Modeling Steady-State and Transient Forces on a Cylinder∗

Osama Marzouk1 , Ali H. Nayfeh1†, Haider N. Arafat2 , Imran Akhtar1

1
Department of Engineering Science and Mechanics, MC 0219
2
Department of Electrical and Computer Engineering, MC 0111
Virginia Tech, Blacksburg, VA 24061
{omarzouk, anayfeh, harafat, akhtar}@vt.edu

April 26, 2006

Abstract
Numerical simulations of flow past a stationary circular cylinder at different Reynolds numbers have
been performed using a computational fluid dynamics (CFD) solver that is based on the Reynolds-
averaged Navier-Stokes equations (RANS). The results obtained are used to develop reduced-order mod-
els for the lift and drag coefficients. The models do not only match the numerical simulation results in
the time domain, but also in the spectral domain. They capture the steady-state region with excellent
accuracy. Further, the models are verified by comparing their results in the transient region with their
counterparts from the CFD simulations and a very good agreement is found. The work performed here
is a step towards building models for vortex-induced vibrations (VIV) encountered in risers, spars, and
other offshore structures.
Keywords: Vortex-induced vibrations, reduced-order modeling, circular cylinders, risers, transient re-
sponse.

1 Introduction
1.1 Literature Summary
When flow passes over a bluff body, a street of shedding vortices (known as the von Kármán vortex street)
develops in the wake and exerts oscillatory forces on the body, which are often decomposed into drag and
lift components along the free-stream and the cross-flow directions, respectively. If the body is capable of
flexing or moving rigidly, these forces can cause it to oscillate and the classical vortex-induced vibration
(VIV) problem takes place. If the frequency of vortex shedding is close to a natural frequency of the body,
the resulting resonance can generate large-amplitude oscillations, which may ultimately cause structural
failure. Understanding this problem is of great interest in the design and maintenance of offshore structures,
like mooring cables, rises, and spars, and of high aspect-ratio structures subject to air streams, like chimneys,
high-rise buildings, bridges, and cable suspensions systems.
In order to model and predict the VIV problem, full simulation of the coupled flow and structure equations
is needed. This involves accounting for three-dimensional effects, turbulence structures, and elasticity, among
∗ Submitted for publication to the Journal of Vibration and Control
† Corresponding Author

1
other considerations. However, in general, many of these structures are quite long, some exceeding 7, 000
feet, such as risers and spars, and flows around them may reach very high Reynolds numbers (Re). Under
these conditions, full simulation of the VIV problem is overwhelming, even with todays computing power.
So, simplifying assumptions about the flow and structure are often made and efforts to produce simple,
accurate representations of the problem via reduced-order modeling are still on-going.
Circular cylinders are extensively used in the study of bluff-body fluid dynamics due to their geometric
simplicity and common use in engineering applications. We give a brief introduction of the studies that have
dealt with VIV of fixed cylinders, cylinders with prescribed motions or forces, elastically mounted cylinders,
and round cross-section elastic structures. A thorough review on the subject of VIV is given by Williamson
and Govardhan (2004).
Bishop and Hassan (1963) were one of the earliest to suggest using a self-excited oscillator to represent
the forces over a cylinder due to vortex shedding. Hartlen and Currie (1970) formulated a remarkable model
for elastically restrained cylinders that are restricted to cross-flow motions. They used a Rayleigh oscillator
to describe the lift force and coupled it to the cylinder motion by a linear velocity term. Currie and Turnball
(1987) proposed a similar model for the fluctuations in the drag. Skop and Griffin (1973) pointed out that the
parameters in the model of Hartlen and Currie (1970) lacked clear connections to physical parameters of the
problem. They proposed a modified van der Pol oscillator to represent the lift coupled to a linear equation
of motion for the structure. They also introduced the Skop-Griffin parameter, which is now commonly used
in studying VIV problems.
Iwan and Blevins (1974) considered the fluid mechanics of the vortex street and developed a model in
terms of a fluid variable that captures the fluid dynamics effects of the problem. Landl (1975) added a
nonlinear aerodynamic damping term of fifth order to the van der Pol oscillator in his two-equation model,
suggesting that this enables better capturing of some physical characteristics. However, the model involved
many constants to be determined.
Several attempts have been made to extend the wake-oscillator models to elastic structures, such as
beams and cables. Iwan (1975) extended the model of Iwan and Blevins (1974) to predict the maximum
VIV amplitude of taut strings and circular cylindrical beams with different end conditions. Iwan (1981) also
derived an analytical model for the VIV of elastic structures under non-uniform flows.
Skop and Griffin (1975) extended their (1973) rigid cylinder model to elastic cylinders. With the goal
of accurately capturing the asymptotic, self-limiting structural response near zero damping, Skop and Bal-
asubramanian (1997) modified the model of Skop and Griffin (1975) by separating the fluctuating lift into
two components: one satisfying a van der Pol equation oscillator that is driven by the transverse motion of
the cylinder and the other linearly proportional to the transverse velocity of the cylinder (called the stall
term). Kim and Perkins (2002) modeled the lift and drag by two nonlinearly coupled van der Pol equations
in their study of resonant responses of suspended elastic cables. The coupling terms were introduced based
on the fact that the main frequency of the drag component is twice the main frequency of the lift component.
Krenk and Nielsen (1999) proposed a two-oscillator model in which the mutual forcing terms are developed
based on the premise that energy flows directly between the fluid and structure. This means that the forcing
terms correspond to the same flow of energy at all times.

