motion is about the Z-axis, with the
following angle driver
⎧ [ ( )2 ( )]
2
⎪
⎪ ωs t + Ts 2πt
⎪
⎪ cos − 1 , 0 ≤ t ≤ Ts
⎨ Ts 2 2π Ts
θ(t) = ( ) (43)
⎪
⎪
⎪
⎪ T
⎩ ωs t − s , t > Ts
2

where the values of ωs and Ts depend on specific numerical examples. The corresponding angular velocity and acceleration are ω =
⋅
θ̇(t)ez and ω = θ̈(t)ez , respectively.

4.1.1. The spin-up maneuver of the straight beam

The rotating beam is a benchmark example that has been used to examine the geometric nonlinearity and the dynamic stiffening
effect in many previous works [42–44], as shown in Fig. 4. Suppose the cross-section of the rotating beam is hollow and circular, and
the material is uniform and isotropic, the geometry and material parameters of the beam are listed in Table 1.
Suppose the shaft rotates about the Z-axis, and Eq. (43) is adopted as the angle driver function with Ts = 15 s. Four different steady
rotational speeds, with ωs = 4, 10, 20, and 40 rad/sec, are selected to highlight the accuracy and efficiency of the present method at
different rotational speeds.
The simulation is carried out by the ANCF [45], the RNCF [22], and the current formulation for comparison. As discussed in [22],

Fig. 4. Schematic of the rotating beam.

10
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Table 1
Parameters of the rotating beam.
Parameter Symbol Value

Length (m) L 8.0

Cross-sectional area (cm2) A 0.7299
Moment of inertia (cm4) I 0.8215
Density (kg/m3) ρ 2766.67
Young’s modulus (GPa) E 68.95
Poisson’s radio ν 0.30

more elements are required to capture the large deformations with the steady rotational speed increase, and the number of elements
required for each steady rotational speed is listed in Table 2. Note that the numbers of beam elements in the ANCF and RNCF are the
same as that in [45] for a certain steady rotational speed.
In the NR-RNCF formulation, with the rotational speed increases, the deformation of the slender beam becomes larger and the
geometric nonlinearity becomes significant, more VMs are required to construct the reduced order basis. The reduced order basis is
constructed as illustrated in Section 3, and the MDs are selected according to the criterion defined by Eq. (39). The weights Wij
associated with the different MDs are shown in the bar plot on Fig. 5. After the normalization operation, the weights are raised to 0.25
power to better show the relative contributions on the bar plot, and only the upper triangle is shown as the matrix is symmetric.
In these four scenarios, all the inertia forces and the elastic forces are always in the X-Y plane, and there are no out-of-plane de­
flections. As shown in Fig. 5, some vibration modes do not contribute since they feature bending motion out of the X-Y plane, and these
modes are not selected into the reduction basis. In other words, the most important VMs can be picked by using the maximum modal
interaction method. Thence, the critical VMs and the MDs with the highest weights (burgundy) are selected. The reduction basis of the
straight beam for four different rotational speeds is also shown in Table 2. The deflections at the tip of the rotating straight beam are
depicted in Fig. 6, where the tip deflections are calculated in the reference frame.
As shown in Fig. 6, the deflections are amplified with the steady rotational speed increase. In the transient state, the deflections are
large since the rotational inertia force is imposed. In the steady state, the angular acceleration disappears and the fluctuation frequency
of the transverse excursions grows with the increase of the steady rotational speeds due to the Coriolis forces and elastic forces.
Note that the first natural frequency of the slender beam is 2.9096 rad/sec, and the considered angular velocities are all higher than
the first critical speed. Therefore, this example is a typical high-speed rotating system. The above results obtained from the three
formulations are in good agreement. In the NR-RNCF, some small deviations can be observed in the steady state as the rotational speed
is higher, while the large deflections in the transient state still close perfectly. When ωs is 40 rad/sec, the maximum transverse
deflection is 2.3309 m in Fig. 6(h), which is almost 30% of the length of the beam and is already a large deformation, and the results are
still highly consistent. Moreover, the comparison of the number of DOFs, adaptive steps, computational times, and the global relative
error for the three methods is as Table 3.
In Table 3, the results obtained from the ANCF and the RNCF are basically the same, but the computational times are significantly
reduced in the RNCF. The results between the NR-RNCF and the RNCF are also in good agreement, and the maximum global relative
error is less than 1.5%.
As the rotating speed increases, the NR-RNCF can maintain sufficient accuracy by adding a few VMs and MDs, the number of which
is much smaller than that in the ANCF and the RNCF. Moreover, it is worth noting that the adaptive steps required by the NR-RNCF are
close to the steps in the RNCF, which are almost the same for different rotational speeds, and both are less than those in the ANCF.
Therefore, the NR-RNCF inherits the large time step-sizes of the RNCF, resulting in higher efficiency.
When ωs is 40 rad/sec, the time consumption of the ANCF, the RNCF, and the NR-RNCF is 203.78 s, 12.71 s, and 1.56 s, respectively,
and the time radios between the NR-RNCF and the ANCF, and between the NR-RNCF and the RNCF are approximately 1/130 and 1/8,
respectively. The results demonstrate that the NR-RNCF can dramatically reduce the computation time and achieve a higher accel­
eration on large-scale computational problems.

4.1.2. The spin-up maneuver of the curved beams

As a benchmark example, the rotating straight beam has been studied in much research [42–44], but rare works exist about the
rotating curved beam. Typically, a large slender beam usually exhibits a coupling of bending and axial displacements when the
geometric nonlinearity is excited. The corresponding bending-stretching coupling could be in principle provided by adding some
high-frequency axial VMs to enrich the reduced order basis containing some low-frequency bending modes [28,46]. However, the

Table 2
The number of degrees of freedom of three formulations for four different steady rotational speed.
ωs (rad/sec) ANCF RNCF NR-RNCF
No. Elements DOFs No. Elements DOFs VMs MDs DOFs

4 6 111 6 111 3 6 9
10 10 183 10 183 4 10 14
20 16 291 16 291 5 12 17
40 24 435 24 435 6 16 22

11
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 5. Maximum modal interaction matrix relative to four different steady rotational speeds: (a) 4 rad/sec, (b) 10 rad/sec, (c) 20 rad/sec, and (d) 40
rad/sec of the spin-up maneuver of the straight beam.

