Multibody Syst Dyn (2021) 51:305–342

https://doi.org/10.1007/s11044-020-09750-0

A referenced nodal coordinate formulation

Hui Ren1 · Kai Yang2

Received: 14 January 2020 / Accepted: 7 June 2020 / Published online: 22 June 2020
© Springer Nature B.V. 2020

Abstract We develop a referenced nodal coordinate formulation (RNCF) to study the dy-
namics of flexible bodies undergoing large-distance travels and/or high-speed rotations.
RNCF is similar to the absolute nodal coordinate formulation (ANCF) but is presented in
a noninertia reference coordinate system (RCS). The position vector and rotation matrix of
the RCS describe translational and rotational motions of the system, whereas the nodal co-
ordinates and slopes in a structure depict its large deformations, such that the generalized
coordinates with multiple scales in length and time are automatically separated. We develop
a parameter-irrelevant technique to derive the rotation equations of the system, where the
influences of large deformations on the rotatory inertia tensors are embodied. The derived
governing equations are simple and elegant, and consistent with the governing equations
for rigid bodies, the floating frame of reference method, as well as ANCF. We verify the
RNCF approach by three typical examples, including the spin-up maneuver, the high speed
motor, and the flexible slider-crank mechanism. The results indicate that to achieve the same
accuracy, the computational cost for RNCF is much lower than that for the corresponding
ANCF in high-speed rotating systems. Moreover, the electrical solar wind sail spacecraft
system is formulated by RNCF, and its propulsive efficiencies with respect to the spin rates
of the E-sails are studied by full-scale models with over ten thousand degrees of freedom.
RNCF provides an effective way to formulate and study the dynamics of vehicles, trains,
ships, aircrafts, and spacecrafts.

Keywords Referenced nodal coordinate formulation · Absolute nodal coordinate

formulation · Multiscale problems · Gyroscopic systems · Electrical solar wind sail

B H. Ren
[email protected]
K. Yang
[email protected]
1 Institute of Aerospace Vehicle Dynamics and Control, School of Astronautics, Harbin Institute of
Technology, Harbin, 150001, P.R. China
2 Graduate Research Assistant, School of Astronautics, Harbin Institute of Technology, Harbin,
150001, P.R. China
306 H. Ren, K. Yang

1 Introduction
Large-distance travels and high-speed rotations are common in the dynamics of vehicles,
trains, ships, aircrafts, and spacecrafts. The lengths of those mechanical systems are in
meters, but their travel distances can be thousands of kilometers and higher. The rotating
parts of those systems can endure angular velocities of hundreds of thousands of rounds per
minute (rpm), whereas their operation durations can be tens of hours, days, months, or even
years. Formulations and calculations of the dynamics of the flexible parts or deformable
bodies in those systems are challenging from the viewpoint of multibody system dynamics.
Both the geometrically exact formulation (GEF) and the absolute nodal coordinate for-
mulation (ANCF) have been successfully implemented to describe deformable bodies in
multibody system dynamics. The GEF [1, 2] is based on independent interpolations on nodal
displacements and rotation coordinates, whereas the ANCF [3] emphasizes on interpolations
on nodal displacements and slopes; both formulations can be applied to deformable bodies
such as beams and plates. The results calculated from the GEF are generally considered to
be more accurate than those from the ANCF [4], providing the same number of generalized
coordinates or even the same number of elements are adopted; but the constant matrices
and efficient calculations of elastic forces [5, 6] in ANCF enable it to be solved very fast
in practice. However, the position vectors in both approaches are described in the absolute
coordinate system, which may cause significant round off errors in large-distance travels,
especially when the strains of the deformable body are small, and different bodies are con-
nected through joints. Moreover, in high-speed rotating scenarios, the absolute coordinate
descriptions can be inefficient in computations, as discussed in Sect. 4.1. A development of
the GEF or ANCF in noninertia coordinate systems could be very useful in practice.
Corotational formulation is another option for deformable bodies with large-distance mo-
tions and high-speed rotations. In corotational formulations for beams and other structures
[7–9], the rotational motion of each element is taken as a reference coordinate system, and
the interpolations of rotational parameters are important but complicated in the implemen-
tations [10]. As a result, the nonlinear computations with respect to rotation coordinates
render the formulations to be complicated. Moreover, the corotational formulations are usu-
ally suitable for numerical simulations, but not for theoretical analyses.
In the study of flexible bodies with small deformations, the floating frame of reference
method (FFRM) has been popular accepted [11]. In vehicle dynamics, the trajectory coordi-
nate system [12] serves as the floating frame of reference, where the moving constraints can
be implemented [13]. The consistent rotation-based formulation [14] is another version of
the FFRM, where the rotation matrix is replaced by the ANCF slopes. Moreover, a similar
approach can be adopted to describe large deformations in rail dynamics [15].
In this work, we propose a referenced nodal coordinate formulation (RNCF) to describe
the dynamics of mechanical systems with large-distance travels and/or high-speed rotations,
as a generalization of FFRM. In Sect. 2, we derive the governing equations of the deformable
body in the reference coordinate system. The governing equations of a rigid body, an FFRM
flexile body, and an ANCF flexible body, are derived from the RNCF equations in Sect. 3,
which are slightly different from the classical equations [3, 11]. In Sect. 4, we apply the
RNCF approach to benchmark examples and the electric solar wind sail spacecraft to show
its correctness and effectiveness. Further comments on the RNCF approach are provided in
Sect. 5. Unlike the ANCF and the GEF approaches, the deformations are described in a coro-
tational reference coordinate system; and unlike the corotational approach, all elements are
described in the same reference coordinate system. Separations of different length and time
scales and significant computational improvement can be achieved in RNCF. Moreover, the
derived governing equations (11)–(13) are simple enough to perform theoretical analyses.
A referenced nodal coordinate formulation 307

2 Theory

The object of this work is to develop a formulation to describe a deformable body under-
going arbitrarily large-distance travels and/or arbitrarily large rotational motions. Consider
the deformable body in a noninertial frame with position vector r(t) and rotational matrix
A(t). The noninertial frame is referred to as the reference coordinate system (RCS). In the
RCS, the position vector of a material particle is p (x, t), where x labels the material co-
ordinates of the particle, and the corresponding position vector in the absolute coordinate
system is
r (x, t) = r(t) + A(t)p (x, t) . (1)
In the RCS, the deformable body is spatially discretized by the nodal position vectors and
slopes as that in ANCF:

p (x, t) = s α (x)q α (t), α = 1, 2, . . . , A, (2)

where the Einstein summation is adopted, and 3A is the number of degrees of freedom of the
deformable body in the RCS. Usually, spatial discretization is performed through the finite
element method, but global trial functions also work in some scenarios. It should be pointed
out that Eq. (2) does NOT stand for the spatial discretization in one finite element, but
provides the spatial discretization for the whole deformable body. Moreover, the reference
coordinate system are usually chosen to be related to the tractive joint of the deformable
body, as will be shown in Sect. 4.

2.1 Kinetic energy

The velocity of the material particle labeled by x can be calculated from Eqs. (1) and (2):

ṙ (x, t) = ṙ(t) + s α (x)A(t)v α (t),

where the angular velocity in the RCS is denoted by ω(t), that is, 
ω = AT Ȧ, and

v α (t) = q̇ α (t) + ω(t) × q α (t). (3)

The kinetic energy of the deformable body is

˚
1 1 1
K = ρ(x)ṙ(x, t) · ṙ(x, t)dv = mṙ · ṙ + d α ṙ · (Av α ) + M αβ v α · v β (4)
2 2 2
V

where
˚ ˚ ˚
m= ρ (x) dv, d =
α α
ρ (x) s (x) dv, M αβ
= ρ (x) s α (x) s β (x) dv,
V V V

in which M αβ stands for the constant mass matrix in ANCF. Direct calculations with respect
to Eq. (4) yield that

d ∂K ∂K d ∂K ∂K
− T = mr̈ + d α Aa α , − T = d α AT r̈ + M αβ a β ,
dt ∂ ṙT ∂r dt ∂ q̇ Tα ∂q α
308 H. Ren, K. Yang

where the acceleration in the RCS is

 
a α = q̈ α + 2ω × q̇ α + ω̇ × q α + ω × ω × q α . (5)

Suppose A(t) is described by a three-component parameter ϑ such that the angular velocity
can be written [11] as ω = D (ϑ) ϑ̇ , and direct calculation yields that

d ∂K ∂K    
T
− T
= D T d α q α × AT r̈ + M αβ q α × a β . (6)
dt ∂ ϑ̇ ∂ϑ

The derivation of Eq. (6) is provided in Appendix A.

2.2 Elastic Potential energy

Suppose the reference configuration of the body is r̊ (x), and denote that
   
∂p ∂p ∂p ∂ r̊ ∂ r̊ ∂ r̊
G(x, t) = , , , G̊(x) = , , .
∂x ∂y ∂z ∂x ∂y ∂z

By Eq. (1), the gradients of the position vectors with respect to x are

∇r(x, t) = A(t)G(x, t), ∇ r̊(x) = G̊(x),

and the deformation gradient can be evaluated by

 −1 −1
F (x, t) = (∇r) ∇ r̊ = A(t)G(x, t)G̊ (x).

Then the Cauchy–Green strain tensor can be written as

1 T  1 −T T −1
ε(x, t) = F · F − I = G̊ (x) GT (x, t) · G(x, t) − G̊ (x) · G̊(x) G̊ (x).
2 2
Moreover, by Eq. (2), the spatial discretization of the strain tensor can be written as

ε (x, t) = S αβ (x) Qαβ (t),

where
 T  
−T ∂s α (x) ∂s β (x) −1
S αβ (x) = G̊ (x) G̊ (x) ,
∂x ∂x
1 
Qαβ (t) = q (t) · q β (t) − q̊ α · q̊ β .
2 α

1. For St. Venant–Kirchhoff elastic materials, suppose that the stiffness tensor is C (x), and
the elastic potential energy can be calculated by
˚
1 1
U = ε (x, t) : C (x) : ε (x, t) dv = k αβμν Qαβ (t)Qμν (t),
2 2
V
A referenced nodal coordinate formulation 309

where V labels the volume of the deformable body in the reference configuration, and
˚
k αβμν = S αβ (x) : C (x) : S μν (x) dv.
V

To damp out high-frequency numerical oscillations, the variation of the elastic potential
energy can be written as
 
δ U = kαβμν Qμν + cQ̇μν δQαβ ,

where c is a visco-elastic damping coefficient. Hence the generalized elastic force is

 
F αe = k αβμν Qμν + cQ̇μν q β . (7)

2. For generic elastic materials, suppose that the elastic potential energy density is W (E),
and the second Piola–Kirchhoff stress tensor is
∂W
σ (x, t) = .
∂ε
To damp out high-frequency oscillations, the generalized elastic force can be written as
˚
F αe = [σ (x, t) + cσ̇ (x, t)] : S αβ (x) dvq β , (8)
V

where the (i, j ) component of the corresponding stress rate tensor is

∂ 2W
σ̇ ij (x, t) = ε̇kl .
∂εij ∂εkl
As a result, the variation of the elastic potential energy can be written as

δ U = δq α · F αe , (9)

where F αe can be taken as either that in Eq. (7) or that in Eq. (8). In practice, to extract
the sparsity in simulations, the flexible body is partitioned into finite elements, and the
calculations above are performed in each element. Then F αe is assembled from the corre-
sponding element elastic forces. The expressions for F αe are usually the same as those in
ANCF.