1.2 Motivation
Most of the aforementioned analytical works are based on modeling the lift by either the Rayleigh or the
van der Pol oscillator. Nayfeh et al. (2003) numerically simulated the two-dimensional flow past a stationary
cylinder for a wide range of Reynolds numbers and calculated the lift and drag. They employed higher-

2
order spectral analyses to determine the phase relations among the different spectral components of the
lift and compared them with the phase information one obtains from closed-form approximate solutions
of the van der Pol and Rayleigh oscillators. They concluded that the van der Pol oscillator is the more
accurate representation of the lift. They also proposed to model the drag as the sum of a mean term and a
time-varying term proportional to the product of the lift and its time derivative.

1.3 Background of Work Presented

In all of the previous models, the emphasis has been on reproducing the steady-state lift and drag. However,
to investigate this fluid-structure interaction problem more thoroughly, modeling the transient as well as the
steady-state lift and drag is crucial. The goals of this work are two-fold:

• First, we extend the results of Nayfeh et al. (2003) on the flow over a stationary cylinder by investigating
the validity of their models in the transient region. In other words, are the models identified based
on the steady-state lift and drag CFD results capable of predicting the transient lift and drag results
obtained with the CFD code?

• Second, we propose modifications and improvements to both of the lift and drag models that take
full and explicit advantage of the phase relations between the different components. These improved
models are validated against the CFD results for both the steady-state and transient behaviors.

The order of this paper is as follows:

• First, we discuss the CFD code used to generate the numerical solutions of lift and drag on a stationary
cylinder.

• Second, we present some of the lift and drag results of Nayfeh et al. (2003) for steady-state responses
and investigate the applicability of their models within the transient response.

• Third, we discuss the improved lift and drag models and compare their performances to the CFD
results in the steady-state as well as transient regions.

• Fourth, we conclude with some remarks and general notes.

2 Numerical Simulation
Direct numerical simulation (DNS) of flows with engineering relevance remains a challenging task, even with
current advances in hardware and software capabilities. Hence, some modeling assumptions or approxi-
mations are often needed to simplify the flow and render their computations feasible. Usually, this entails
representing the turbulent scales with turbulence models, which significantly reduce the spatial and temporal
resolution requirements. One alternative to DNS is the large-eddy simulation (LES) approach. In LES, the
energy-containing turbulent scales are resolved, whereas the subgrid scales (SGS) are modeled. LES provides
information about a wide range of spatial and temporal scales in the flow at a cost that is significantly lower
than DNS. Nonetheless, for high Reynolds-number flows over complex geometries, LES computations still
remain to be a formidable task.
In the conventional Reynolds-averaged Navier-Stokes (RANS) modeling, the effects of all of the turbulent
fluctuations are modeled. Extensively used within the fluid engineering community, the RANS approach and
its variants are efficient and much cheaper than DNS and LES in computation cost, yet it yields results that
are accurate enough to capture most of the important physical characteristics of the problem.

3
 Therefore, we compute the time-dependent, incompressible flow past a cylinder by solving the Reynolds-
averaged Navier-Stokes (RANS) equations. We employ the artificial compressibility method to speed up
the solution convergence. This method introduces an artificial time-derivative of pressure term into the
continuity equation, thereby explicitly coupling the pressure and the velocity field (Rogers et al., 1987;
Chorin, 1997).
The nondimensional equations governing the two-dimensional flow have the form
∂q ∂F ∂G 1
+ + − H∇2q = 0 (1)
∂t ∂x ∂y Re
where Re ≡ U D/ν is the Reynolds number and
       

 p 
 
 βu   
 βv 
 0 0 0
2  
q= u , F = u +p ,G = uv ,H =  0 1 0  (2)

 
 
 
 