extraction of axial and bending VMs can be challenging for curved beams. Fortunately, this nonlinear effect can be captured by the
MDs.
In this subsection, three types of curved beam configurations are studied. Note that the main focus of these three examples is to
show the effectiveness of the current approach, so only one steady rotating speed is shown.
The beam can bend in the opposite direction of the rotation (upwind), as shown in Fig. 7(a); in the same direction of the rotation
(downwind), as in Fig. 7(b), or bend out of the plane, as in Fig. 7(c). Supposed the cross-section of all three curved beams is hollow and
circular, and the material is uniform and isotropic. The length of all three curve beams is 8 m, the initial curvature radius is 32/π m, and
all other parameters are the same as the straight beam.
The dynamics of all three curved beams with Ts = 15 s and ωs = 4 rad/sec are calculated by RNCF and NR-RNCF approaches. Note
that the first natural frequency of the curve beam is 2.9248 rad/sec, and the considered angular velocity is higher than the first critical
speed. In RNCF, 8 C1 beam elements described in [45] are adopted to discretize these curved beams, and the RNCF model consists of
144 degrees of freedom.
In the NR-RNCF, the reduced order basis is constructed as described in Section 3 for all three curved beams, which consists of the
first six VMs and a few MDs selected according to the criterion defined by Eq. (39). The maximum modal interaction matrix is shown in
the bar plot on Fig. 8.
As shown in Fig. 8(a) and 8(b), the MDs θ22, θ24, θ26, θ44, θ46, and θ66 are more critical for the upwind and downwind curved beam
intuitively. In these two configurations, all the inertia forces and the elastic forces are always in the X-Y plane, and there is no out-of-
plane deflection. Therefore, the first, third, and fifth modes are not involved because they are characterized by bending motion outside
the X-Y plane. As a result, these six MDs (burgundy) are selected to enrich the linear basis.
As shown in Fig. 8(c), all six modes have contribution since the out-of-plane deflection appears in the third configuration of the
curved beam. The MD θ11 is the most important, and the cross terms relating the first two VMs to all the other VMs are more important
than others. Thence, the first ten MDs with the highest weights (burgundy) are selected.
The reduction basis of the three curved beams for the NR-RNCF is shown in Table 4. The deflections at the tip of the rotating curve
beams are shown in Fig. 9, where the tip deflections are calculated in the reference frame, and the deflections with respect to its three

12
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig 6. Tip deflections of the rotating straight beam with the steady rotational speeds (a) and (b) 4 rad/sec, (c) and (d) 10 rad/sec, (e) and (f) 20 rad/
sec, (g) and (h) 40 rad/sec.

axes are labeled as u, v, and w, respectively.

The first two curved beams still execute planar motions, and there is no out-of-plane deflection. In these two scenarios, the results
calculated from both approaches are in good agreement, as shown in Fig. 9(a) and 9(b). The dynamics in the third scenario are more
complex. In the transient state, out-of-plane deflection emerges because the elastic forces have the z-component, while in the steady

13
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Table 3
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the three formulations for four
different speed.
ωs = 4 rad/sec ωs = 10 rad/sec
ANCF RNCF NR-RNCF ANCF RNCF NR-RNCF

DOFs 111 111 9 183 183 14

Adaptive steps 4050 2014 2319 9277 2014 2381
Time (sec) 6.31 3.09 0.58 22.17 4.98 0.72
χrel (%) 0.06 — 0.54 0.04 — 0.82

ωs = 20 rad/sec ωs = 40 rad/sec
ANCF RNCF NR-RNCF ANCF RNCF NR-RNCF

DOFs 291 291 17 435 435 22

Adaptive steps 18,127 2014 2459 35,903 2036 3076
Time (sec) 67.87 9.17 1.02 230.78 12.71 1.56
χrel (%) 0.08 — 0.97 0.08 — 1.41

Fig. 7. The (a) upwind, (b) downwind, and (c) out-of-plane configurations of the spin-up maneuver of curved beams.

state, the centrifugal forces are only in the x-z plane and the y-component of the deflection is zeros because the Coriolis force and the
rotational inertia force vanish. As shown in Fig. 9(c), the results still agree perfectly.
Note that the transverse deformations achieve approximately 1 m in all three curve beams, which are excited in the nonlinear
regime. The results indicate that the MDs are essential to describe the nonlinear bending-stretching coupling effect of the curve beams.
A comprehensive comparison of the two methods is presented in Table 5.
As shown in Table 5, the results from NR-RNCF yield excellent approximations to those from the RNCF, and the maximum global
relative error is only 0.86%. Although the out-of-plane configuration requires more MDs to achieve sufficient accuracy, the number of
generalized coordinates in the NR-RNCF is still much less than that in the RNCF. Moreover, the current formulation is about 6 times
faster than the RNCF. Note that the time steps required by the NR-RNCF in these three scenarios are 2280, 2280, and 2295, respec­
tively, which are close to the 2007 steps in the RNCF. Since the NR-RNCF can simulate with the large time step size while maintaining a
low scale DOFs, a higher efficiency can be achieved.

4.1.3. Rotating membrane

Rotating membranes are widely utilized in space flexible membranes antennas, flexible solar sail spacecraft, and other fields.
Because of the small stiffness and high-speed rotation, the geometric nonlinear effect of the structure needs to be considered in the
application process. As shown in Fig. 10(a), a circular rotating membrane is presented to demonstrate the accuracy of the NR-RNCF.
Supposed the thickness of the rotating membrane is uniform, and the material is isotropic. The geometry and material parameters
of the beam are listed in Table 6.
First, only the in-plane deformation is considered, i.e., wz = 0. Suppose the rotating membrane rotates in the X-Y plane, i.e., the
shaft rotates about the Z-axis, and the Z-axis coincides with the center of the membrane. The angle driver function is defined by Eq. (43)
with Ts = 2 s. Two different steady rotational speeds ωs are calculated, with ωs = 7 and 10 rad/sec. Note that the first natural frequency
of the rotating membrane is only 0.448 rad/sec, and the considered angular velocities are all higher than the first critical speed.
Therefore, this example is also a typical high-speed system.

14
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 8. Maximum modal interaction matrix relative to the first six VMs: (a) upwind, (b) downwind, and (c) out-of-plane configurations of the spin-
up maneuver of curved beams.

Table 4
Size of reduction order basis for rotating curved beam models.
the configuration of curved beams VMs MDs Total

upwind 3 6 9
downwind 3 6 9
out of plane 6 10 16

As shown in Fig. 10(b), the circular rotating membrane is modeled by the RNCF approach with 1350 Hermite quadratic strain
elements [34]. The whole model consists of 18,912 degrees of freedom, which often accompanies a massive computational cost.
In this case, the deformation of the rotating membrane only occurs within the plane, so the vibration mode only considers the in-
plane mode. For membrane structures, the in-plane modes are usually higher-order VMs because in-plane stiffness is much greater than
their bending stiffness. As shown in Fig. 11(a), only the 19th mode is in-plane mode among the first 100 VMs of the circular membrane.
Therefore, the 19th vibration mode and its corresponding unique modal derivative are selected to form the reduced-order basis of
the NR-RNCF. As shown in Fig. 11(b), it can be found that the shape of this modal derivative is also in-plane deformation. It is worth
noting that these two modes are orthogonal because the inner product of their mode vector is less than 10− 8. The deformations at the
edge of the rotating membrane are shown in Fig. 12, which are measured in the reference frame.
The numerical results reveal that the deformation of the rotating membrane grows with the steady rotational speed increase. The
edge’s deflection along the Y axis is about one order of magnitude higher than that along the X axis, which is a typical in-plane torsional
deformation like the mode shape of the 19th mode (Fig. 11(a)). However, a clear difference can be observed if only the 19th vibration
mode is applied as the modal basis in the NR-RNCF, while the results are in good agreement with those from the RNCF when the modal
derivative is adopted. The comparison of the number of DOFs, adaptive steps, computational times, and the global relative error
between the two methods is in Table 7.
As shown in Table 7, the global relative errors are significantly decreased with the addition of one MD, and the NR-RNCF can

15
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 9. Tip deflections of the (a) upwind, (b) downwind, and (c) out-of-plane spin-up maneuver with steady rotational speeds of 4 rad/sec,
calculated by the RNCF approach (dashed) and the NR-RNCF (double-dashed).