2.3 Virtual work

Suppose the applied force on the deformable body is f (x, t), which is provided in the
RCS, and the corresponding expression in the absolute coordinate system can be written as
A (t) f (x, t). The virtual work done by the applied force is
˚
   
δW = δr (x, t) · A (t) f (x, t) dv = δr · F + δq α · Qα + δπ · q α × Qα , (10)
V

where δπ is an infinitesimal change of orientation in the RCS, that is, δπ = AT δA, and
˚ ˚
F (t) = A (t) f (x, t) dv, Qα (t) = s α (x) f (x, t) dv.
V V
310 H. Ren, K. Yang

2.4 Governing equations

Suppose the body is installed to a multibody system through constraint equations

C(r, A, q α , qo , t) = 0,

where qo are the generalized coordinates of the other parts in the multibody system.
The discretized governing equations can be derived from the Lagrange equations and
their equivalences in Eq. (34) in Appendix A. The translation equations are
 T
∂C
mr̈ + d α Aa α + λ = F, (11)
∂r

where a α is the same as that in Eq. (5); the rotation equations are
 
  ∂C T
d q α × A r̈ + M q α × a β +
α T αβ
λ = q α × Qα , (12)
∂π

which can be derived from Eq. (34) in Appendix A; and the deformation equations are
 T
α ∂C
d α AT r̈ + M αβ a β + F e + λ = Qα , α = 1, 2, . . . , A. (13)
∂q α

The governing equations (11), (12), and (13) are derived from the Lagrange equations with
constraints and are the fundamental equations in this work. The deformable body is spatially
discretized by nodal coordinates and slopes in the reference coordinate system, and the
current approach will be referred to as the referenced nodal coordinate formulation (RNCF).

3 Discussions

Equations (11), (12), and (13) are universal in multibody system dynamics, and the govern-
ing equations for the rigid bodies and flexible bodies can be derived from them.

3.1 Rigid body

The body coordinate system of a rigid body is automatically its RCS, and the nodal position
vector q α = q ◦α are constant vectors. Hence, the acceleration of the node is
 
a α = ω̇ × q ◦α + ω × ω × q ◦α ,

and direct calculation yields

  
d α a α = m ω̇ × r ◦c + ω × ω × r ◦c , M αβ q α × a β = J̊ω̇ + ω × J̊ω ,

where r ◦c is the mass center of the rigid body in the RCS, and J̊ is the rotatory inertia tensor
of the rigid body in the RCS, which can be evaluated by
˚ ˚
mr ◦c = ρ (x) r ◦ (x) dv = d α q ◦α , J̊ = ρ (x) r◦ (x) r◦ (x) dv = M αβ q◦α q◦β . (14)
T T

V V
A referenced nodal coordinate formulation 311

The governing equations (11) and (12) become

 
   ∂C T
mr̈ + mA ω̇ × r ◦c + ω × ω × r ◦c + λ = F,
∂r
 
  ∂C T
mr ◦c × AT r̈ + J̊ω̇ + ω × J̊ω + λ = t,
∂π

which are exactly the equations of motions for a constrained rigid body, where the applied
torque in the reference coordinate system can be evaluated by
˚
 T 
t= A (t) [r (x, t) − r(t)] × f (x, t) dv = q ◦α × Qα (t).
V

Especially, when the origin of RCS coincides with the mass center of the rigid body, such
that r ◦c = 0, the governing equations become the Euler–Poincaré equations with constraints

 T  T
∂C ∂C
mr̈ + λ = F, J̊ω̇ + ω × J̊ω + λ = t.
∂r ∂π

3.2 Floating frame of reference

If the displacement in the deformable body is small, the position vector can be written as

q α (t) = q ◦α + uα (t) , uα (t) = φ iα pi (t) , φ iα = φ i (xα ) , i = 1, 2, . . . , K,

where the displacements uα are expressed by modal coordinates φ i (x), xα labels the particle
corresponding to the undeformed nodal coordinates q ◦α = r ◦ (xα ), and the Einstein summa-
tion is also adopted. By direct calculations, the equations of motions (11) become
 
       ∂C T
mr̈ + A p̈ − A mr ◦c + p × ω̇ + A ω × 2 ṗ + ω × mr ◦c + p + λ
∂r
= F, (15)

where r ◦c is the rigid mass center as that in Eq. (14), and

˚
 
ϕi = ρ (x) φ i (x) dv = d α φ iα , = ϕ1, ϕ2, . . . , ϕK , p = (p1 , p2 , . . . , pK )T .
V

The equations of motions (12) become

       
mr ◦c + p × AT r̈ + Jω̇ + γ i − ηij pj p̈i + ω × (Jω) + 2ṗi i
+ ϒ ij pj ω
 
∂C T
+ λ = t, (16)
∂π
312 H. Ren, K. Yang

where
˚
γi = ρ (x) r◦ (x) φ i (x) dv = M αβ q ◦α × φ iβ ,
V
˚
j
ηij = ρ (x) φ i (x) × φ j (x) dv = M αβ φ iα × φ β ,
V
˚
ρ (x) r◦ (x) φi (x) dv = M αβ q◦α φ iβ ,
T T
i
=
V
˚
ρ (x) φi (x) φj (x) dv = M αβ φ iα φ β .
T T j
ϒ =ij

 
and the current rotatory inertia tensor J = J̊ + pi i + iT + ϒ ij pi pj , in which J̊ the rigid
rotatory inertia tensor in Eq. (14). The equations of motions (13) become
 T  T
ϕ iT AT r̈ + γ i − ηij pj ω̇ + I ij p̈j − 2ωT ηij ṗj − ωT i
+ ϒ ij pj ω + Fei
 
∂C T
+ λ = Qi , i = 1, 2, . . . , K, (17)
∂pi

where the geometrically nonlinear effects in Fei can be taken into considerations, and
˚
j
I ij = ρ (x) φ iT (x) φ j (x) dv = M αβ φ iα · φ β ,
V
˚
Qi (t) = φ iT (x) f (x, t) dv = φ iα · Qα (t).
V

The governing equations (15), (16), and (17) are mathematically equivalent to the floating
frame of reference method [3, 11, 15] and provide the equations of motions for flexible
bodies with small or moderate deformations and large rotations.
In practice, the displacements are small compared with the nodal coordinates, that is,
 
q α (t) ≈ q ◦α , v α ≈ u̇α + ω × q ◦α , a α ≈ üα + 2ω × u̇α + ω̇ × q ◦α + ω × ω × q ◦α .

Then the equations of motions (15) become

 T
  ∂C
mAT r̈ − mr ◦c × ω̇ + p̈ + ω × 2 ṗ + mω × r ◦c + AT λ = AT F . (18)
∂r

The equations of motions (16) become

 T
  ∂C
mr ◦c × AT r̈ + J̊ω̇ +  p̈ + ω × J̊ω + 2ṗi i
ω+ λ = t, (19)
∂π

where
 
 = γ 1, γ 1, . . . , γ K
A referenced nodal coordinate formulation 313

Fig. 1 Schematic of the C 1

beam element

The equations of motions (17) become

 T
∂C
ϕ iT AT r̈ + γ iT ω̇ + I ij p̈j − 2ωT ηij ṗj − ωT i
ω + Fei + λ = Qi , i = 1, 2, . . . , K,
∂pi
(20)
where Fei are calculated from the linear elasticity. In this simplification, the most compli-
cated parameter ϒ ij will not be calculated. Moreover, it should be pointed out that if AT r̈,
ω̇, and p̈ are taken to be the acceleration variables, then the corresponding mass matrix in
Eqs. (18), (19), and (20) is actually a constant matrix, which can be useful in practice.

3.3 Absolute nodal coordinate formulation

When the RCS is chosen to be an inertia coordinate system, then the derived governing
equations are the same as those of the ANCF. In this scenario,

r̈ = 0, ω = 0, ω̇ = 0, a α = q̈ α ,

and the governing equations become

 T
∂C
M αβ q̈ β + F αe + λ = Qα .
∂q α

The mass matrix is a constant matrix, and both M αβ and F αe can be evaluated by the finite
element techniques.

4 Applications

The RNCF approach is developed to study the dynamics of flexible body systems with
high-speed rotations and/or large-travels. In practice, the reference coordinate system can be
flexibly chosen for convenience. Most of the flexible components in the following examples
are beam/rod structures, and the three-dimensional ANCF beam C 1 elements developed
in [5] will be adopted to discretize those structures. A schematic of the element is shown
in Fig. 1. The position and slope vectors at both terminal nodes, and the cross-section slope
vectors at the middle node are taken as the generalized coordinates, such that the generalized
314 H. Ren, K. Yang

Fig. 2 Schematic of the rotating

beam

1/2 1/2
coordinate vectors q α in Eq. (2) with α = 1, 2, . . . , 10 are r 0 , r 0x , r 0y , r 0z , r y , r z , r 1 , r 1x ,
r 1y , and r 1z . The position vector r (x, t) and the slope vectors r y (x, t) and r z (x, t) are
independently interpolated, so that those position and slope vectors can be written as

ds α
r (x, t) = s α (x) q α (t) , r x (x, t) = (x) q α (t) ,
dx
r y (x, t) = y α (x) q α (t) , r z (x, t) = zα (x) q α (t) ,

where the nonzero shape functions in s α (x), y α (x), and zα (x) are

s 1 = 1 − 3ξ 2 + 2ξ 3 , s 2 = l(ξ − 2ξ 2 + ξ 3 ), y 3 = z4 = 1 − 3ξ + 2ξ 2 , y 5 = z6 = 4ξ(1 − ξ ),
s 7 = 3ξ 2 − 2ξ 3 , s 8 = l(ξ 3 − ξ 2 ), y 9 = z10 = ξ(2ξ − 1),

in which l is the length of the element in the reference configuration, and ξ = x/ l. The
position vector r (x, t) is interpolated by Hermite polynomials, and r y (x, t) and r z (x, t)
are interpolated by the second-order Lagrange polynomials such that the slope vectors r x ,
r y , and r z are all second-order polynomials and can achieve the same order of accuracy.
The high accuracy of the element has been validated in [5]. This beam/rod element will be
adopted to both ANCF and RNCF approaches in the following examples.