 2 

v uv v +p 0 0 1

Here, U is the free-stream velocity, D is the cylinder diameter, ν is the kinematic viscosity, p is the pressure,
u and v are the velocity components in the horizontal (streamwise) and vertical (cross-stream) directions,
respectively, and β is an artificial compressibility parameter. The original continuity equation corresponds
to having β −→ ∞. The problem is solved on a curvilinear O-grid with a 25D radius in order to avoid any
blocking effects without using buffer domains. The convective terms are discretized using a second-order
upwinding difference scheme. The physical time terms, which represent flow unsteadiness, are switched to
the right-hand side and used as source terms; they are discretized using a second-order three-level backward-
difference formula.
To integrate the RANS equations with inflow, outflow, and no-slip boundary conditions, radial and
circumferential stretching is employed and a second-order finite-difference scheme is used. At the inflow
boundary, the velocity components are specified and the pressure is extrapolated from the interior points.
At the outflow boundary, the pressure is specified and the velocity components are extrapolated from the
interior points. For turbulent closure, a one-equation model, such as the Spalart-Allamras (SA) method
(Spalart and Allamras, 1992), is used to represent the unresolved scales. Equivalent boundary conditions
are also set for the turbulence quantities.
The outcome of the simulation is the pressure distribution over the cylinder surface, which ultimately is
integrated to calculate the lift and drag coefficients. The time histories of the lift and drag coefficients are
then converted to the spectral domain using Fourier analysis. The time and frequency are nondimensionalized
using the free-stream velocity and the cylinder diameter. The nondimensional frequency (Strouhal number,
St) is defined as St ≡ fU/D, where f is the dimensional cyclic frequency in Hz.
Figure 1 shows typical time histories and corresponding power spectra for the case when Re = 4, 000.
The lift coefficient CL is presented in the top two subfigures and the drag coefficient CD is presented in
the bottom two subfigures. The lift power spectrum shows a large peak with amplitude a1 at the shedding
frequency fs ≈ 0.2 and smaller peaks (a3 and a5) at the odd harmonics 3fs and 5fs . The corresponding time
history shows the lift coefficient fluctuating periodically about the origin. Therefore, one infers that the lift
coefficient oscillates at the shedding frequency and that its behavior is influenced by cubic, and, to lower
extent, higher-order odd nonlinearities.
By contrast, the drag power spectrum shows a large peak a2 at twice the shedding frequency (2fs ) and
smaller peaks (a4 and a6 ) at the even harmonics 4fs and 6fs . The corresponding time history shows the drag
fluctuating periodically about a non-zero mean that reaches a constant value at steady state. This implies
that the drag coefficient consists of a time-varying mean term and a fluctuating term. The fluctuating term
mainly varies quadratically with the lift coefficient, as the influence of the higher-order even nonlinearities is

4
quite small. The corresponding drag-polar plot in Figure 2 clearly illustrates this quadratic coupling between
the drag and lift coefficients.

3 Modeling Background
3.1 Lift
Nayfeh et al. (2003) investigated two wake-oscillator models to represent the lift, namely, the van der Pol
and Rayleigh oscillators. Using higher-order spectral moments analysis, they calculated, for a few cases, the
phase angle φ13 between the lift components at fs and 3fs to fall around 90◦ . Based on their findings, they
concluded that the van der Pol oscillator

C̈L + ω2 CL = µĊL − αCL2 ĊL (3)

is the more suitable choice as an efficient and simple model for the lift coefficient in steady state. The
angular frequency ω in equation (3) is related (but not equal) to the angular shedding frequency ωs = 2πfs
and the parameters µ and α represent the linear and nonlinear damping coefficients, respectively. The values
of µ and α are taken positive, so that the linear damping is destabilizing while the nonlinear damping is
stabilizing. As a consequence, small disturbances grow and large ones decay, both eventually approaching a
stable limit cycle. The values of the parameters in equation (3) depend on the Reynolds number and their
values are based on matching the steady-state CFD lift data.
Recently, the authors (Nayfeh et al., 2005) examined the possibility of using equation (3) to model the
lift coefficient in the transient region as well. Using the method of multiple scales (Nayfeh, 1973, 1981)
and assuming that the oscillator is weakly damped (i.e., µ = O() and α = O() where   1 is a small
bookkeeping parameter), the following second-order approximate solution was obtained:
s
 2
1 1 2
CL(t) = a(t) 1 + µ − αa(t) sin[ωt + θ(t) + η(t)]
16ω2 4

1.6 1.6
0.8 1.4
L
C 0 CD
-0.8 1.2
-1.6 1
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
Time Time

1x10 1
1x10 1

<CD>ss
a1

1x10 0
1x10 0

a 2

1x10 -1
a 1x10 -1

L D
C C
3

a 4
1x10 -2
a 1x10 -2

a
5

1x10 -3
1x10 -3

1x10 -4
1x10 -4

fs
0 0.3 0.6 0.9 1.2 1.5 0 0.3 2fs 0.6 0.9 1.2 1.5
Strouhal Number (St) Strouhal Number (St)

Figure 1: Time histories and corresponding power spectra of the lift and drag coefficients obtained from the
CFD simulation at Re = 4, 000.