Table 5
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the two formulations with a
steady rotational speed of 4 rad/sec.
ωs = 4 rad/sec upwind curve beam downwind curve beam out-of-plate curve beam
RNCF NR-RNCF RNCF NR-RNCF RNCF NR-RNCF

DOFs 144 9 144 9 144 16

Adaptive steps 2007 2280 2007 2280 2007 2295
Time (sec) 4.072 0.635 4.102 0.636 4.132 0.797
χrel (%) — 0.48 — 0.47 — 0.86

achieve sufficient accuracy with only two generalized coordinates (1VM+1MD), while the RNCF method requires 18,912 degrees of
freedom. Therefore, the results confirm that the modal derivative technique is effective for flexible multibody systems with
geometrically nonlinear characteristics. In addition, the modal derivative effectively avoids the calculation and selection of higher-
order in-plane VMs, which are expensive for the membrane structure. Although the adaptive steps of NR-RNCF are slightly larger
than those in RNCF, the comparison of degrees of freedom is too disparate, thus the computational efficiency is significantly improved.
In the above example, the deformation mode is relatively simple. To further confirm the validity of the NR-RNCF in the deformation

16
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 10. (a) Schematic diagram of the rotating membrane, (b) its division with 1350 triangular elements.

Table 6
Parameters of the rotating membrane.
Parameter Thickness Radius Density Young’s modulus Poisson’s radio

Value 0.002 m 1.50 m 3200 kg/m3 2.07 GPa 0.30

Fig. 11. (a) The 19th vibration mode and (b) the corresponding modal derivative of the rotating membrane (the blue line denotes the undeformed
configuration, and the red line denotes mode shapes).

coupling problem, a bending-torsional coupling problem will be considered. As shown in Fig. 10(a), suppose the membrane is excited
by a sinusoidal displacement excitation along the Z-axis
wz = 0.012 ⋅ sin(0.25 ⋅ 2π ⋅ t) (meter)

while the membrane is also rotating about the Z-axis, and the angle driver function is defined by Eq. (43) with Ts = 2 s and ωs = 4 rad/
sec.
According to the characteristics of its external excitation, the 3rd and 8th order vibration modes, as shown in Fig. 13, need to be
added to the reduction basis to describe the corresponding out-of-plate deformation.
The mode shape of these two modes is similar to the beam’s low-frequency bending modes, and both of them are rotationally
symmetric. To capture the nonlinear response, the MDs related to these vibration modes also need to be enriched into the reduction
basis. The deformations at the edge of the rotating membrane are shown in Fig. 14, which are measured in the reference frame.
Results from the present formulation are highly consistent with those from the RNCF when the corresponding MDs are taken into
consideration. If only the VMs is adopted in the NR-RNCF, the results’ amplitude and trend have significant deviations relative to the

17
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 12. The deflections at the edge of the rotating membrane with the steady rotational speeds (a) and (b) ωs = 7 rad/sec, (c) and (d) ωs = 10
rad/sec.

Table 7
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the two formulations with
different steady rotation speeds.
ωs = 7 rad/sec ωs = 10 rad/sec
RNCF NR-RNCF (1VM) NR-RNCF (1VM+1MD) RNCF NR-RNCF (1VM) NR-RNCF (1VM+1MD)

DOFs 18,912 1 2 18,912 1 2

Adaptive steps 507 609 654 508 664 672
Time (sec) 3308 0.122 0.163 3591 0.118 0.191
χrel (%) — 56.2 0.26 — 64.3 0.43

RNCF, and the global relative error is even up to 43.8%, as shown in Table 8.
As shown in Table 8, the accuracy of the current formulation shows a marked improvement when the modal derivatives are
considered, and the maximum global relative error is only 0.71%. It is worth noting that this example is a typical bending-torsional
coupling problem, and the modal derivative is essential to capture the nonlinear coupling deformation. Therefore, the result confirms
that the NR-RNCF is also applicable to the deformation coupling problem.
At the same time, the computational cost is greatly reduced due to the dramatic reduction of the dimension of equations, which is

18
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 13. (a) The 3rd vibration mode and (b) the 8th vibration mode of the rotating membrane (the blue line denotes the undeformed configuration,
and the red line denotes mode shapes).

Fig. 14. The deflections at the edge of the rotating membrane with the steady rotational speeds ωs = 4 rad/sec and displacement excitation wz.

19
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Table 8
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the two formulations in
the bending-torsional problem.
RNCF NR-RNCF (3VMs) NR-RNCF (3VMs+6MDs)

DOFs 18,912 3 9
Adaptive steps 507 634 662
Time (sec) 3462 0.167 0.384
χrel (%) — 43.8 0.71

about three orders of magnitude faster than the RNCF. The result further indicates that the high efficiency of the present method does
not rely on a particular example, but benefits from the great reduction of the number of DOFs.

4.2. Flexible pendulum

To verify the generality of the current formulation in multibody systems, a flexible pendulum consisting of three beams is
considered in Fig. 15. The adjacent beams are connected by revolute joints, and the first beam is hinged to the ground. The flexible
pendulum is released at the horizontal position and subject to a constant gravitational force with g = 9.81 m/sec2.
Suppose the cross-section of the beam is circular with a radius of 5 mm; the length of the beam is 1.0 m; the material is uniform and
isotropic, with density ρ = 1100 kg/m3, Young’s modulus E = 5 × 106 Pa, and Poisson’s ratio ν = 0.3. In both RNCF and NR-RNCF, the
left end position vector and the local coordinate frame of each beam are selected as the reference frame.
In the RNCF, each beam is divided into 10 C1 beam elements [45] and the number of DOFs in the RNCF model is 558. In the
NR-RNCF, constraint reactions must be taken into account. Fortunately, constraints are usually acting only on some specific boundary
nodes, it is therefore convenient to adopt the extended Craig-Bampton method [28] here to obtain the reduction basis.
As discussed in [28], the elastic coordinates are partitioned into two categories. The DOFs that are related to joints are referred to as
boundary DOFs and denoted as qB , which should be preserved exactly to make connections with other bodies. The other DOFs are
referred to as interior DOFs and denoted as qI , which need to be reduced. Accordingly, Elements of mass matrix M and stiffness matrix
K in Eq. (27) needs to be re-arranged associated with qB and qI , which are
[ ] [ ]
MBB MBI KBB KBI
M= , K= (44)
MIB MII KIB KII