4.1 High-speed rotating systems

Suppose the reference coordinate systems are prescribed with given r(t) = 0 and A(t). Then
Eqs. (11) and (12) are not necessary. For high-speed rotation systems, the reference coor-
dinate system is chosen to be rotating with a tractive joint and the rotation motion A(t) is
prescribed. Hence ω(t) and ω̇(t) are both known. The governing equations (13) become
 T
   ∂C
M αβ q̈ β + 2ω × q̇ β + ω̇ × q β + ω × ω × q β + F αe + λ = Qα . (21)
∂q α

4.1.1 Spin-up maneuver

The spin-up of a rotating beam, as shown in Fig. 2, is a standard benchmark for beams
with large rotations [16, 17]. The original maneuver is a planar mechanism, and to calcu-
late its dynamics by three-dimensional formulations, the maneuver will be generalized to a
spatial mechanism. Suppose the cross section of the beam is hollow and circular, with area
0.7299 cm2 and moment of inertia 0.8215 cm4 ; the length of the beam is 8 m; the material
of the beam is uniform and isotropic, with density ρ = 2766.67 kg/m3 , Young’s modulus
E = 68.95 GPa, and Poisson’s ratio ν = 0.3. One end of the beam is fixed at the hub, and its
rotation motion is about the z axis, prescribed by
  2
ωs t 2
+ Ts cos 2π t
− 1 , 0 ≤ t ≤ Ts ,
ϑ(t) = Ts  2 T 2π Ts (22)
ωs t − 2s , t > Ts ,
A referenced nodal coordinate formulation 315

Fig. 3 The (a) longitudinal and (b) transverse deflections at the tip of the spin-up manoeuvre, calculated by
the RNCF approach (dash-dotted) and the ANCF approach (dashed), respectively

where Ts = 15 s and ωs = 4 rad/s. The corresponding angular velocity and acceleration are
ω(t) = zϑ̇(t) and ω̇(t) = zϑ̈(t), with
 
ωs
t − Ts
sin 2π t
, 0 ≤ t ≤ T , ωs
1 − cos 2π t
, 0 ≤ t ≤ Ts ,
ϑ̇(t) = Ts 2π Ts s
ϑ̈(t) = Ts Ts
ωs , t > Ts , 0, t > Ts .

The rotating beam can be formulated by both ANCF and RNCF approaches, and the beam
structure will be spatially partitioned into multiple beam elements of uniform lengths, with
each element described by Fig. 1. In the ANCF approach the position and slope vectors in
the spatial coordinate system are directly taken to be the generalized coordinates, whereas in
the RNCF approach the origin of the RCS is located at the hub such that r(t) = 0, and A(t)
describes the coordinate system revolving with the rotation shaft, that is, rotating about the
z axis and following the motion described by ϑ(t) in Eq. (22). The generalized coordinates
are the position and slope vectors of each node in the reference coordinate system. Since
the cross section of the beam is circular, the stretch and bending deformations, which are
mainly described by the position vectors of the centroid line, are decoupled with the twist
deformations, which are mainly described by the cross section slopes, and the configuration
of the beam depends only on the the position vectors. Moreover, the other end of the beam
is free with no constraint equation attached, so the equations of motions in Eq. (21) become
  
M αβ q̈ β + 2ω × q̇ β + ω̇ × q β + ω × ω × q β + F αe = 0.

The position vectors q α = 0 except at the hub, and the excitation loads are provided by the
RCS. The tip deflections are measured with respect to the RCS, and the deflections with
respect to its three axes are denoted by u, v, and w, respectively. The dynamics of the
beam is calculated by the current RNCF approach and the ANCF approach in [5], respec-
tively, and six elements described by Fig. 1 are adopted in each approach. The calculated
results are in good agreement with each other, as shown in Fig. 3. By Eq. (21), besides
M αβ q̈ β , there are three other types of inertia forces applied on the manoeuvre: the rotational
   
inertia force ω̇ × M αβ q β , the Coriolis force 2ω × M αβ q̇ β , and the centrifugal force
  
ω × ω × M αβ q β , which are all perpendicular to the z axis. The position vectors q β are
within the x-y plane, such that all those three inertia forces are within that plane, which can
only lead to deflections with in the x–y plane; the elastic forces in Eqs. (7) and (8) are linear
combinations of the position vectors, whose z components are zeros; this further guarantee
316 H. Ren, K. Yang

Table 1 Comparisons of the number of elements, and adaptive stepsizes and computational times required by
the adaptive-stepsize HHT integrator for different formulations and steady rotational speeds in the manoeuvre

ωs = 4 rad/s ωs = 10 rad/s ωs = 40 rad/s ωs = 100 rad/s

Planar ANCF RNCF Planar ANCF RNCF Planar ANCF RNCF Planar ANCF RNCF

No. 4 6 6 8 10 10 16 24 24 48 72 72
elements
Adaptive 4050 4050 2007 9289 9294 2007 35971 38675 2028 96479 89502 2106
steps
Time 2.15 12.6 6.0 19.1 49.0 9.4 314 469 24.2 4132 3862 91.0
(sec)

the z-component of q̈ β to be zeros, and no z-direction motions can be excited. As a result,

there is no out-of-plane deflection. The deflections of the manoeuvre in the x–y plane can
be partitioned into two states:
1. In the transient state, ω̇ = 0, and the deflection motion is not zero under the rotational
inertia force, which further leads to q̇ β = 0, so that the Coriolis force is also imposed,
and the manoeuvre endues enforced deflection motions;
2. In the steady state, ω̇ = 0, and the hub rotates at a constant angular velocity along the
z axis. As time goes by, q̇ β → 0, and both the rotational inertia and Coriolis forces will
vanish. The centrifugal force is along the x axis and is balanced by the elastic forces,
which leads to a steady-state stretch along the rod, as shown in Fig. 3(a).
The manoeuvre is essentially a planar mechanism and can also be formulated by the planar
ANCF described in [3]. All three formulations are calculated by the adaptive-stepsize HHT
integrator [18] with the relative error 10−6 . The results calculated from all three approaches
agree very well, as shown in Fig. 4. As the ultimate angular velocity ωs increases, the in-
ertia forces increase, and the deflections are amplified. Moreover, the comparisons of the
computational costs are listed in Table 1. More elements are required to capture the large
deformations, and a series of tests have been performed to find out the minimum numbers
of elements required for each ultimate rotational speed ωs . The RNCF approach requires
the same number of elements as the ANCF approach to describe the same large deformation
to the same accuracy, but the computational costs are significantly reduced in RNCF. It is
interesting to notice that the adaptive stepsizes required in the RNCF approach are almost
the same for different ultimate rotational speeds, even if the number of elements are quite
different, whereas the numbers of stepsizes required in ANCF are highly influenced by the
rotational speeds. Furthermore, the same approach also works for curved beam configura-
tions, where the beam can bend in the opposite direction of the rotation (upwind), as shown
in Fig. 5(a); in the same direction of the rotation (downwind), as in Fig. 5(b), or bend out of
plane, as in Fig. 5(c). All three curved beams are of length 8 m with initial curvature radius
R = 32/π m, and all the other parameters are the same as before. The dynamics of the three
curved beams are calculated by the current RNCF approach with 8 C 1 elements in [5] and 20
elements of the ANCF elastic line approach in [16]. The first two curved beams still execute
planar motions, and the results calculated from both approaches are in good agreement, as
shown in Figs. 6(a) and 6(b). The dynamics in the third scenario are more complex, and the
results agree perfectly well, as shown in Fig. 6(c). In the first two scenarios, all the position
vectors q β are within the x–y plane, such that the inertia forces and the elastic forces are
always in the x–y plane, and there is no out-of-plane deflection. However, the steady-state
transverse deflections v are not zeros anymore, because the centrifugal forces are no longer
 A referenced nodal coordinate formulation

Fig. 4 Tip deflections of the spin-up manoeuvre with steady rotational speeds (a) and (d) 10 rad/s, (b) and (e) 40 rad/s, as well as (c) and (f) 100 rad/s, calculated by the planar
ANCF (solid), spatial ANCF (dotted), and the RNCF (dash-dotted) approaches, respectively
317
318 H. Ren, K. Yang

Fig. 5 The (a) upwind, (b) downwind, and (c) out-of-plane configurations of the spin-up manoeuvre of
curved beams

along the beam. In the third scenario, the initial position vectors q β have z-components, the
elastic force can have z-component in the transient state, and then out-of-plane deflection
emerges. In the steady state, both the rotational inertia force and the Coriolis force die out.
The centrifugal force will be along the x-direction if all the position vectors are in the x–z
plane, and the y-component of the steady deflections will stay zeros, as in Fig. 6(c). The
deflections can be easier to be calculated and physically explained in RNCF than in ANCF.

4.1.2 High-speed motor

Consider a simple model for a high-speed motor, as shown in Fig. 7, which is a high-speed
rotating shaft with an inertia imbalanced rotor at the middle. Suppose the material of the
motor is uniform and isotropic. The length of the shaft is L = 1.02 m, Young’s modulus is
210 GPa, Poisson’s ratio is 0.3, the radius of the shaft is r0 = 0.05 m, and the mass density is
7800 kg/m3 . The parameters of the rotor include the radius R = 0.1 m, thickness d = 0.01
m, and imbalance deviation h = 0.05 m. Suppose one end of the shaft is clamped at O
on the ground, whereas the other end is simply pinned at the position Lx. The clamped
boundary condition at O indicates that r(t) = 0 and A(t) is prescribed, so that the shaft can
only revolve about the x axis. The pinned boundary condition at the other end of the shaft
provides the constraint equations for Eq. (21), which can be written as

r (L, t) − Lx = 0.