5
 α
− a(t)3 sin[3ωt + 3θ(t)] + · · ·
32ω
≡ a1 (t) sin[ωt + θ(t) + η(t)] − a3(t) sin[3ωt + 3θ(t)] + · · · (4)
h i
16ω
where η(t) = tan−1 αa(t)2 −4µ and the amplitude a(t) and phase θ(t) are governed by the modulation
equations
da 1
= (4µa − αa3) (5)
dt 8  
dθ 1 2 3 2 11 2 4
=− µ − αµa + α a (6)
dt 8ω 2 32

Setting ȧ = 0 in Eqn. (5), the steady-state value of a is obtained from the solution of a(4µ − αa2) = 0,
p
which is either the trivial solution a = 0 or the nontrivial solution a = 2 µ/α. For the nontrivial solution,
it follows from equation (4) that
r r
µ µ µ
a1 = 2 and a3 = (7)
α 4ω α

Moreover, the angle η = π2 and, from equation (6), the corresponding expression for θ̇ = −µ2 /16. Conse-
quently, the angular shedding frequency is given by

µ2
ωs = ω + θ̇ = ω − (8)
16ω
Equation (8) shows that the angular frequency ω of the van der Pol oscillator is not exactly equal to the
angular shedding frequency ωs , as one would predict from a first-order expansion (Nayfeh et al., 2003).
Hence, an improved second-order approximate expression for the steady-state lift coefficient becomes

CL (t) ≈ a1 cos(ωs t) + a3 cos(3ωs t + π2 ) (9)

The methodology used to identify the system parameters for a given Reynolds number is as follows:

1. The CFD solver is used to calculate the time history of the lift coefficient.

2. Spectral analysis is performed on the steady-state part of the CFD data to extract the values of a1,
a3, and fs (or ωs ).

1.5
1
0.5
L
0
C

-0.5
-1
-1.5
1 1.2 1.4 1.6
CD

Figure 2: A drag-polar plot of CL v.s. CD at Re = 4, 000.

6
 1.5 Lift model
Lift model
Re 100,000 CFD
0.8 Re 200 CFD
1
0.6
0.4 0.5
0.2

L
0

C
CL

0
−0.2 −0.5
−0.4
−1
−0.6
−0.8 −1.5
80 85 90 95 100 80 85 90 95 100
Time Time

(a) Re = 200 (b) Re = 100, 000

Figure 3: Comparisons between the simulated and modeled steady-state lift coefficients.

3. Equations (7) and (8) are then solved for the nonlinear and linear damping coefficients α and µ and
the angular frequency ω.

4. With all of the parameters identified, equation (3) is numerically integrated using a Runge-Kutta
routine and the results are compared with the CFD results.

In Figure 3, we compare the time history of the steady-state lift obtained from the CFD simulation with
that obtained by integrating equation (3) for the Reynolds numbers Re = 200 and Re = 100, 000. We note
that the steady-state solution of the van der Pol equation is independent of the initial conditions. It is clear
that the van der Pol oscillator accurately models the CFD results for a wide range of Reynolds number flows.
Next, we check whether the van der Pol oscillator identified using the steady-state lift data can also
model the transient lift. This serves two purposes:

1. It gives confidence that the physics is modeled correctly.

2. It confirms that the model is capable of simulating the transient as well as the steady-state response.

Using the values of α, µ, and ω defined earlier and specifying the initial conditions CL (t0 ) and ĊL (t0),
equation (3) is integrated for a long time. Because the CFD results only yield the values of CL , the initial
values of ĊL are approximated using a five-point fourth-order central finite-difference formula. In Figures 4
and 5, we compare the transient lift obtained from the model and the CFD analysis for Re = 200 and
Re = 100, 000, respectively. For each case, results initiated at four different initial times are presented.
For the case Re = 200, we simulate the transient lift using the initial starting times t0 = 34, 39, 45, and
50. We see from Figure 4 that, the larger the starting time is, the more accurate the model is in simulating
the CFD results. For example, when t0 = 34, the van der Pol oscillator underestimates the amplitude of
the lift markedly; also, there is a phase shift between its prediction and the CFD lift. These differences,
however, diminish as the starting time t0 is increased, as shown in the results for t0 = 50, which are in very
good agreement with the CFD results.
For the case Re = 100, 000, we show in Figure 5 the predicted transient lift using the initial starting
times t0 = 10, 12, 14, and 16. It follows from Figure 5 that the discrepancy in modeling the lift amplitude
for moderate Reynolds number flows is resolved. However, there is still a phase difference between the two
results, which again is minimized when using larger starting times.

7
 1 1
Lift model
CFD

CL
L
0 0

C
to=34
to=45
−1 −1
20 30 40 50 60 70 80 20 30 40 50 60 70 80
Time Time
1 1

L
L

0 0

C
C

to=39
to=50
−1 −1
20 30 40 50 60 70 80 20 30 40 50 60 70 80
Time Time

Figure 4: Simulated and modeled transient lift coefficients for different starting times for Re = 200.

2 Lift model 2
CFD

L
L

C
0 0
C

to=10 to=14
−2 −2
0 20 40 60 0 20 40 60
Time Time

2 2
to=12 to=16
L

CL

0 0
C

−2 −2
0 20 40 60 0 20 40 60
Time Time

Figure 5: Simulated and modeled transient lift coefficients for different starting times for Re = 100, 000.