The reduction can be written as

( ) [ ]( )
qB I O uB
q= o
=q + = qo + Ψp (45)
qI ΦIB ΨII pI

in which
[ ] ( )
I O uB
Ψ= , p= (46)
ΦIB ΨII pI

where the constraint mode matrix

ΦIB = − K−II 1 KIB (47)

and the transformation matrix ΨII can be obtained by the procedure described in Section 3, except that the mass matrix M and the
tangent stiffness matrix K need to be replaced by MII and KII, respectively.
Note that uB is the increments of the boundary DOFs around the equilibrium configuration, and pI is the modal coordinate vector
related to the interior DOFs, both of them form the generalized modal coordinate vector p. In this way, the expression of Eq. (45) is
consistent with Eq. (13) and can be directly applied to the NR-RNCF.
In the NR-RNCF, the reduction basis ΨII can be constructed through the procedure in Section 3 for all three beams, which consists of
the first ten VMs and a few MDs selected based on the MMI criterion. The MMI matrices are shown in the bar plot in Fig. 16.
Since Beam-1 and Beam-2 are two-end constrained and Beam-3 is single-end constrained, the vibration modes excited by the three

Fig. 15. Schematic of a flexible pendulum consisting of three beams.

20
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 16. Maximum modal interaction matrix relative to the first ten VMs: (a) Beam-1, (b) Beam-2, and (c) Beam-3 of the flexible pendulum.

beams are different, as shown in Fig. 16. During the swinging process, all the inertia forces and the elastic forces are always in the X-Y
plane, and there are no out-of-plane deflections. Some vibration modes do not participate since they are characterized by out-of-plane
bending, and these VMs and the corresponding MDs are not involved in the reduction basis. Thence, the critical five VMs and the
corresponding fifteen MDs (burgundy) are selected as the modal basis for each beam. Also, considering the rigid coordinates and
boundary DOFs of each beam, the number of DOFs in the NR-RNCF model is 84. The dynamic responses of the flexible pendulum at 0.0
s, 0.5 s, 1.0 s, 1.5 s, and 2.0 s are shown in Fig. 17.

Fig. 17. The dynamic response of the flexible pendulum during the two second.

21
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

As shown in Fig. 17, the dynamic responses calculated from the NR-RNCF are highly consistent with those from the RNCF when the
MDs are used, whereas a stiff response can be observed if only VMs are adopted as the modal basis. The comparison indicates that the
MDs are essential to describe geometric nonlinearities. The displacements and deflections at the tip of each beam are shown in Fig. 18,
where the NR-RNCF considers the MDs and the tip deflections are described in the reference frame for each beam.
Note that the bending stiffness of the beam is only 2.45 × 10− 3 N⋅m2, and the beam is as soft as a cable. The tip displacements of the

Fig. 18. The displacements and deflections at the tip of each beam, (a) and (b) Beam-1, (c) and (d) Beam-2, (e) and (f) Beam-3.

22
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

two methods show almost no deviation in Fig. 18(a), 18(c), and 18(e). From Beam-1 to Beam-3, a certain sequence of the deflection
peaks can be observed in Fig. 18(b), 18(d), and 18(f), and the maximum transverse deflection of each beam exceeds 50% of the length
of the beam, which is very large deformation, and the results from the NR-RNCF are still in good agreement with those from the RNCF.
The number of DOFs, adaptive steps, computational times, and the global relative error between the two methods are further compared
in Table 9.
The results from NR-RNCF are consistent with the RNCF results, and the global relative error of the two methods is about 1% in
Table 9. Although the boundary DOFs, VMs, and corresponding MDs are all considered for each beam, the number of generalized
coordinates of the NR-RNCF is still much smaller than that of the RNCF, and the difference in time steps between the two methods is
also not significant. In the NR-RNCF, the computation time for the simulation is 76.31 s, which is about five times faster than the RNCF.
The result demonstrates that the NR-RNCF can significantly improve computational efficiency and save huge computer resources.
Note that the flexible pendulum is a typical flexible multibody system with large deformation, and the constraints can be easily
considered by the extended Craig-Bampton method [28] in the NR-RNCF, which further guarantees the generality of the current
formulation in flexible multibody systems.

4.3. Nonlinear beamlike truss

Nowadays, truss structures composed of structural elements are widely welcomed in astronautics because of their excellent
stiffness-to-weight and strength-to-weight ratios. As a result, the geometric nonlinearity effects can no longer be neglected for large-
scale truss structures. However, the finite element model of the truss system generally involves high degrees of freedom, which bears a
prohibitive computation cost and cannot be exploited conveniently for the controller design of the space structure. To address this
drawback, many techniques were developed to obtain the reduced-order model of truss structure [47–49]. In this subsection, the
NR-RNCF is adopted to achieve the model order reduction of the space nonlinear beamlike truss (NBT).

4.3.1. Description of the NBT

The triangular prism beamlike truss [49] is considered in this paper, which is composed of longerons, battens, and diagonals, as
shown in Fig. 19(a) and 19(b). The nonlinear beamlike truss is produced by 36 spatial repeating elements along the axial direction, and
the total length of the beamlike truss is 98.64 m.
The local enlarged joint of the adjacent members is shown in Fig. 19(c), the connection of the adjacent longerons and battens is
rigid-jointed, and the diagonal is connected to the truss by pin joints. As shown in Fig. 19(d) and 19(e), the cross-section of the longeron
and batten is a hollow cylinder, and that of the diagonal members is a solid cylinder. The longerons and battens are made of aluminum
alloy, and the diagonals are made of carbon fiber. The longerons and battens are beams, the diagonals are two-force rods, and the
detailed material and dimensional parameters of the members are summarized in Table 10.

4.3.2. Modal analysis of the NBT

To obtain a RNCF model of the beamlike truss, each longeron and batten is discretized by 6 and 4 C1 beam elements [45],
respectively, and a linear rod element describes each diagonal. The total degree of freedom of the RNCF model is 18,360.
To verify the validity of the RNCF model, the natural characteristics of the NBT with cantilevered boundary conditions are given by
Eq. (28) and compared with the results obtained by the business software ABAQUS. Note that the bending modes along the y-axis and
the z-axis are symmetrical, so only the bending modes along the z-axis are shown in this paper. The first five order modes of the NBT are
selected and compared in Table 11, and the relative errors can be defined by
⃒ ⃒
⃒fRNCF − fABAQUS ⃒
Re = × 100%
fABAQUS

where fRNCF and fABAQUS represent the frequencies obtained by the RNCF model and ABAQUS software. The NBT established by
ABAQUS has 1092 beam elements with B32 and 216 link elements with T3D2.
According to Table 11, it can be known from the relative errors that the results of the RNCF model are precise enough and in good
agreement with those of ABAQUS, where the maximum error is no more than 1.93%. That demonstrates the validity of the RNCF
model. Furthermore, the first five mode shapes are depicted in Fig. 20 and the deformation scale factor is 100. It can be seen that the
first five vibration mode shapes are bending/twisting dominant modes and the results obtained by ABAQUS and the RNCF model are
consistent. The validity of the RNCF model is further verified.