The shaft is considered as a deformable rod, whereas the rotor is taken as a rigid body. Sup-
pose that the constitutive relation of the material remains to be St. Venant–Kirchhoff elastic
even in large deflections, whereas the geometrically nonlinear effects must be taken into con-
siderations. The rotating shaft can be formulated by both ANCF and RNCF approaches and
discretized by the element described in Fig. 1, and the governing equations are calculated by
the adaptive-stepsize HHT integrator [18] with the relative error 10−7 . Suppose the motor
speeds up from the static configuration, and the spin-up process is also described by Eq. (22)
with Ts = 1 s. Then the motor rotates in a steady rotational speed ωs . The deflections of the
shaft for different steady speeds are shown in Fig. 8. The results calculated by the ANCF
approach and those by the RNCF approach agree very well. The computational efficiencies
of the ANCF approach and the RNCF approach are compared in Table 2. The computa-
tional costs of the RNCF approach are much lower than those of the ANCF one. The details
 A referenced nodal coordinate formulation

Fig. 6 Tip deflections of the (a) upwind, (b) downwind, and (c) out-of-plane spin-up manoeuvre of curved beams, calculated by the RNCF approach (dashed) and the ANCF
approach (dash-dotted)
319
 320

Table 2 Comparisons of the number of elements, and adaptive stepsizes and computation times required by the adaptive-stepsize generalized-α integrator for different formu-
lations and steady rotational speeds in the rotor

ωs = 7.5 krpm ωs = 10 krpm ωs = 12.5 krpm ωs = 15 krpm ωs = 17.5 krpm ωs = 20 krpm

ANCF RNCF ANCF RNCF ANCF RNCF ANCF RNCF ANCF RNCF ANCF RNCF

No. elements 4 4 8 8 12 12 16 16 24 24 32 32
Adaptive steps 84377 127 112710 131 187668 145 213713 1036 222129 1655 253934 3246
Time (sec) 174 0.50 398 0.79 1057 1.68 1809 41.0 4660 105 6460.5 309
H. Ren, K. Yang
A referenced nodal coordinate formulation 321

Fig. 7 Schematic of the rotating

beam

Table 3 Inertia properties of the two rigid bodies in the slider-crank mechanism

m (kg) Jx (kg m2 ) Jy (kg m2 ) Jz (kg m2 )

crankshaft 0.36 0.002727 0.002727 0.001

slider block 0.075552 0.01 0.01 0.01

in the wavy oscillations at the steady state are calculated and shown in Fig. 9. It seems that
the wavy deflections do not influence the computational efficiency of the ANCF approach,
but can significantly increase the computational costs in the RNCF approach. However, the
RNCF approach is still much faster than the ANCF approach in all the scenarios. In this
example, the imbalance of the motor is the chief source of excitation; the deflections of the
shaft are secondary sources of excitations; and their relations lead to three scenarios:
1. When the steady speed is small or moderate, the deflections caused by the chief excita-
tion are small, such that those secondary excitations are insignificantly small, and their
influences are actually filtered out by the error control strategies of the integrator, and the
deflections appear to be steady states, as those in the first two scenarios of Fig. 8;
2. When the steady speed is high enough, the deflections caused by the chief excitation are
quite large, such that the wavy motions caused by the secondary excitations are obvious,
as those in the third scenario of Fig. 8 and those in Fig. 9;
3. When the steady speed is extremely high, the wavy motions caused by the secondary
excitations could be more important, and dynamical instability could appear in the shaft
deflections, which is a self-excitation phenomenon.
Shannon’s sampling theorem requires that the stepsizes of the integrator should be small
enough to catch all the significant variations in the generalized coordinates. In this sense,
the RNCF approach separates the background rigid-body rotations from deformations, and
provides an efficient method to formulate and calculate the deflections in a high-speed motor.

4.2 Slider-Crank mechanism

Consider a spatial flexible slider-crank mechanism shown in Fig. 10. The mechanism con-
sists of four bodies: the ground, the crankshaft, the connecting rod, and the slider block. The
crankshaft and the slider block are formulated by rigid bodies with inertia parameters listed
in Table 3. The crankshaft is connected to the ground at A through a revolute joint rotating
about the x axis, and its length L2 = 0.14 m. The slider block is connected to the ground at
B through a translational joint, such that it always moves along the x axis. The connecting
 322

Fig. 8 Deflections of the motor with steady rotational speeds (a) and (d) 104 rpm, (b) and (e) 1.25 × 104 rpm, as well as (c) and (f) 1.5 × 104 rpm, calculated by the ANCF
approach (dotted) and the RNCF approach (dash-dotted)
H. Ren, K. Yang
 A referenced nodal coordinate formulation

Fig. 9 The wavy deflections of the motor with high speeds (a) and (d) 1.5 × 104 rpm, (b) and (e) 1.75 × 104 rmp, as well as (c) and (f) 2 × 104 rpm, calculated by the ANCF
approach (dotted) and the RNCF (dash-dotted) approach
323
324 H. Ren, K. Yang

Fig. 10 (a) configuration and (b) schematic of the slider-crank mechanism

rod is a uniform flexible rod, whose unstretched length L1 = 0.30 m, and its cross-section
is circular with radius 0.008 m. The mass density of the connecting rod is 7870 kg/m3 ; its
Young’s modulus and Poisson’s ratio are 2.0 × 108 N/m2 and 0.29, respectively. The con-
necting rod is connected to the crankshaft at C through a universal joint, and the slider block
through a spherical joint at B. The connecting rod can be formulated by either the ANCF
approach or the corresponding RNCF approach, and spatially discretized by the element
described by Fig. 1. In the RNCF approach the reference coordinate system is chosen to be
the position vector and the local coordinate frame at C on the connecting rod, which is also
the tractive joint of the connecting rod. The initial configuration of the mechanism is that
the universal joint C is located at the origin O, such that the x coordinate of B is 0.3 m,
and the initial velocity of each part of the mechanism is zero. The mechanism is driven by a
rotational motion of the revolute joint A, which is prescribed to be

ϑ(t) = STEP5 (t, 0 sec, 0, 0.8 sec, ϑs ) , (23)

where the STEP5 function is defined as

⎧
⎨ f0 ,  t < t0 ,
STEP5 (t, t0 , f0 , t1 , f1 ) = f0 + (f1 − f0 ) ζ 3 10 − 15ζ + 6ζ 2 , t0 ≤ t ≤ t1 , (24)
⎩
f1 , t > t1 ,

where ζ = (t − t0 ) / (t1 − t0 ). The function STEP5 is C 2 continuous with respect to t , so

that the rotational motion ϑ(t), the angular velocity ϑ̇(t), and the angular acceleration ϑ̈(t)
are all continuous functions. Moreover, suppose the gravitational acceleration is negligible.
The connecting rods are formulated by the ANCF and RNCF approaches with the element
described by Fig. 1, and the number of elements are taken to be 10 and 20 for comparisons.
To speed up the simulation, a small viscous damping is imposed in the elastic forces, as
described in Eq. (7), with the coefficients c = 10−3 and 10−2 s−1 . Five different options of
ϑs are calculated, with ϑs = 2π, 4π, 6π, 8π , and 10π , and the governing equations of the
mechanism are solved by a Lie group generalized-α integrator [19], with the local relative
error threshold prescribed to be 10−7 . The adaptive time step required in each scenario is
listed in Table 4. The motions of the slider block are calculated in each scenario. Moreover,
when the connecting rod is considered to be a rigid body, the corresponding motions of the
slider block are also calculated. The differences between the motions calculated from flexi-
ble rods and those calculated from the rigid rod, which are referred to as the deflections of
A referenced nodal coordinate formulation 325

Table 4 Comparisons of the adaptive time steps required in the calculations of the flexible slider-crank
mechanism

c = 10−3 s−1 c = 10−2 s−1

ϑs = 2π ϑs = 4π ϑs = 6π ϑs = 8π ϑs = 10π ϑs = 2π ϑs = 4π ϑs = 6π ϑs = 8π ϑs = 10π

20 ANCF 22243 32297 40070 44233 61476 10448 21696 23323 29991 37978
elements
10 ANCF 21727 32803 39180 42639 65381 10603 21775 25866 31498 39466
elements
20 RNCF 10264 17727 21047 28952 35051 3595 9646 12797 15089 20261
elements
10 RNCF 10127 17105 17902 23676 31009 3679 9810 13285 15919 21006
elements

the rods with respect to rigid rods, are computed to evaluate the accuracies in the descrip-
tions of the flexible bodies. The deflections of the rods with respect to rigid rods in different
scenarios are depicted in Fig. 11. The results indicate that:
1. The results calculated from the ANCF approach and those from the RNCF approach are
almost the same, when the same number of elements is adopted;
2. The adaptive number of steps required by the integrator are insensitive to the number of
elements but strongly depend on the underlying physics and formulations;
3. The number of time steps required by RNCF is fewer than those required by ANCF;
4. Small damping coefficient can significantly improve the simulation efficiency, but the
deflections are slightly changed;
5. Faster rotational motions require more elements to achieve the same accuracy and more
time steps to track and describe the large deformations.

4.3 Electric solar wind sail spacecraft

The dynamics of a spacecraft with a new propulsion concept in the solar system, the electric
solar wind sail [20], will be studied by the RNCF. The electric solar wind sail (E-Sail) uses
charged long tethers to extract thrusts from the solar wind. A schematic diagram of the
E-sail spacecraft is shown in Fig. 12. The lengths of the tethers are tens of kilometers, and
protons in the solar wind are repelled by those charged tethers, and then the resulting electric
Coulomb field pushes the spacecraft outward from the sun. The solar wind strength decays
as r −2 , where r is the distance from the Sun to the spacecraft, and the effective area of the
sail is proportional to the electron sheath width surrounding the tethers, which scales as r in
the solar wind environment. The E-sail thrust force is proportional to the product of the solar
wind strength and the effective sail area, which scales as r −1 and decreases slower than the
thrust forces of the other concepts, such as the solar sail (r −2 ). The E-sail is suitable for deep-
space missions inside the solar system with r ≥ r⊕ , where r⊕ = 1 AU ≈ 1.5 × 108 km is the
average radius of the Earth orbit. Janhunen [20] proposed to build such a spacecraft with
an E-sail of ∼ 1 N thrust force, ∼ 100 kg mass, and ∼ 10 year lifetime for such missions.
Theoretical analysis [21] and experiments [22] show that the thrust force on a unit length
tether can be written as

dF
≈ 0.18 max (0, V0 − V1 ) 0 mp no u2⊥ ,
dl
 326

Fig. 11 The deflections of the slider block with respect to rigid rods calculated with different ϑs and damping coefficient c
H. Ren, K. Yang
A referenced nodal coordinate formulation 327