One possible explanation for these deviations is the fact that the CFD results contain two types of
transients, those arising physically and those arising from numerical analysis effects (e.g., impulsive initial
conditions), and it seems that the rate of decay of the latter depends on the Reynolds number. We note
from Figures 4 and 5 that the larger the Reynolds number is, the faster the rate of convergence is to the
physical solution. For Re = 200, it follows from Figure 4 that the transients due to numerical effects in the
CFD simulation might propagate for some time after the initial conditions and might take up to 50 time
units to decay completely. On the other hand, for Re = 100, 000, it follows from Figure 5 that the transients
due to numerical effects decay much faster, lasting for about 16 time units.
Clearly, the agreement between the van der Pol oscillator and CFD results is quite good once the transients
arising from numerical effects have died out. Therefore, the van der Pol oscillator seems to be capable of
modeling both of the steady-state lift and the transient lift on a cylinder in uniform flow.

3.2 Drag
Referring to Figure 1, we note that the drag consists of two major components. The first is a mean compo-
nent that monotonously approaches a constant value in the steady state; this component is assumed to be
independent of the lift. The second is an oscillatory component related to the lift and has a frequency equal

8
to twice the lift frequency of oscillation.
Since the lift and drag have a common source, the pressure distribution of the cylinder surface, and in
view of this two-to-one frequency relationship, Nayfeh et al. (2003) reasoned that the drag is quadratically
related to the lift in some fashion. They examined the phase relation between the periodic components of
the drag and lift and found that it falls around 270◦ . Hence, they inferred that the periodic component of
the drag must be proportional to −CL ĊL and proposed the drag model
a2
CD (t) = hCD i − 2 CL (t)ĊL (t) (10)
ωs a21

where a2 is the amplitude of the drag component at 2fs and hCD i is the mean component of drag. For
steady-state behavior, hCD i is constant, while, for transient behavior, it increases monotonously with time.
The constant value of hCD i is determined from the CFD steady-state time history of the drag and the value
of a2 is determined from its spectral analysis.
In Figure 6, the drag results obtained from equation (10) are compared with the CFD calculations for
the Reynolds numbers Re = 200 and 100, 000. For Re = 200, excellent agreement is found between the two
solutions. For Re = 100, 000, the agreement is also quite good, but there appears to be a slight deviation in
phase between the two results.
We also examined the capability of equation (10) to predict the transient drag. However, because the
mean drag hCD i is time-dependent in the transient region, modeling it as a constant following the steady-
state results brings about a mismatch with the CFD solution. To account for this discrepancy, we first
extracted the transient profile of the mean drag from the CFD simulation and fitted it into a polynomial
function in time to replace the constant value of hCD i. Then, substituting the transient lift results at different
initial times into equation (10), we calculated the corresponding transient drag.
In Figures 7 and 8, we compare the transient results from the CFD code and the drag model for the
Reynolds numbers Re = 200 and 100, 000. Again, we find that, in general, the accuracy of the proposed
model improves for later starting times as the transient effects due to numerical computations diminish. For
the low Reynolds number flow in Figure 7, we find that the proposed model underestimates the amplitude
of CD when starting at t0 = 40, but it does an excellent job when starting at t0 ≥ 60, which is still in
the transient region. For the moderate Reynolds number flow in Figure 8, we find that the model does an

1.24
Drag model 0.94 Drag model
Re 200 Re 100,000
CFD CFD
1.22 0.92

0.9
1.2
0.88
D
D

C
C

1.18 0.86

0.84
1.16
0.82

1.14 0.8

80 85 90 95 100 0.78
80 85 90 95 100
Time Time

(a) Re = 200 (b) Re = 100, 000

Figure 6: Comparisons between the simulated and modeled steady-state drag coefficients.

9
 1.25 1.25
t =40 t =60
1.2 o o
1.2
1.15 1.15

D
C

C
1.1 Drag mode 1.1
1.05 CFD 1.05
1 1
40 50 60 70 80 40 50 60 70 80
Time Time
1.25 1.25
t =50 to=65
o
1.2 1.2
1.15 1.15
D

D
C

C
1.1 1.1
1.05 1.05
1 1
40 50 60 70 80 40 50 60 70 80
Time Time

Figure 7: Simulated and modeled transient drag coefficients for different starting times for Re = 200.

1 1
to=10 to=14
0.9 0.9

0.8 0.8
CD
CD

0.7 Drag mode 0.7

CFD
0.6 0.6
10 20 30 40 50 60 10 20 30 40 50 60
Time Time
1 1
to=12 t =16
o
0.9 0.9
0.8 0.8
CD
CD

0.7 0.7
0.6 0.6
10 20 30 40 50 60 10 20 30 40 50 60
Time Time

Figure 8: Simulated and modeled transient part of the drag coefficients for different starting times for Re
= 100, 000.

excellent job in predicting the amplitude of CD ; however, there exists a small phase difference between the
two solutions.