4.3.3. Dynamic analysis

In this section, dynamic responses of the space truss structure are investigated to validate the reliability of the NR-RNCF. The large
truss structure features a large size and more flexibility, which induces the weak ability to resist deformation. Therefore, such
structures exhibit large deformations under the influence of external excitation, such as the attitude maneuver or other orbital op­
erations. These scenarios can be modeled as the space truss structure with cantilevered-free boundary conditions [50], as shown in
Fig. 21.
Suppose the cantilevered end of the space truss is subjected to sinusoidal displacement excitations along the Y and Z directions

W = [ 0, 0.05sin(0.25 ⋅ 2π ⋅ t), 0.25sin(0.20 ⋅ 2π ⋅ t) ]T

23
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Table 9
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the two formulations for the
flexible pendulum.
Method DOFs Adaptive steps Time (sec) χrel (%)
RNCF 558 4470 443.41 —
NR-RNCF 84 5913 76.31 1.07

Fig. 19. Schematic view of (a) nonlinear beamlike truss composed of two spatial repeating elements, (b) the cross-section of the NBT, (c) the local
enlarged joint of the adjacent members of the NBT, (d) the cross-section of the longeron and batten of the NBT, and (e) the cross-section of the
diagonal of the NBT.

Table 10
Material and dimensional parameters of the nonlinear beamlike truss.
L / mm D / mm h / mm Density (kg / m3) Young’s Modulus (GPa) Poisson’s ratio

Longeron 2740 100 1 2700 69 0.33

Batten 3200 50 1 3960 370 0.22
Diagonal 4212.8 6 — 1600 88 —

In the NR-RNCF, the reduced order basis is constructed as described in Section 3 for the NBT with the cantilevered end. As shown in
Fig. 22, the cross terms relating the first two VMs to all the other VMs are more critical than others. Mode 3 does not participate since
this mode features torsional motion along the X-axis, as illustrated in Fig. 20(b). As the modal derivative with the maximum weight, the

24
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Table 11
Comparisons of natural frequencies between the RNCF model and the ABAQUS model.
Mode Order ABAQUS (Hz) RNCF (Hz) Re (%)
b
1 0.2642 0.2693 b 1.9304
t
2 1.5156 1.5209 t 0.3297
b
3 1.5696 1.5973 b 1.7648
b
4 4.0743 4.1386 b 1.5782
t
5 4.5413 4.5575 t 0.3567
b
Bending mode
t
Torsional mode

Fig. 20. The (a) first, (b) third, (c) fourth, (d) sixth, and (e) eighth mode shape of the NBT with cantilevered boundary conditions (the blue line
denotes the undeformed configuration, and the red line denotes mode shapes) and the deformation scale factor is 100; in each subfigure, the top one
is from ABAQUS, and the bottom one is from the RNCF.

25
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 21. The full expended space truss (36 units) with cantilevered-free boundary conditions.

Fig. 22. Maximum modal interaction matrix relative to the first five VMs of the NBT with the cantilevered end.

Fig. 23. The mode shape of the MDs with the highest weight, i.e., θ22 (the blue line denotes the undeformed configuration, and the red line denotes
mode shapes).

mode shape of θ22 is shown in Fig. 23.

The MD θ22 is a typical bending-stretching mode, while the corresponding VMs are bending-dominated modes. Therefore, the
reduced order basis in the NR-RNCF is formed by the first four bending modes and the first six MDs with the highest weights (bur­
gundy), which yields an excellent approximation to the RNCF model, as shown in Fig. 24.
As is evident, the results obtained by the present NR-RNCF agree excellently with those computed by the RNCF when the MDs are
taken into consideration, which can further verify the proposed NR-RNCF is effective and reliable. Moreover, the frequencies of
external excitation are close to the first natural frequency ω1 = 0.2693 Hz, as illustrated in Table 11, and thus the vibration amplitude
of the NBT is relatively large because of resonance behavior. Also, dynamic responses with the basis formed by only the first four
bending modes are shown. The result is overly stiff and clearly wrong. The comprehensive comparison between the two methods is
shown in Table 12.
As clarified in Table 12, the accuracy is significantly improved by adding a few MDs. The results indicate that the MDs are essential
to describe the nonlinear bending-stretching coupling effect of the NBT, and it further confirms that the proposed approach could
achieve high accuracy with fewer degrees of freedom than the RNCF. The results demonstrate that the NR-RNCF has a great advantage
over the RNCF in terms of efficiency and can achieve a higher acceleration in large-scale problems. Consequently, it is important to

26
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Fig. 24. Responses of the midpoint at the free end under the external excitation, (a) the deflection along the y-axis, and (b) the deflection along the
z-axis.

Table 12
Comparison of the number of degrees of freedom, adaptive steps, computational times, and the global relative error of the two formulations under the
influence of external excitation.
Formulation DOFs Adaptive steps Time (sec) χrel (%)
RNCF 18,360 2007 4886 —
NR-RNCF (4VM) 4 2643 0.193 32.7
NR-RNCF (4VM+6MD) 10 3254 0.301 0.36

study the active vibration control and the real-time simulation of the space truss based on the NR-RNCF in further work.

5. Comments and conclusions

The generalized coordinates with respect to the rigid motions and those with respect to large deformations are separated in the
RNCF. The generalized coordinates related to large deformations in the RNCF can be significantly reduced by a linear combination of
linear vibration modes and mode derivatives, such that the geometrical nonlinearities are preserved. The derived formulation, named
NR-RNCF, has the following characteristics:

1. The Green-Cauchy strain tensor is adopted to describe geometrically nonlinear deformations, and the yielding elastic modal forces
can be expressed as cubic polynomials of modal coordinates, which is highly efficient in computations;
2. Similar to the RNCF, the time step of the current formulation is much large than ANCF in simulations of fast-rotating systems.
Moreover, the proposed approach has a great advantage over the RNCF in terms of efficiency owing to the reduction of the
dimension of equations.
3. To describe large deformation and high-speed rotations of the flexible body, modal derivatives are adopted to enrich the linear
reduction basis in the NR-RNCF. As a simulation-free approach, modal derivatives have been proven as a simple but effective
technique to capture nonlinear responses;
4. It is not accurate enough to describe large deformations by only linear vibration modes, and the accuracy can be significantly
improved with the addition of a few modal derivatives;
5. The formulation is similar to the floating frame of reference method and can be directly implemented in the classical framework of
multibody system dynamics;
6. Much fewer number of generalized coordinates are required to achieve a high accuracy, and this formulation can be directly
adopted in controller designs.

27
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

No data was used for the research described in the article.