Fig. 12 A schematic diagram of

the E-Sail spacecraft

where 0 = 8.854 × 10−12 F/m is the vacuum permittivity of electric fields, mp = 1.67 ×
10−27 kg is the mass of a proton, u⊥ is the speed of the solar wind perpendicular to the
tangent of the tether, V0 is the electric potential of the wire, V1 ≈ 1 kV is the environmental
electric potential, and no is the solar wind electron density, which is proportional to r −2 . The
thrust force is always perpendicular to tether and away from the Sun and can be written as
dF  r⊕ r
≈ 0.18 max (0, V0 − V1 ) 0 mp n⊕ u2 i × ×i
dl r r
r⊕ r
≡ c max (0, V0 − V1 ) i × ×i , (25)
r r
where n⊕ = 7.3 cm−3 is the solar wind proton density at r⊕ , u = 400 km/s is the speed of the
solar wind, r is the vector from the Sun to the spacecraft such that r = r , c = 2.4 × 10−5
N/(kV·km), and i is the unit tangential vector of the tether. Typically, the voltage in the tether
is about 20 kV–40 kV, and the thrust is about 10−6 N/m. The gravitational acceleration at r⊕
is g = 5.93 × 10−3 m/s2 , and it is balanced by the centrifugal acceleration ωg2 r⊕ of the orbit,
where ωg = 1.99 × 10−7 rad/s, and the corresponding orbital speed is ωg r⊕ = 29.78 km/s.
The space tether in the E-sail is usually of length L ∼ 10 km, the gravitational gradients
can be represented by the acceleration gL/r⊕  10−9 m/s2 , and its action on the spacecraft
(∼ 100 kg) is ∼ 10−7 N, which can be neglected because both the gravitational and thrust
forces are ∼ 1 N. Hence it is accurate enough to take the gravitational acceleration for the
spacecraft to be uniform as −gr⊕2 r/r 3 .
In this work, we will study two E-sails, and their geometric configurations are similar,
as shown in Fig. 13, with different tether lengths. There is one remote unit attached at the
end of each main tether, and those tether ends are also connected by auxiliary tethers. The
proposed tethers are made of tens of µm-diameter aluminum wires [27], and the parameters
of the spacecrafts are listed in Table 5. The maximum thrust forces and the total gravitational
forces of the spacecraft at r⊕ in both scenarios can be estimated, where the prior ones are
slightly larger than the latter ones. The maximum thrust force decreases as r −1 , whereas the
gravitational force decreases as r −2 . So when r > r⊕ , the maximum thrust will always be
able to overcome the gravitational force and drive the spacecraft outward to the deep space.
From the MBS viewpoint, the E-sail spacecraft consists of the core module, that is, the
main cabin, and the engine, that is, the E-sail, and those two parts are connected by a rev-
olute joint. The core module can be considered as either a rigid body or a point mass, and
the E-sail is a flexible body, which consists of a series of main tethers, auxiliary tethers, and
328 H. Ren, K. Yang

Fig. 13 Geometric configurations of the E-sails with tether lengths (a) 10 km and (b) 20 km

Table 5 Parameters of the E-sail

spacecraft Physical quantities E-sail 1 E-sail 2

Mass of the core module mc (kg) 100

Mass of each remote unit mr (kg) 0.5
Material density of each tether ρ (kg/m3 ) 2700
Young’s modulus of each tether E (GPa) 70
Number of main tethers 100
Length of each main tether (km) 10 20
Diameter of each tether (µm) 80 78
Electric potential bias V0 (kV) 40 22
Maximum thrust force at r⊕ (N) 0.980 1.056
Total gravitational force at r⊕ (N) 0.975 1.052

remote units. Suppose the origin of absolute coordinate system (X, Y , Z) is located at the
mass center of the Sun, as shown in Fig. 12, and the spacecraft is described in a reference
coordinate system (RCS). The origin of the RCS is located at the revolute joint, and its po-
sition vector with respect to the absolute coordinate system is denoted by r(t). The rotation
matrix for the RCS is chosen to be the E-sail rotation coordinate system A = (x, y, z), and
the rotational axis of the revolute joint is along the z axis. Since the size of the core module
(∼ 10 m) is much smaller than that of the E-sail (∼ 10 km), the core module will be formu-
lated by a point mass, located at the origin of the RCS. The generalized coordinates of the
system are the position vector r(t), the rotation matrix A(t), and the relative nodal vectors
of the E-sail in the RCS q α (t). The kinetic energy of the core module and the remote unit
labeled by a, where a = 1, 2, . . . , 100, can be written as

1 1
Kc = mc ṙ · ṙ, Ka = mr (ṙ + v a ) · (ṙ + v a ) ,
2 2
where v a is the velocity vector for the remote unit, which coincides with that for the node at
the end of the ath tether. The tethers can be discretized by the elements in Appendix B, and
the kinetic energy of the element labeled by n is Kn , which is similar to that in Eq. (35). So
A referenced nodal coordinate formulation 329

the total kinetic energy can be written as the same formula in Eq. (4),
  1 1
K = Kc + Kn + Ka = mṙ · ṙ + d α ṙ · (Av α ) + M αβ v α · v β ,
n a
2 2

where q α are the nodal coordinates of the E-sail in the RCS, d α and M αβ are assembled
ij
from the corresponding element terms dni and Mn as those in Eq. (36), as well as the masses
of the remote units, and
 
m = mc + mn + ma , v α (t) = q̇ α (t) + ω(t) × q α (t).
n a

The elastic forces are only applied on the generalized coordinates q α , and the variation of
the elastic potential energy is
δ U = δq α · F αe ,
where the generalized elastic force F αe can be assembled from the corresponding element
elastic forces in Eq. (37) in Appendix B. The virtual work done by the gravitational force on
the core module and the remote unit labeled by a are
   
δ Wc = δr · F cg , δ Wa = δ r + Aq a · F ag = δr · F ag + δπ · q a × Qag + δq a · Qag ,

where q a is the position vector of the remote unit, which coincides with that of the node at
the end of the ath tether; and the generalized forces F cg , F ag , and Qag are

r⊕2 r r⊕2 r AT r
F cg = −mc g 3
, F ag = −mr g 3
, Qag = −mr gr⊕2 .
r r r 3
The virtual work done by the gravitational force and the solar wind thrust force on the tethers
can be written as
      
δ Wt = δr · F t + F tg + δπ · q α × Qαt + Qαg + δq α · Qαt + Qαg ,

where the generalized forces F t , F tg , Qαt , and Qαg are assembled from the corresponding
element generalized forces in Eq. (38). So the total virtual work done by the gravitational
force and the thrust force can be written as
  
δ W = δ Wc + δ Wt + δ Wa = δr · F + δπ · q α × Qα + δq α · Qα ,
a

where the lumped load F and nodal load Qα are

r⊕2 r r⊕2 AT r
F = F t − mg 3
, Qα = Qαt − d α g 3
.
r r
The expressions for the kinetic energy, potential energy, and the virtual work done by the
applied forces are the same as those in Eqs. (4), (9), and (10), so the derived governing
equations are the same as those in Eqs. (11)–(13). The constraint equations come from the
navigation algorithm of the spacecraft. In this work, we suppose that the spacecraft moves
in such a way that the spin axis of the E-sail is always pointed to the sun, that is,
r(t) r(t)
· x(t) = 0, · y(t) = 0, (26)
r⊕ r⊕
330 H. Ren, K. Yang

which provide two constraint equations to the system dynamics and can also ensure the
stability of the motion. Next, it is time to consider the geometrical boundary conditions for
each main tether. In the RCS, one end of each main tether is fixed to the hub, and the position
vector and the slope at the corresponding end of the ath main tether are

pa = 0, pax = ua t a , a = 1, 2, . . . , 100, (27)

where the norm of the tangent ua are taken as the generalized coordinates instead of pax , and
 T
2aπ 2aπ
t a = cos , sin ,0 .
N N

Imposing the boundary conditions in Eq. (27) includes eliminating the equations with re-
spect to pa and making a dot product of t a to the equations with respect to pax , to get equa-
tions for ua . Moreover, suppose the spacecraft is initially located at the X axis, traveling as
a planet whose orbit is circular in the X–Y plane; the −z direction of its body coordinate
system is pointed to the sun; and the E-sail is spinning about the z axis of its body coordinate
system, with an initial spin rate ωs . Thus, the initial conditions for the RCS are
⎛ ⎞ ⎛ ⎞ ⎛ ⎞ ⎛ ⎞
r⊕ 0 0 1 0 −ωg
r(0) = ⎝ 0 ⎠ , A(0) = ⎝ 0 1 0 ⎠ , ṙ(0) = ⎝ ωg r⊕ ⎠ , ω = ⎝ 0 ⎠ ,
0 −1 0 0 0 ωs
√
where ωg = g/r⊕ is the orbit angular velocity. The initial conditions for q α should be
calculated from the equilibrium state of the tethers. At time t = 0− the tether electric voltage
is lower than V1 , and no thrust force is applied to the tethers, that is, F t (0) = 0 and Qαt (0) =
0. Suppose that the deformation of q α is balanced by the centrifugal force, that is,

q̇ α (0) = 0, M αβ ωs k̂ × ωs k̂ × q α (0) + F αe (0) = 0.