4 Improved Lift and Drag Models

The results presented in the previous section are solely based on matching the amplitudes and frequencies
of the CFD and model results. Even with the good agreement found, especially at steady state, we believe
there is still room for improvement. Although the use of higher-order spectral analysis of the CFD results
showed that the phase φ13 between the lift components at fs and 3fs is nearly 90◦, it actually differs from
one case of Reynolds number to another. In fact, in some of our calculations, we have found separation from
90◦ as large as ±25◦ . Obviously, this wide variation in the phase is not fully accounted for in the van der
Pol model in equation (3). This fact does not produce a real problem as long as one is concerned with the
time histories because a3 is usually two orders of magnitude smaller than a1. However, because we intend to
extend this model to cases of forced and freely vibrating cylinders, we would like to have it be very accurate
in both the time domain and the spectral domain.
Similarly, for the drag CFD calculations, we have found that the phase φ12 between the lift component

10
at fs and the drag component at 2fs varies from 270◦ by as much as 85◦ . Qin (2004) proposed that the
quadratic term in the drag model should be of the form CL2 instead of −CL ĊL . He also found linear coherence
between the drag and lift components at fs and 3fs and introduced a linear lift term in the drag model to
account for it.

4.1 Lift
Here, we modify the van der Pol oscillator by introducing a Duffing-type nonlinearity to equation (3),
resulting in the new lift model
C̈L + ω2 CL = µĊL − αCL2 ĊL − γCL3 (11)

The coefficient γ is determined based on matching precisely the phase φ13 obtained from the CFD data to
the phase obtained from solving equation (11). In this process, we use the method of harmonic balance to
determine approximate solutions of the model. These solutions along with spectral analysis of the CFD data
are used to identify all of the parameters in equation (11).
We seek a solution for the lift coefficient of the form

CL (t) = c1 cos(ωst) + c2 sin(ωst) + c3 cos(3ωs t) + c4 sin(3ωs t) (12)

Then, upon substituting equation (12) into equation (11) and separating the terms multiplying the different
sine and cosine functions, we obtain the linear system

Ay = b (13)

where y = {ω2 , µ, α, γ}, b = {c1, c2, 9c3, 9c4}ωs2, and the entries of the matrix A are

A11 = c1, A21 = c2 , A31 = c3 , A41 = c4 ,

A12 = −c2 ωs, A22 = c1ωs , A32 = −3c4ωs , A42 = 3c3 ωs,
1 3

A13 = c2 − c4c22 + c21c2 + 2c23c2 + 2c24 c2 − 2c1c3 c2 + c21 c4 ωs ,
4
1

A23 = −c31 − c3 c21 − c22 c1 − 2c23c1 − 2c24c1 − 2c2 c4c1 + c22 c3 ωs,
4
1

A33 = −c32 + 6c4c22 + 3c21c2 + 3c34 + 6c21 c4 + 3c23c4 ωs ,
4
1

A43 = −c31 − 6c3c21 + 3c22c1 − 3c33 − 3c3 c24 − 6c22c3 ωs ,
4
1

A14 = 3c31 + 3c3c21 + 3c22 c1 + 6c23c1 + 6c24c1 + 6c2 c4c1 − 3c22 c3 ,
4
1

A24 = 3c32 − 3c4c22 + 3c21 c2 + 6c23c2 + 6c24c2 − 6c1 c3c2 + 3c21 c4 ,
4
1 3

A34 = c1 + 6c3c21 − 3c22 c1 + 3c33 + 3c3 c24 + 6c22c3 ,
4
1

A44 = −c32 + 6c4c22 + 3c21c2 + 3c34 + 6c21 c4 + 3c23c4
4
The cj are determined by matching the solution in equation (12) with the steady-state CFD simulation
solution which is expressed as

CL (t) = a1 cos(ωst) + a3 cos(3ωs t + φ13) (14)

This yields c1 = a1 , c2 = 0, c3 = a3 cos φ13, and c4 = −a3 sin φ13.

11
 Table 1: Lift parameters at different Reynolds numbers.
Re = 200 Re = 4, 000 Re = 100, 000
fs 0.192 0.23 0.254
a1 0.621 1.23 1.06
a3 0.0043 0.042 0.021
φ13 78◦ 107◦ 100◦
ωs 1.206 1.445 1.596
ω 1.185 1.562 1.641
µ 0.066 0.386 0.251
α 0.68 1.050 0.902
γ 0.192 −0.297 −0.166

0.8 (a) 1.6 (b)

0.4 0.8
CL 0 0
-0.4 -0.8
-0.8 -1.6
100 104 108 112 116 120 80 84 88 92 96 100
Time Time
1.6 (c)
0.8
CL 0 CFD
-0.8 Model
-1.6
100 104 108 112 116 120
Time

Figure 9: Comparisons of the steady-state time histories of the lift coefficients obtained with the CFD
simulation and the improved model: (a) Re = 200, (b) Re = 4, 000, and (c) Re = 100, 000.