Acknowledgement

This work is supported by the National Natural Science Foundation of China [grant No. 11772101], the National Key Research and
Development Program of China [grant No. 2018YFF0300506], the Fundamental Research Funds for the Central Universities [grant No.
FRFCU5710052721], and the Heilongjiang Touyan Innovation Team Program.

Appendix A. Derivation of the nonlinear elastic force in Eq. (24)

Substituting Eq. (14) into Eq. (10) yields

1( )
Qμν = q ⋅q − ̂ qμ ⋅ ̂
qν
2 μ ν
1 [( o ) ( ) ]
= qμ + ψμ p ⋅ qoν + ψν p − q ̂μ ⋅ q
̂ν
2
(A-1)
1 [( o o ) ( ) ]
= ̂μ ⋅ ̂
qμ ⋅ qν − q q ν + ψiμ ⋅ qoν + ψiν ⋅ qoμ pi + ψiμ ⋅ ψjν pi pj
2
1 ij
= Qoμν + Xμν
i
pi + Yμν pi pj
2

in which
1( o o ) 1( i ) 1( i )
Qoμν = ̂μ ⋅ ̂
q ⋅q − q i
q ν , Xμν = ψμ ⋅ qoν + ψiν ⋅ qoμ , Yμν
ij
= ψμ ⋅ ψjν + ψiν ⋅ ψjμ
2 μ ν 2 2
Note that Xμν
i
and Yμν
ij
are satisfied with the following property
i
Xμν i
= Xνμ ij
, Yμν ij
= Yνμ ji
= Yμν ji
= Yνμ (A-2)

Substituting Eq. (9) into Eq. (23) yields

Fei = ψiα ⋅ Fαe = ψiα ⋅ καβμν Qμν qβ (A-3)

Substituting Eqs. (A-1) and (14) into Eq. (A-3) yields

Fei = ψiα ⋅ καβμν Qμν qβ

( )( )
1
= ψiα ⋅ καβμν Qoμν + Xμν
j jk
pj + Yμν pj pk qoβ + ψlβ pl
2
( )( )
1 jk
= καβμν Qoμν + Xμν j
pj + Yμν pj pk Xαi β + Yαilβ pl
2
( ) ( )( )
1 jk
= Θαβ Xαi β + Yαilβ pl + καβμν Xμν j
pj + Yμν pj pk Xαi β + Yαilβ pl
2
(A-4)
= Θαβ Xαi β + Θαβ Yαilβ pl + καβμν Xαi β Xμν
j
pj + καβμν Yαilβ Xμν
j
pj pl
1 1
+ καβμν Yμν Xαβ pj pk + καβμν Yαilβ Yμν
jk i jk
pj pk pl
2 2
( )
= Θαβ Xαi β + Θαβ Yαijβ + καβμν Xαi β Xμν
j
pj

1 ( ) 1
+ καβμν Yαijβ Xμν
k
+ Yαkiβ Xμν
j
+ Yαjkβ Xμν
i
pj pk + καβμν Yαijβ Yμν
kl
pj pk pl
2 2
= f i + K ij pj + Lijk pj pk + N ijkl pj pk pl

where the constant coefficient f i , Kij , Lijk , and Nijkl are expressed by

28
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

Θαβ = καβμν Qoμν , f i = Θαβ Xαi β , K ij = Θαβ Yαijβ + καβμν Xαi β Xμν
j

1 ( ) 1 ( ) (A-5)
Lijk = καβμν Yαijβ Xμν
k
+ Yαjkβ Xμν
i
+ Yαkiβ Xμν
j
, N ijkl = καβμν Yαijβ Yμν
kl
+ Yαikβ Yμν
jl
2 4

Appendix B. Derivation of the derivative of the stiffness matrix in Eq. (35)

According to Eqs. (11) and (27), the tangent stiffness matrices can be written as
⃒ ∂Fα ⃒ ( )⃒
K⃒eq = e ⃒q=qo = καβμν Qμν I3 + καμβν qμ qTν ⃒q=qo (B-1)
∂qβ
Assumed a displacement given in the direction of the linear vibration mode ϕj, the elastic coordinates can be written as

qx = qox + ϕjx ηj (B-2)

Substituting Eq. (B-2) into Eq. (10) yields

1( )
Qμν = q ⋅q − ̂ qμ ⋅ ̂
qν
2 μ ν
1 [( o ) ( ) ]
= qμ + ϕjμ ηj ⋅ qoν + ϕjν ηj − ̂
qμ ⋅ ̂
qν (B-3)
2
1 [( o o ) ( ) ( ) ]
= qμ ⋅ qν − q q ν + qoμ ⋅ ϕjν + ϕjμ ⋅ qoν ηj + ϕjμ ⋅ ϕjν η2j
̂μ ⋅ ̂
2
Similarly
( )( )T ( )
(B-4)
T T T
qμ qTv = qoμ + ϕjμ ηj qoν + ϕjν ηj = qoμ qoν + qoμ ϕjν + ϕjμ qoν T ηj + ϕjμ ϕjν η2j

Substituting Eqs. (B-3) and (B-4) into Eq. (B-1) yields

( )
K qo + ϕj ηj = καβμν Qμν I3 + καμβν qμ qTν
( [( ) ( ) ( ) ])
1
= καβμν qoμ ⋅ qoν − q q ν + qoμ ⋅ ϕjν + ϕjμ ⋅ qoν ηj + ϕjμ ⋅ ϕjν η2j I3
̂μ ⋅ ̂ (B-5)
2
[ ( ) ]
T T
+κ αμβν
qoμ qoν T + qoμ ϕjν + ϕjμ qoν T ηj + ϕjμ ϕjν η2j

Conducted the partial derivatives of Eq. (B-5) with respect to ηj, yields
( ) [ ( ) ( ) ] [( ) ]
∂K qo + ϕj ηj 1 o j
(B-6)
T T
= καβμν qμ ⋅ ϕν + ϕjμ ⋅ qoν + ϕjμ ⋅ ϕjν ηj I3 + καμβν qoμ ϕjν + ϕjμ qoν T + 2ϕjμ ϕjν ηj
∂ηj 2

According to Eq. (34), we can get

( ) { [ ( ) ( ) ] }⃒
∂K qo + ϕj ηj ⃒⃒ 1 o j
ηj =0 = καβμν qμ ⋅ ϕν + ϕjμ ⋅ qoν + ϕjμ ⋅ ϕjν ηj I3 ⃒ηj =0
∂ηj 2
{ [( ) ]}⃒
T T
+ καμβν qoμ ϕjν + ϕjμ qoν T + 2ϕjμ ϕjν ηj ⃒ηj =0 (B-7)
[ ( )] [( )]
1 o j T
= καβμν qμ ⋅ ϕν + ϕjμ ⋅ qoν I3 + καμβν qoμ ϕjν + ϕjμ qoν T
2

so
[ ]
∂K ⃒⃒ 1( o j ) [( )]
(B-8)
T
= καβμν q ⋅ ϕ + ϕjμ ⋅ qoν I3 + καμβν qoμ ϕjν + ϕjμ qoν T
∂ηj eq 2 μ ν

Note that the tangent stiffness matrix derivative’s expression is consistent and general for the beam element and the plate/shell
element in the procedure of computing the MDs. Moreover, the detailed expression of the constant four-order tensor καβμν can be found
in Refs. [45] and [34].