Then at time t = 0+ the electric voltage is set to V0 , and the thrust force is immediately
imposed. The tether will deform at the thrust force, then the deformation will change the
geometric configuration of the E-sail, and the thrust force will be significantly reduced. The
spin motions of the E-sail can provide tensions to the tethers to resist the deformation and
maintain the thrust force. The bending stiffness of each tether is very small, and the spin
rate ωs can hardly be achieved by the spin-up process in Sect. 4.1.1. The E-sail spins are
proposed to be achieved by two effects [23]:
1. Tether Deployment. During the tether deployment, the remote units can be speeded up,
and the conservation of angular momentum will provide the start-up spin rate ωs ;
2. Coriolis Effect. By an additional modulation on the electric voltage [24, 25] when the
spacecraft orbits around the sun, a lateral thrust force can be generated to drive the tether
rotation to accumulate gradually.
Unfortunately, neither the deploy process nor the voltage modulation strategy will be studied
in this work. To study the deploy process, an arbitrarily Lagrangian–Euler (ALE) technique
[26] will be necessary, because the lengths of the tethers are time variant; the control strategy
of the electric voltage can be studied by the current formulation, and this will be discussed
in following works. In this work, the sail spin rate will be assumed, and the dynamics of the
spacecraft will be studied.
A referenced nodal coordinate formulation 331

Fig. 14 Radial displacements of the E-sails with radii of (a) 10 km and (b) 20 km for several spin rates

Each main tether is discretized by 16 tether elements. The auxiliary tethers are automat-
ically partitioned into 100 segments, and each segment is formulated by one tether element.
As a result, there are 1700 elements in total, and the degree of freedom of the system is
10006. There are two constraint equations in the system, and the corresponding constraint
torque can also be calculated. The governing equations are calculated by a generalized-α Lie
group DAE integrator [19] with adaptive stepsizes, and the relative error in the integrator is
set to be 10−7 . All the formulations are programmed in Matlab-2019. The RNCF technique
is suitable to calculate the dynamics of the E-sail spacecraft, because:
1. The reference coordinate system is much more efficient in descriptions of the spin mo-
tions of the E-sail than the absolute coordinate system, as studied in Sect. 4.1;
2. The simplified formulations in Appendix B can significantly reduce the computational
costs in each element.
The RNCF approach is quite efficient, and it takes about 1 hour to run a 200-second-long
simulation of the full-scale Matlab model in the desktop. The trajectory of the spacecraft
can be accurately calculated, where the spacecraft is generally moving along the orbital
trajectory, but the thrust force will push the spacecraft radially outward to the deep space,
and this is an acceleration process, as shown in Fig. 14. It appears that if the configuration
of the E-sail is close to planar, then it can provide enough thrust force to drive the spacecraft
off the orbital trajectory into the deep space. The deformed configurations of the E-sails at
the 200-second simulation are shown in Fig. 15. The z deflections of the E-sails are less
than 80 m, which are small compared with the radii 10 km and 20 km. Under the control
strategy in Eq. (26), the spin rates seem to be conserved in 2000 s, as shown in Fig. 16,
with insignificant wrinkles. The deformation of an E-sail depends on the balance of the
tensions and the applied forces, that is, the gravity and the thrust. The tension in a main
tether is mostly generated by the centrifugal force, and it is proportional to ωs2 , whereas the
thrust load in Eq. (25) is roughly proportional to the electric voltage. The dimensionless
number in the E-sail is the tension over the maximum trust force in the tether. The total
maximum thrust force at the E-sail is about 1 N, and the maximum thrust force at each main
tether is about 0.01 N. The tension in each tether is calculated, and the results are shown in
Fig. 17. The tensions in the auxiliary tethers are positive and slightly smaller than those of
the corresponding ones in the main tethers, which indicates that the circumferencial shape
of the E-sail is maintained. As shown in Fig. 17(a), when ωs = 0.5◦ /s, the tension of the 10
km E-sail is about 40 times of that of the thrust force, and the tension becomes wavy, which
332 H. Ren, K. Yang

Fig. 15 Configurations of the E-sails with radii of (a) 10 km and (b) 20 km at the 200 second with different
initial spin rates

Fig. 16 Calculated spin rates of the E-sails with radii of (a) 10 km and (b) 20 km, with different initial spin
rates

Fig. 17 Tensions of the E-sails with radii of (a) 10 km and (b) 20 km at different initial spin rates

means that the shape of the E-sail is not well retained. When ωs gets lower, the tension may
not be large enough to resist the thrust force, and the shape of the E-sail can be distorted.
As shown in Fig. 18, when ωs = 0.25◦ /s in the 10 km scale E-sail, and the steady tension
 A referenced nodal coordinate formulation

Fig. 18 The (a) tensions, (b) radial displacements, and (c) constraint torques for the 10-km E-sail spacecraft with small spin rates
333
334 H. Ren, K. Yang

Fig. 19 The constraint torques of the E-sails with radii of (a) 10 km and (b) 20 km at different spin rates

is only about 10 times of that of the maximum thrust force on the tether, a bifurcation
emerges. The tensions in the auxiliary tethers can decrease dramatically as in Fig. 18(a),
which demonstrates that those tethers are buckled. In this scenario, the configuration of the
E-sail is screwed along the circumference; the total thrust force decreases significantly as in
Fig. 18(b); and the constraint forces become unpredictable as in Fig. 18(c), which indicates
that the spacecraft is hard to be manoeuvred. Those phenomena are all to be avoided in
practice, and the spin rate of the E-sail must be large enough to provide enough tensions
to the tethers to ensure the E-sail to work properly. On the other hand, the stretches in the
tethers are still small in practice. The axial stiffnesses of the tethers are all greater than 300
N, whereas the tensions are all less than 4 N, so that the stretch strains in the tethers are less
than 1.3%, which indicates that the assumption r x ≈ 1 is generally satisfied, and thus the
element developed in Appendix B is good enough for descriptions of the E-sail. The angular
momentum of the spacecraft is not conserved, and the constraint torques are calculated, as
shown in Fig. 19, to provide the manoeuvre torques against the orbital motions. The higher
the spin rates, the larger the constraint forces required; the wider the E-sail spans, the harder
the spacecraft can be manoeuvred. To achieve a proper thrust force while at the same time
to minimize the control energy, the size of the E-sail should not be too large, and the spin
rate should be neither too large nor too small, which will be studied in the future.

5 Comments
To describe a deformable body undergoing large-distance travels and high-speed rotations,
it is recommended to separate the generalized coordinates with respect to the rigid motions
and those with respect to large deformations. RNCF provides an effective way to achieve
this purpose. It is different from other approaches in the following aspects:
1. Unlike the corotational formulations, all the elements are described in the same RCS,
and the large deformations are described by the nodal coordinates and slopes in the RCS
with no interpolation on rotational motions;
2. Unlike GEF or ANCF, all the generalized coordinates related to large deformations are
described in a corotational RCS, to be separated from the rigid motions with large-
distance travels and high-speed rotations;
3. The governing equations in RNCF are derived equivalently from the traditional Lagrange
equations with constraints, and they are simple and elegant, as in Eqs. (11)–(13), which
can be used for theoretical studies;
A referenced nodal coordinate formulation 335

4. The inertia forces caused by large distance travels and high-speed rotations are handled
as common applied forces, and the resultant deformations can be effectively calculated;
5. RNCF is suitable to describe mechanical systems with multiple scales, such as:
(a) Two different length scales: a large scale for large-distance travels, and a small scale
for deformations;
(b) Two different time scales: a large scale for long-time travels, and a small scale for
high-speed rotation periods.
6. The generalized coordinates for large-distance traveling and high-speed rotating rigid
motions and those of the deformations are automatically separated, and cosimulation
techniques can be directly applied to the simulations;
7. The governing equations for rotational motions in Eq. (34) are universal and can be
adopted in generic scenarios;
8. Lie group integrators can be applied to calculate high-speed rotational motions, which
seems very efficient.

6 Conclusions
The governing equations for RNCF are more general than other equations in multibody
system dynamics, and a simplified version of the floating-frame of reference formulation
can be derived from them, where the mass matrix is almost a constant one. To calculate the
dynamics of a flexible body in high-speed rotating system, the same number of elements
are required for both ANCF and RNCF approaches to achieve the same accuracy, but the
computational costs in RNCF are much lower than those in ANCF. The dynamics of the
electric solar wind sail can be easily formulated by RNCF, and the full-scale governing
equations are suitable for long-time simulations. The spin rate of the E-sail should be able
to provide tensions of about 100 times of the thrust load to resist the chaotic deformations
in the tether and provide enough thrust force for the spacecraft. RNCF is suitable to study
the dynamics of mechanical systems with multiple scales, such as a large-scale travel and a
small-scale vehicle size, a high-frequency rotation, and a long time of operation.

Acknowledgement This work is supported by the National Natural Science Foundations of China (NSFC)
under grant number 11772101, the Shanghai Academy of Spaceflight Technology under grant number
SAST 2017-025, and the National Key Research and Development Program of China under Grant number
2018YFF0300506.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Appendix A: Derivations of Eqs. (6) and (12)

The angular velocity ω(t) and infinitesimal rotation δπ in the body coordinate system for a
rotation motion A(t) are calculated by


ω = AT Ȧ, δπ = AT δA.

The variation of an arbitrary function c (A) can be calculated by

∂c ∂c     ∂c
3 3 3 3
δc = δA = Aδπ = Aac bcd δπd ,
∂A ∂A a=1 b=1 c=1 d=1
∂Aab
336 H. Ren, K. Yang

3 3
where abc is the alternative tensor such that (x × y)a = b=1 c=1 abc xb yc , and hence

∂c  3 3
∂c
3
= bcd Aac ,
∂πd a=1 b=1 c=1
∂Aab

which is equivalent to
 T 3
∂c ∂c
= × (Aa: )T . (28)
∂π a=1
∂A a:

Now consider a special case that c = a T AT (t)c, where a and c are constant vectors. Then
direct calculations yield
 T
∂c ∂ 2c  
=
a AT c, a
= AT c . (29)
∂π ∂π ∂π T

Moreover, suppose the rotation motion is described by a three-component parameter ϑ , and

the angular velocity and the infinitesimal rotation can be written as [11]

ω = D (ϑ) ϑ̇, δπ = D (ϑ) δϑ, (30)

respectively, and by Eqs. (4) and (30),

 T  T
∂K ∂ω ∂K ∂K ∂K ∂ω ∂K ∂K
T
= = DT , = + DT .
∂ ϑ̇ ∂ ϑ̇ ∂ωT ∂ωT ∂ϑ T ∂ϑ ∂ωT ∂π T

Then direct calculation yields

    T 
d ∂K ∂K d ∂K ∂ω
T ∂K ∂K
− = DT + Ḋ − − DT . (31)
dt ∂ ϑ̇ T ∂ϑ T dt ∂ωT ∂ϑ ∂ωT ∂π T

In Appendix C, the following formula will be proved:

 T
T ∂ω
Ḋ − = DT 
ω. (32)
∂ϑ

Substituting Eq. (32) into Eq. (31) yields

 
d ∂K ∂K d ∂K ∂K ∂K
− = D T
+ ω × − . (33)
dt ∂ ϑ̇ T ∂ϑ T dt ∂ωT ∂ωT ∂π T

By Eqs. (3), (4), and (29),

∂K     ∂K    
= d α q α × AT ṙ + M αβ q α × q̇ β + ω × q β , = d α q̇ α + ω × q α × AT ṙ ,
∂ωT ∂π T
which can be substituted into Eq. (33), and Eq. (6) can be derived by direct calculations.
Moreover, suppose the virtual work done by the applied force can be written as

δ W = δπ · t + · · · = δϑ T D T t + · · · ,
A referenced nodal coordinate formulation 337

and the Lagrange equations with constraints can be written as

 
d ∂K ∂K ∂C T
− + λ = D T t,
dt ∂ ϑ̇ T ∂ϑ T ∂ϑ

where C = 0 indicates the holonomic constraint equations. Direct calculations yield that

∂C ∂C ∂π ∂C
= = D.
∂ϑ ∂π ∂ϑ ∂π
The Lagrange equations with constraints for the three-component parameter ϑ are
    
d ∂K ∂K ∂C T d ∂K ∂K ∂K ∂C T
− + λ=D T
+ω× − + λ = D T t.
dt ∂ ϑ̇ T ∂ϑ T ∂ϑ dt ∂ωT ∂ωT ∂π T ∂π

The choice of the three-component parameter ϑ is arbitrary, and in any specific configura-
tion, it is always possible to find a proper ϑ such that D is invertible. Hence the rotation
equations for constrained flexible bodies can be written as
 
d ∂K ∂K ∂K ∂C T
+ω× − + λ = t, (34)
dt ∂ωT ∂ωT ∂π T ∂π

where the partial derivatives with respect to the infinitesimal rotation δπ can be evaluated
by Eq. (28) or Eq. (29). Equation (12) can be derived by substitution of Eqs. (6) and (10)
into Eq. (34).