Listed in Table 1 are the results for the cases Re = 200, 4, 000, and 100, 000, covering a fairly wide range
of Reynolds numbers. Typically, risers operate in the lower subset of that range while spars operate in the
upper subset and beyond it. From the values of α and γ, it is clear that the influence of the Duffing-type
cubic term should not be neglected in the modeling. In Figure 9, we plot the steady-state time histories of
the lift coefficient obtained from the improved model and the CFD simulations. For all three cases of Re,
there is excellent agreement between the two solutions.
Next, we examine the validity of this model in predicting the transient behavior of the lift coefficient. In
Figure 10, we show the transient variations of the lift coefficient with time for Re = 200. The results of the
improved model are compared with the CFD results for the starting times t0 = 40, 50, and 60 time units.
It is clear that the improved model for the lift matches the CFD simulation results in amplitude and phase
with excellent accuracy for t0 ≥ 50. In Figure 11, the transient variations of the lift coefficient with time
for Re = 4, 000 are shown. The results of the improved model are compared with the CFD results for the
starting times t0 = 38, 42, and 45 time units and the improved model matches very well the CFD simulation
for t0 ≥ 45. Lastly, we present in Figure 12 the transient variations of the lift coefficient with time for
Re = 100, 000. The results of the improved model are compared with the CFD results for the starting times
t0 = 12, 15, and 18 time units. The improved lift model matches very well the CFD simulation for t0 ≥ 18.

12
 0.8 (a) (b)
0.4 t0 = 40
CL
0
-0.4 t0 = 50
-0.8
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
0.8 (c) Time
0.4
CFD
CL 0 Model
-0.4
t0 = 60
-0.8
30 40 50 60 70 80 90 100
Time

Figure 10: Time histories of the lift coefficients in the transient region from the CFD and improved van der
Pol model for Re = 200.

1.6 (a) ( b)
0.8 t0 = 38
C 0
L

-0.8 t0 = 42
-1.6
30 40 50 60 70 80 30 40 50 60 70 80
1.6 (c) Time
0.8
CFD
C 0
L
Model
-0.8 t0 = 45
-1.6
30 40 50 60 70 80
Time

Figure 11: Time histories of the lift coefficients in the transient region from the CFD and improved van der
Pol model for Re = 4, 000.

Nonetheless, a small deviation in phase still remains in the lift transient results for t0 = 18.

4.2 Drag
In the simulations we have conducted thus far, we have found that the phase φ12 is around 270◦. However,
the actual value of the phase may vary from one case of Reynolds number to another by almost 85◦, which
is quite significant. Therefore, we propose a new model in which the drag is proportional to both CL ĊL and
CL2 in the following manner:
a2

a2
CD (t) = hCD i + 2k1 CL2 − CL2 + 2k2 CLĊL (15)
a21 ωsa21

where the brackets h i denote the mean value of the term inside.
The first expression in equation (15) stands for the mean component of the drag, which reaches a constant


value hCD iss in the steady-state region. The term − CL2 in the second expression in equation (15) negates
the DC component introduced by CL2 . The amount of contribution from both quadratic expressions to the

13
 1.4 (a) ( b) t0 = 15
0.7
C 0
L

-0.7
-1.4 t0 = 12
10 20 30 40 50 60 10 20 30 40 50 60
1.4 (c) Time
0.7
CFD
C 0
L
Model
-0.7 t0 = 18
-1.4
10 20 30 40 50 60
Time

Figure 12: Time histories of the lift coefficients in the transient region from the CFD and improved van der
Pol model for Re = 100, 000.

Table 2: Drag parameters at different Reynolds numbers.

Re = 200 Re = 4, 000 Re = 100, 000
hCD iss 1.18 1.42 0.85
a2 0.039 0.152 0.065
φ12 −35◦ −18◦ −7◦
k1 0.82 0.95 0.992
k2 −0.57 −0.314 −0.127

overall behavior of the drag is determined by matching precisely the amplitude a2 and phase φ12 obtained
from the model with the CFD results at steady state.
To this end, we substitute equation (14) into equation (15), expand the result, and obtain

CD (t) = hCD iss + a2 [k1 cos(2ωst) − k2 sin(2ωs t)] + · · · (16)

Then, by comparing equation (16) with the CFD solution

CD (t) = hCD i + a2 cos(2ωs t + φ12) + · · · (17)

we obtain k1 = cos φ12 and k2 = sin φ12. In Table 2, we present the drag model parameters for Re =
200, 4, 000, and 100, 000. It is clear from the values of k1 and k2 that the term CL2 can play a role nearly as
influential as the term CLĊL on the behavior of the drag coefficient.
In Figure 13, we plot the steady-state time histories of the drag coefficient for Re = 200, = 4, 000,
and = 100, 000. Results obtained from the improved drag model are compared with the CFD simulation
results and excellent agreement is demonstrated for all of the cases presented.
To adapt this model to capture the transient behavior of the drag as well, we only need to express hCD i
as a time-dependent function, which is extracted from the CFD data and asymptotes hCD iss with time. We
present in Figure 14 the transient variations of the drag coefficient with time for Re = 200. The results
of the improved model are compared with the CFD results for the starting times t0 = 40, 50, and 60 time
units. It is clear that the improved drag model matches the CFD simulation results in amplitude and phase