29
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

References

[1] A.A. Shabana, Definition of ANCF finite elements, J. Comput. Nonlinear Dyn. 10 (5) (2015), 054506, https://doi.org/10.1115/1.4030369.
[2] P. Lan, Q.L. Tian, Z.Q. Yu, A new absolute nodal coordinate formulation beam element with multilayer circular cross section, Acta Mech. Sin. 36 (1) (2020)
82–96, https://doi.org/10.1007/s10409-019-00897-4.
[3] Q. Tian, P. Zhang, K. Luo, Dynamics of soft mechanical systems actuated by dielectric elastomers, Mech. Syst. Signal Proc. 151 (2021), 107392, https://doi.org/
10.1016/j.ymssp.2020.107392.
[4] W. Fan, S.H. Zhang, W.D. Zhu, H. Zhu, An efficient dynamic formulation for the vibration analysis of a multi-span power transmission line excited by a moving
deicing robot, Appl. Math. Model. 103 (2022) 619–635, https://doi.org/10.1016/j.apm.2021.10.040.
[5] M. Pieber, K. Ntarladima, R. Winkler, J. Gerstmayr, A hybrid ALE formulation for the investigation of the stability of pipes conveying fluid and axially moving
beams, J. Comput. Nonlinear Dyn. 17 (5) (2022), 051006, https://doi.org/10.1115/1.4053505.
[6] V. Sonneville, M. Scapolan, M. Shan, O.A. Bauchau, Modal reduction procedures for flexible multibody dynamics, Multibody Syst. Dyn. 51 (4) (2021) 377–418,
https://doi.org/10.1007/s11044-020-09770-w.
[7] M. Géradin, D.J. Rixen, A fresh look at the dynamics of a flexible body application to substructuring for flexible multibody dynamics, Int. J. Numer. Methods
Eng. 122 (14) (2021) 3525–3582, https://doi.org/10.1002/nme.6673.
[8] M. Ellenbroek, J. Schilder, On the use of absolute interface coordinates in the floating frame of reference formulation for flexible multibody dynamics,
Multibody Syst. Dyn. 43 (3) (2018) 193–208, https://doi.org/10.1007/s11044-017-9606-3.
[9] J. Schilder, K. Dwarshuis, M. Ellenbroek, A. de Boer, The tangent stiffness matrix for an absolute interface coordinates floating frame of reference formulation,
Multibody Syst. Dyn. 47 (3) (2019) 243–263, https://doi.org/10.1007/s11044-019-09689-x.
[10] A. Zwölfer, J. Gerstmayr, The nodal-based floating frame of reference formulation with modal reduction, Acta Mech. 232 (3) (2021) 835–851, https://doi.org/
10.1007/s00707-020-02886-2.
[11] R. Vidoni, A. Gasparetto, M. Giovagnoni, Design and implementation of an ERLS-based 3-D dynamic formulation for flexible-link robots, Robot. Comput. Integr.
Manuf. 29 (2013) 273–282, https://doi.org/10.1016/j.rcim.2012.07.008.
[12] R. Vidoni, P. Gallina, P. Boscariol, A. Gasparetto, M. Giovagnoni, Modeling the vibration of spatial flexible mechanisms through an equivalent rigid-link system/
component mode synthesis approach, J. Vib. Control. 23 (12) (2017) 1890–1907, https://doi.org/10.1177/1077546315604495.
[13] N. Kobayashi, T. Wago, Y. Sugawara, Reduction of system matrices of planar beam in ANCF by component mode synthesis method, Multibody Syst. Dyn. 26 (3)
(2011) 265–281, https://doi.org/10.1007/s11044-011-9259-6.
[14] K. Otsuka, K. Makihara, Deployment simulation using absolute nodal coordinate plate element for next-generation aerospace structures, AIAA J. 56 (3) (2018)
1266–1276, https://doi.org/10.2514/1.J056477.
[15] Y.X. Tang, H.Y. Hu, Q. Tian, Model order reduction based on successively local linearizations for flexible multibody dynamics, Int. J. Numer. Methods Eng. 118
(3) (2019) 159–180, https://doi.org/10.1002/nme.6011.
[16] Q.L. Tian, P. Lan, Z.Q. Yu, Model-order reduction of flexible multibody dynamics via free-interface component mode synthesis method, J. Comput. Nonlinear
Dyn. 15 (10) (2020), 101008, https://doi.org/10.1115/1.4047868.
[17] E. Kim, H. Kim, M. Cho, Model order reduction of multibody system dynamics based on stiffness evaluation in the absolute nodal coordinate formulation,
Nonlinear Dyn 87 (3) (2017) 1901–1915, https://doi.org/10.1007/s11071-016-3161-y.
[18] E. Kim, M. Cho, Design of a planar multibody dynamic system with ANCF beam elements based on an element-wise stiffness evaluation procedure, Struct.
Multidiscip. Optim. 58 (3) (2018) 1095–1107, https://doi.org/10.1007/s00158-018-1954-y.
[19] K. Luo, H.Y. Hu, C. Liu, Q. Tian, Model order reduction for dynamic simulation of a flexible multibody system via absolute nodal coordinate formulation,
Comput. Methods Appl. Mech. Eng. 324 (2017) 573–594, https://doi.org/10.1016/j.cma. 2017.06.029.
[20] Y.S. Hou, C. Liu, H.Y. Hu, Component-level proper orthogonal decomposition for flexible multibody systems, Comput. Methods Appl. Mech. Eng. 361 (2020),
112690, https://doi.org/10.1016/j.cma.2019.112690.
[21] K. Otsuka, K. Makihara, H. Sugiyama, Recent advances in the absolute nodal coordinate formulation: literature review from 2012 to 2020, J. Comput. Nonlinear
Dyn. 17 (8) (2022), 080803, https://doi.org/10.1115/1.4054113.
[22] H. Ren, K. Yang, A referenced nodal coordinate formulation, Multibody Syst. Dyn. 51 (3) (2020) 305–342, https://doi.org/10.1007/s11044-020-09750-0.
[23] H. Ren, T.F. Yuan, M.Y. Huo, C. Zhao, S.J. Zeng, Dynamics and control of a full-scale flexible electric solar wind sail spacecraft, Aerosp. Sci. Technol. 119 (2021),
107087, https://doi.org/10.1016/j.ast.2021.107087.
[24] C. Touze, A. Vizzaccaro, O. Thomas, Model order reduction methods for geometrically nonlinear structures: a review of nonlinear techniques, Nonlinear Dyn
105 (2) (2021) 1141–1190, https://doi.org/10.1007/s11071-021-06693-9.
[25] S.R. Idelsohn, A. Cardona, A reduction method for nonlinear structural dynamic analysis, Comput. Methods Appl. Mech. Eng. 49 (3) (1985) 253–279, https://
doi.org/10.1016/0045-7825(85)90125-2.
[26] S.R. Idelsohn, A. Cardona, A load-dependent basis for reduced nonlinear structural dynamics, Comput. Struct. 20 (1985) 203–210, https://doi.org/10.1016/
0045-7949(85)90069-0.
[27] O. Weeger, U. Wever, B. Simeon, On the Use of Modal Derivatives for Nonlinear Model Order Reduction, Int. J. Numer. Methods Eng. 108 (13) (2016)
1579–1602, https://doi.org/10.1002/nme.5267.
[28] L. Wu, P. Tiso, Nonlinear model order reduction for flexible multibody dynamics: a modal derivatives approach, Multibody Syst. Dyn. 36 (4) (2016) 405–425,
https://doi.org/10.1007/s11044-015-9476-5.
[29] L. Wu, P. Tiso, K. Tatsis, E. Chatzi, F. van Keulen, A modal derivatives enhanced Rubin substructuring method for geometrically nonlinear multibody systems,
Multibody Syst. Dyn. 45 (1) (2019) 57–85, https://doi.org/10.1007/s11044-018-09644-2.
[30] S. Jain, P. Tiso, Model order reduction for temperature-dependent nonlinear mechanical systems: a multiple scales approach, J. Sound Vib. 465 (2020), 115022,
https://doi.org/10.1016/j.jsv.2019.115022.
[31] E. Delhez, F. Nyssen, J.C. Golinval, A. Batailly, Reduced order modeling of blades with geometric nonlinearities and contact interactions, J Sound Vib 500
(2021), 116037, https://doi.org/10.1016/j.jsv.2021.116037.
[32] J. Marconi, P. Tiso, F. Braghin, A nonlinear reduced order model with parametrized shape defects, Comput. Methods Appl. Mech. Eng. 360 (2020), 112785,
https://doi.org/10.1016/j.cma.2019.112785.
[33] J. Marconi, P. Tiso, D.E. Quadrelli, F. Braghin, A higher order parametric nonlinear reduced order model for imperfect structures using Neumann expansion,
Nonlinear Dyn 104 (4) (2021) 3039–3063, https://doi.org/10.1007/s11071-021-06496-y.
[34] H. Ren, Fast and robust full-quadrature triangular elements for thin plates/shells with large deformations and large rotations, J. Comput. Nonlinear Dyn. 10 (5)
(2015), 051018, https://doi.org/10.1115/1.4030212.
[35] A. Cammarata, Global modes for the reduction of flexible multibody systems: methodology and complexity, Multibody Syst. Dyn. 53 (1) (2021) 59–83, https://
doi.org/10.1007/s11044-021-09790-0.
[36] A. Cardona, M. Geradin, Time integration of the equations of motion in mechanism analysis, Comput. Struct. 33 (3) (1989) 801–820, https://doi.org/10.1016/
0045-7949(89)90255-1.
[37] J.B. Rutzmoser, D.J. Rixen, P. Tiso, S. Jain, Generalization of quadratic manifolds for reduced order modeling of nonlinear structural dynamics, Comput. Struct.
192 (2017) 196–209, https://doi.org/10.1016/j.compstruc.2017.06.003.
[38] Tiso, P.: Optimal second order reduction basis selection for nonlinear transient analysis. In: Proulx, T. (ed.) Modal Analysis Topics, vol. 3, 27–39. Springer, New
York (2011). https://doi.org/10.1007/978-1-4419-9299-4_3.
[39] S. Jain, P. Tiso, J.B. Rutzmoser, D.J. Rixen, A quadratic manifold for model order reduction of nonlinear structural dynamics, Comput. Struct. 188 (2017) 80–94,
https://doi.org/10.1016/j.compstruc.2017.04.005.