Appendix B: A simplified ANCF/RNCF element for space tethers

Consider the kth element of the tether with a uniform isotropic material and a uniform cross
section, whose undeformed configuration is a straight line, with the arclength coordinate
xk−1 ≤ x ≤ xk . The generalized coordinates of the element are

q 1 (t) = p (xk−1 , t) , q 2 (t) = px (xk−1 , t) , q 3 (t) = p (xk , t) , q 4 (t) = px (xk , t) ,

and Hermite polynomials are adopted to interpolate the position vector, so the corresponding
shape functions are

s 1 (x) = 1 − 3ξ 2 + 2ξ 3 , s 2 (x) = lk (ξ − 2ξ 2 + ξ 3 ),
s 3 (x) = 3ξ 2 − 2ξ 3 , s 4 (x) = lk (ξ 3 − ξ 2 ),

where lk = xk − xk−1 and ξ = (x − xk−1 ) /lk . The kinetic energy of the element in the RCS
can be written as
1 1 ij
Kk = mk ṙ · ṙ + dki ṙ · (Av i ) + Mk v i · v j , (35)
2 2
where mk is the mass of the element, and
ˆxk ˆxk
ij
v i = q̇ i + ω × q i , dki = i
ρs (x) dx, Mk = ρs i (x) s j (x) dx. (36)
xk−1 xk−1
338 H. Ren, K. Yang

The reference configuration can be described by

q̊ 1 = (xk−1 , 0, 0)T , q̊ 2 = (1, 0, 0)T , q̊ 3 = (xk , 0, 0)T , q̊ 4 = (1, 0, 0)T ,

! !
and it is easy to check that !p̊x (x)! = 1. The axial strain of the cable can be written as

1
ε (x, t) = (r x (x, t) · r x (x, t) − 1) .
2
where r x (x, t) = A (t) px (x, t) by Eq. (1). The unit tangential vector in the element is

r x (x, t)
i (x, t) = ,
r x (x, t)

and the curvature of the tether can be evaluated by

∂i (x, t)/∂x r x (x, t) × r xx (x, t)

κ (x, t) = = ,
r x (x, t) r x (x, t) 3

The stretch in the tether is expected to be small in practice, such that r x (x, t) ≈ 1, and

(r x · r x ) (r xx · r xx ) − (r x · r xx )2
i ≈ rx, κ ·κ = ≈ (px · px ) (pxx · pxx ) − (px · pxx )2 .
rx 6

Then the elastic potential energy of the element can be calculated by

ˆxk ˆxk
1 1
Uke = EAε 2 (x, t) dx + EI κ (x, t) 2
dx
2 2
xk−1 xk−1

1 ijmn 1 ijmn
= κ Qij (t)Qmn (t) + υk Pij (t)Pmn (t)
2 k 2
for i, j, m, n = 1, 2, 3, 4,

ˆxk
ijmn 1 
κk = EAsxi (x) sxj (x) sxm (x) sxn (x) dx, Qij (t) = q i (t) · q j (t) − q̊ i · q̊ j ,
2
xk−1

1
Pij (t) = q i (t) · q j (t),
2
and
ˆxk
ijmn   
υk = EI 2 sxi sxj sxx
m n
sxx + sxx
i j m n
sxx sx sx
xk−1
 i j m n 
− sxx sx sxx sx + sxi sxx sxx sx + sxx
j m n
sx sx sxx + sxi sxx
i j m n j m n
sx sxx dx.

Then the elastic force is

ijmn ijmn
F ike = κk Qmn + υk Pmn q j . (37)
A referenced nodal coordinate formulation 339

The virtual work done by the thrust force and the gravitational force on the element is

ˆxk "    #
r⊕ r r⊕2 r
δ Wk = δr · c max (0, V0 − V1 ) i ×
p
× i − ρg dx
r 2 r 3
xk−1

ˆxk "    #
r⊕ r r⊕2 r
≈ δr p · c max (0, V0 − V1 ) r x × × r x − ρg dx
r 2 r 3
xk−1
      
= δr · F kt + F kg + δπ · q i × Qikt + Qikg + δq i · Qikt + Qikg ,

where the generalized forces are

 
ij   r⊕ r   r2 r
F kt = ζk Aq i × 2
× Aq j , F kg = −mk g ⊕ 3 ,
r r
  (38)
r⊕ AT r r 2 AT r
Qikg = −gdki ⊕ 3 ,
ijm
Qikt = ιk q j × 2
× qm ,
r r
ij ij ijm ijm
in which ζk = c max (0, V0 − V1 ) ζ̂k and ιk = c max (0, V0 − V1 ) ι̂k with

ˆxk ˆxk
ij ijm
ζ̂k = sxi (x) sxj (x) dx, ι̂k = s i (x) sxj (x) sxm (x) dx.
xk−1 xk−1

ij ij
The constants dki , Mk , and ζ̂k can be analytically evaluated to get the following matrices

⎛ ⎞ ⎛ ⎞ ⎛ 6 1 ⎞
1 13
35
11lk
210
9
70
− 13l
420
k 1
− 5l6k 10
5lk 10
⎜ lk
2
⎟ ⎜ 11lk lk2 lk2 ⎟ ⎜ 1 2lk − 1 − lk ⎟
⎜ 210
13lk
− 140 ⎟
dk = mk ⎜
⎝ 1
12 ⎟ , Mk =
⎠ mk ⎜ 9 105 420 ⎟ , ζ̂ k = ⎜ 10 15 10
⎝− 6 − 1 6 − 1 ⎠
30 ⎟
2 ⎝ 70 13lk
420
13
35
− 11l
210
k
⎠ 5lk 10 5lk 10
− 12
lk
− 13l k
− 140
lk2
− 11l k lk2 1
10
− 30
lk
− 10
1 2lk
15
420 210 105

ij m
and the constant tensor ι̂k can be evaluated to get the following ι̂ik matrices with

⎛ ⎞ ⎛ ⎞
3
− 70
1
− 5l3k 4 9
70
lk
140
− 70
9 3lk
140
5lk 35
⎜− 1 ⎟ ⎜ lk lk2 lk2 ⎟
⎜ 70
43lk 1
− 60
lk
⎟, ⎜ 140 − 140 − 840
lk
⎟
ι̂1k = ⎝− 3
420 70
⎠ ι̂2k = ⎜ 9 120 ⎟
5lk
1
70
3
5lk
− 354
⎝ − 70 − 140lk 9
70
− 140
3lk
⎠
lk2 lk2
4
35
− 60 − 35 420
lk 4 13lk 3lk
− 840 − 140 168
3lk
140
⎛ 1 ⎞
⎛ ⎞
3 4
− 5l3k − 70 − 70
9
− 140
3lk
70
9
− 140
lk
5lk 35
⎜ 4 lk ⎟ ⎜ 3lk −lk2 lk2 ⎟
⎜ 353
13lk
− 35
4
− 60 ⎟ ⎜−
3lk
⎟
ι̂3k = ⎝−
420
1 ⎠, ι̂4n = ⎜ 9140 168 140 840 ⎟
5lk
− 354 3
5lk 70 ⎝ 70 3lk
140
− 70 9 lk
140 ⎠
lk2 −lk2
− 701
− 60
lk 1
70
43lk
420 − 140
lk lk
840 140 120

ijmn ijmn
Moreover, the stiffness tensors κk and υk can be evaluated analytically too, but since
they are complicated, it is better to calculate them numerically using Gauss quadratures.
340 H. Ren, K. Yang

Appendix C: Proof of formula (32)

Suppose the rotational matrix A = (x, y, z) is described by a three-component-parameter

ω = AT Ȧ = (x, y, z)T (ẋ, ẏ, ż).
vector ϑ . The angular velocity ω can be calculated from 
Hence the angular velocity can be explicitly written as
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
z · ẏ 3 z · ya z · y1 z · y2 z · y3
ω = ⎝ x · ż ⎠ = ⎝ x · za ⎠ ϑ̇a = D (ϑ) ϑ̇, D = ⎝ x · z1 x · z2 x · z3 ⎠ , (39)
y · ẋ a=1 y · xa y · x1 y · x2 y · x3

where x a ≡ ∂x/∂ϑa , y a ≡ ∂y/∂ϑa , and za ≡ ∂z/∂ϑa , and D is the same as that in Eq. (30).
So the time derivative of D is
⎛ ⎞ ⎛ ⎞
 3 za · y 1 za · y 2 za · y 3 3 z · y 1a z · y 2a z · y 3a
Ḋ = ⎝ x a · z1 x a · z2 x a · z3 ⎠ ϑ̇a + ⎝ x · z1a x · z2a x · z3a ⎠ ϑ̇a , (40)
a=1 ya · x1 ya · x2 ya · x3 a=1 y · x 1a y · x 2a y · x 3a

and the partial derivative ∂ω/∂ϑ can be directly calculated by

⎛ ⎞ ⎛ ⎞
z ·y z2 · y a z3 · y a z · y 1a z · y 2a z · y 3a
∂ω  ⎝ 1 a 
3 3
= x 1 · za x 2 · za x 3 · za ⎠ ϑ̇a + ⎝ x · z1a x · z2a x · z3a ⎠ ϑ̇a . (41)
∂ϑ a=1 y · x y2 · xa y3 · xa a=1 y · x 1a y · x 2a y · x 3a
1 a

Then by Eqs. (40) and (41) the left-hand side of Eq. (32) becomes
⎛ ⎞
 T 
3 za · y 1 − z1 · y a , x a · z1 − x 1 · za , ya · x1 − y1 · xa
∂ω ⎝ za · y 2 − z2 · y a , y a · x 2 − y 2 · x a ⎠ ϑ̇a . (42)
T
Ḋ − = x a · z2 − x 2 · za ,
∂ϑ a=1 za · y 3 − z3 · y a , x a · z3 − x 3 · za , ya · x3 − y3 · xa

Since x, y, z are unit vectors, x · x a = y · y a = z · za = 0, and their partial derivatives can

be expressed in the basis {x, y, z} as
   
x a = (y · x a ) y + (z · x a ) z, y a = z · y a z + x · y a x, za = (x · za ) x + (y · za ) y.

Hence their dot products can be calculated by

   
x a · y b = (z · x a ) z · y b , y a · zb = x · y a (x · zb ) , za · x b = (y · za ) (y · x b ) ,

which can be substituted into Eq. (42) to yield that

 T
T ∂ω
Ḋ −
∂ϑ
⎛               ⎞

3 (x · za ) x · y 1 − x · z1 x · y a , (y · x a ) y · z1 − y · x 1 (y · za ) , z · y a z · x 1 − z · y 1 (z · x a )
             
= ⎝ (x · za ) x · y 2 − x · z2 x · y a , (y · x a ) y · z2 − y · x 2 (y · za ) , z · y a z · x 2 − z · y 2 (z · x a ) ⎠ϑ̇a .
             
a=1 (x · za ) x · y 3 − x · z3 x · y a , (y · x a ) y · z3 − y · x 3 (y · za ) , z · y a z · x 3 − z · y 3 (z · x a )

(43)

On the other hand, the right-hand side of Eq. (32) can be directly calculated by Eq. (39),
A referenced nodal coordinate formulation 341

which yields

DT 
ω
⎛             ⎞

3 x · z1 (y · x a ) − y · x 1 (x · za ) , y · x 1 z · y a − z · y 1 (y · x a ) , z · y 1 (x · za ) − x · z1 z · y a
             
= ⎝ x · z2 (y · x a ) − y · x 2 (x · za ) , y · x 2 z · y a − z · y 2 (y · x a ) , z · y 2 (x · za ) − x · z2 z · y a ⎠ϑ̇a .
             
a=1 x · z3 (y · x a ) − y · x 3 (x · za ) , y · x 3 z · y a − z · y 3 (y · x a ) , z · y 3 (x · za ) − x · z3 z · y a

(44)

Since the vectors x, y, z are perpendicular to each other, that is, x · y = y · z = z · x = 0,

their time derivatives indicate that

y · x a = −x · y a , z · y a = −y · za , x · za = −z · x a ,

which can be substituted to Eq. (44), yielding that

DT 
ω
⎛              ⎞

3 x · y 1 (x · za ) − x · z1 x · y a , y · z1 (y · x a ) − y · x 1 (y · za ) , z · x 1 z · y a − z · y 1 (z · x a )
             
= ⎝ x · y 2 (x · za ) − x · z2 x · y a , y · z2 (y · x a ) − y · x 2 (y · za ) , z · x 2 z · y a − z · y 2 (z · x a ) ⎠ϑ̇a .
             
a=1 x · y 3 (x · za ) − x · z3 x · y a , y · z3 (y · x a ) − y · x 3 (y · za ) , z · x 3 z · y a − z · y 3 (z · x a )

(45)

Equalities (43) and (45) are obvious, which can be checked term by term. So Eq. (32) is
proved. 

Moreover, since the three-component parameter ϑ can be arbitrarily chosen, the proof is
independent of any specific ϑ .

References
1. Simo, J.C.: A finite strain beam formulation. The three-dimensional dynamic problem. Part I. Comput.
Methods Appl. Mech. Eng. 49(1), 55–70 (1985). https://doi.org/10.1016/0045-7825(85)90050-7
2. Simo, J.C., Vu-Quoc, L.: Three dimensional finite strain rod model. Part II: computational aspects. Com-
put. Methods Appl. Mech. Eng. 58, 79–116 (1986). https://doi.org/10.1016/0045-7825(86)90079-4
3. Shabana, A.A.: Computational Continuum Mechanics, pp. 270–320. Cambridge University Press, Cam-
bridge (2008)
4. Bauchau, O.A., Han, S., Mikkola, A., Matikainen, M.K.: Comparison of the absolute nodal coor-
dinate and geometrically exact formulations for beams. Multibody Syst. Dyn. 32, 67–85 (2014).
https://doi.org/10.1007/s11044-013-9374-7
5. Ren, H.: A simple absolute nodal coordinate formulation for thin beams with large deformations and
large rotations. J. Comput. Nonlinear Dyn. 10(6), 061005 (2015). https://doi.org/10.1115/1.4028610
6. Ren, H.: Fast and robust full-quadrature triangular elements for thin plates/shells with large deformations
and large rotations. J. Comput. Nonlinear Dyn. 10(5), 051018 (2015). https://doi.org/10.1115/1.4030212
7. Crisfield, M., Galvanetto, U., Jelenić, G.: Dynamics of 3-D co-rotational beams. Comput. Mech. 20,
507–519 (1997). https://doi.org/10.1007/s004660050271
8. Hsiao, K.M., Lin, J.Y., Lin, W.Y.: A consistent co-rotational finite element formulation for geometrically
nonlinear dynamic analysis of 3-D beams. Comput. Methods Appl. Mech. Eng. 169(1–2), 1–18 (1999).
https://doi.org/10.1016/S0045-7825(98)00152-2
9. Jonker, J.B., Meijaard, J.P.: A geometrically nonlinear formulation of a three-dimensional beam element
for solving large deflection multibody system problems. Int. J. Non-Linear Mech. 53, 63–74 (2013).
https://doi.org/10.1016/j.ijnonlinmec.2013.01.012
10. Jelenić, G., Crisfield, M.: Interpolation of rotational variables in nonlinear dynamics of 3D
beams. Int. J. Numer. Methods Eng. 43(7), 1193–1222 (1998). https://doi.org/10.1002/(SICI)1097-
0207(19981215)43:73.0.CO;2-P
342 H. Ren, K. Yang

11. Shabana, A.A.: Dynamics of Multibody Systems 3rd edn. pp. 28–158. Cambridge University Press,
Cambridge (2005)
12. Recuero, A.M., Escalona, J.L.: Application of the trajectory coordinate system and the moving modes
method approach to railroad dynamics using Krylov subspaces. J. Sound Vib. 332(20), 5177–5191
(2013). https://doi.org/10.1016/j.jsv.2013.04.015
13. Sugiyama, H., Shabana, A.A.: Trajectory coordinate constraints in multibody railroad vehicle systems.
J. Syst. Des. Dyn. 1(3), 481–490 (2007). https://doi.org/10.1299/jsdd.1.481
14. Shabana, A.A.: Geometrically accurate floating frame of reference finite elements for the small defor-
mation problem. Proc. Inst. Mech. Eng., Proc., Part K, J. Multi-Body Dyn. 232, 286–292 (2018)
15. Shabana, A.A., Chamorro, R., Rathod, C.: A multi-body system approach for finite-element modelling
of rail flexibility in railroad vehicle applications. Proc. Inst. Mech. Eng., Proc., Part K, J. Multi-Body
Dyn. 222(1), 1–15 (2008). https://doi.org/10.1243/14644193JMBD117
16. Sugiyama, H., Suda, Y.: A curved beam element in the analysis of flexible multi-body systems using
the absolute nodal coordinates. Proc. Inst. Mech. Eng., Proc., Part K, J. Multi-Body Dyn. 221, 219–231
(2007)
17. Gerstmayr, J., Shabana, A.A.: Analysis of thin beams and cables using the absolute nodal coordinate
formulation. Nonlinear Dyn. 45, 109–130 (2006)
18. Negrut, D., Rampalli, R., Ottarsson, G., Sajdak, A.: On an implementation of the Hilber–Hughes–Taylor
method in the context of index 3 differential-algebraic equations of multibody dynamics. J. Comput.
Nonlinear Dyn. 2(1), 73–85 (2007). https://doi.org/10.1115/1.2389231
19. Brüls, O., Cardona, A., Arnold, M.: Lie group generalized–alpha time integration of constrained
flexible multibody systems. Mech. Mach. Theory 48(1), 121–137 (2012). https://doi.org/10.1016/
j.mechmachtheory.2011.07.017
20. Janhunen, P.: Status report of the electric sail in 2009. Acta Astronaut. 68(5–6), 567–570 (2009).
https://doi.org/10.1016/j.actaastro.2010.02.007
21. Janhunen, P.: Increased electric sail thrust through removal of trapped shielding electrons by or-
bit chaotisation due to spacecraft body. Ann. Geophys. 27, 3089–3100 (2009). https://doi.org/
10.5194/angeo-27-3089-2009
22. Janhunen, Pe., Toivanen, P., Polkko, J., Merikallio, S., Salminen, P., Haeggström, E., Seppänen, H.,
Kurppa, R., Ukkonen, J., Kiprich, S., Thornell, G., Kratz, H., Richter, L., Krömer, O., Rosta, R., Noorma,
M., Envall, J., Lätt, S., Mengali, G., Obraztsov, A.: Invited article: electric solar wind sail: toward test
missions. Rev. Sci. Instrum. 81, 111301 (2010). https://doi.org/10.1063/1.3514548
23. Toivanen, P., Janhunen, P.: Electric solar wind sail: deployment, long-term dynamics, and control
hardware requirements. In: MacDonald, M. (ed.) Advances in Solar Sailing. Springer, Berlin (2014).
https://doi.org/10.1007/978-3-642-34907-2_58
24. Toivanen, P.K., Janhunen, P.: Spin plane control and thrust vectoring of electric solar wind sail. J. Propuls.
Power 29(1), 178–185 (2013). https://doi.org/10.2514/1.B34330
25. Janhunen, P.: Photonic spin control for solar wind electric sail. Acta Astronaut. 83, 85–90 (2013).
https://doi.org/10.1016/j.actaastro.2012.10.017
26. Hong, D., Ren, G.: A modeling of sliding joint on one-dimensional flexible medium. Multibody Syst.
Dyn. 29, 91–105 (2011). https://doi.org/10.1007/s11044-010-9242-7
27. Seppänen, H., Kiprich, S., Kurppa, S., Janhunen, P., Hæggström, E.: Wire-to-wire bonding of µm-
diameter aluminum wires for the electric solar wind sail. Microelectron. Eng. 83(11), 3267–3269 (2011).
https://doi.org/10.1016/j.mee.2011.07.002