30
T. Yuan et al. Mechanism and Machine Theory 185 (2023) 105290

[40] M. Arnold, O. Brüls, Convergence of the generalized-alpha scheme for constrained mechanical systems, Multibody Syst. Dyn. 18 (2) (2007) 185–202, https://
doi.org/10.1007/s11044-007-9084-0.
[41] O. Brüls, A. Cardona, M. Arnold, Lie group generalized-alpha time integration of constrained flexible multibody systems, Mech. Mach. Theory. 48 (1) (2012)
121–137, https://doi.org/10.1016/j.mechmachtheory.2011.07.017.
[42] J.C. Simo, L. Vu-Quoc, On the dynamics of flexible beams under large overall motions - the plane case: part II, J. Appl. Mech. 53 (4) (1986) 855–863, https://
doi.org/10.1115/1.3171871.
[43] K.M. Hsiao, R.T. Yang, A.C. Lee, A consistent finite element formulation for nonlinear dynamic analysis of planar beam, Int. J. Numer. Methods Eng. 37 (1)
(1994) 75–89, https://doi.org/10.1002/nme.1620370106.
[44] T. Kim, A.M. Hansen, K. Branner, Development of an anisotropic beam finite element for composite wind turbine blades in multibody system, Renew. Energy. 59
(2013) 172–183, https://doi.org/10.1016/j.renene.2013.03.033.
[45] H. Ren, A simple absolute nodal coordinate formulation for thin beams with large deformations and large rotations, J. Comput. Nonlinear Dyn. 10 (6) (2015),
061005, https://doi.org/10.1115/1.4028610.
[46] Q. Li, T.S. Wang, X.R. Ma, Geometric nonlinear effects on the planar dynamics of a pivoted flexible beam encountering a point-surface impact, Multibody Syst.
Dyn. 21 (3) (2009) 249–260, https://doi.org/10.1007/s11044-008-9138-y.
[47] G. Piccardo, F. Tubino, A. Luongo, Equivalent Timoshenko linear beam model for the static and dynamic analysis of tower buildings, Appl. Math. Model. 71
(2019) 77–95, https://doi.org/10.1016/j.apm.2019.02.005.
[48] M. Liu, D.Q. Cao, D.F. Zhu, Equivalent dynamic model of the space antenna truss with initial stress, AIAA J. 58 (4) (2020) 1851–1863, https://doi.org/10.2514/
1.J058647.
[49] M. Liu, D.Q. Cao, X.Y. Zhang, J. Wei, D.F. Zhu, Nonlinear dynamic responses of beamlike truss based on the equivalent nonlinear beam model, Int. J. Mech. Sci.
194 (2021), 106197, https://doi.org/10.1016/j.ijmecsci.2020.106197.
[50] M. Liu, D.Q. Cao, J.P. Li, X.Y. Zhang, J. Wei, Dynamic modeling and vibration control of a large flexible space truss, Meccanica 57 (2022) 1017–1033, https://
doi.org/10.1007/s11012-022-01487-8.

31