14
 1.24 (a) 1.6 (b)
1.2 1.5
D
C 1.4
1.16 1.3
1.12 1.2
100 104 108 112 116 120 80 84 88 92 96 100
Time Time
0.94 (c)
0.88
D
C CFD
0.82 Model
0.76
100 104 108 112 116 120
Time

Figure 13: Comparisons of the steady-state time histories of the drag coefficient obtained from the CFD
simulation and the improved model: (a) Re = 200, (b) Re = 4, 000, and (c) Re = 100, 000.

with excellent accuracy for t0 ≥ 50. We present in Figure 15 the transient variations of the drag coefficient
with time for Re = 4, 000. The results of the improved model are compared with the CFD results for the
starting times t0 = 39, 43, and 47. The improved model matches very well with the CFD simulation for
t0 = 47 time units. Lastly, we present in Figure 16 the transient variations of the drag coefficient with time
for Re = 100, 000. The results of the improved model are compared with the CFD results for the starting
times t0 = 12, 20, and 30 time units. The improved model matches very well with the CFD simulation for
t0 = 30.

5 Conclusions
Lift and drag coefficients for the two-dimensional flow over a stationary circular cylinder were examined
and modeled as a preliminary step to understanding the complex problem of vortex-induced vibrations in
risers and other offshore slender structures. The models account for the coupling between these coefficients

1.3 (a) ( b)
1.2
D
C1.1 t0 = 50
1 t0 = 40
CFD
Model
0.9
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
1.3 (c) (d)
1.2 >D
D
C1.1 t0 = 60 C< <CD>ss
1
0.9
30 40 50 60 70 80 90 100 30 40 50 60 70 80 90 100
Time Time

Figure 14: Time histories of the drag coefficient in the transient region from the CFD and improved van der
Pol model for Re = 200.

15
 1.6 (a) (b)
D
1.4 t0 = 43
C t0 = 39
1.2 CFD
Model
1
30 40 50 60 70 80 30 40 50 60 70 80
1.6 (c) (d)
D
1.4 >D
C C
< <CD>ss
1.2 t0 = 47
1
30 40 50 60 70 80 30 40 50 60 70 80
Time Time

Figure 15: Time histories of the drag coefficient in the transient region from the CFD and improved van der
Pol model for Re = 4, 000.

1 (a) (b)
0.9
CD0.8
0.7 CFD
t0 = 20
t = 12 Model
0.6 0
10 20 30 40 50 60 10 20 30 40 50 60
1 (c) (d)
0.9 >D
CD0.8 C
<
0.7 t0 = 30 <CD>ss
0.6
10 20 30 40 50 60 10 20 30 40 50 60
Time Time

Figure 16: Time histories of the drag coefficient in the transient region from the CFD and improved van der
Pol model for Re = 100, 000.

and cover the steady-state and transient behaviors. Initially, these models were based on representing the
lift coefficient by the van der Pol equation and representing the drag coefficient by the sum of a mean
drag term and a nonlinear term proportional to the lift coefficient times its time derivative. Values for
the parameters in these models were obtained by matching a second-order approximate solution of the van
der Pol equation with the CFD steady-state lift. The latter was obtained by numerically integrating the
unsteady Reynolds-averaged Navier-Stokes equations (RANS).
These initial models assumed that the phase between the main and third harmonic components of the lift
and the phase between the main components of the lift and the drag are fixed at φ13 = 90◦ and φ12 = 270◦,
respectively. However, the fact is that these phases vary from the aforementioned values, sometimes quite
considerably, with the Reynolds number. Therefore, we set out to present improved models for the lift and
drag coefficients that explicitly account for these phases.
For the lift coefficient, the van der Pol oscillator was modified by adding a Duffing-type cubic nonlinearity
term. Then, the values of the model parameters were computed by matching the model solution, obtained

16
by the method of harmonic balance, to the steady-state CFD results.
As for the drag coefficient, the initial model was modified by introducing a second quadratic term that


is proportional to CL2 − CL2 . The amount of contribution from each quadratic term was computed based
on exactly matching the phase φ12 to the phase computed from the CFD data.
These improved models were then compared with the CFD solutions in both the steady-state and transient
regimes for the three Reynolds numbers: Re = 200, Re = 4, 000, and Re = 100, 000. For the steady-state lift
and drag coefficients, we found that the improved lift and drag models do an excellent job in matching the
CFD results. For the transient lift and drag coefficients, we found that the accuracy of the model depends
on the starting point at which the initial conditions are taken. The reason is that part of the CFD transient
results arise from numerical inaccuracies due to the impulsive initial conditions imposed. Because these
numerical effects diminish with time, it stands to reason that the agreement between the models and the
CFD results in the transient regimes improves with later starting times. Nonetheless, it is clear that these
models predict the transient behavior quite well. This solidifies our conviction in these models as an initial
step towards modeling the full fluid-structure problem.

6 Acknowledgment
This work was supported by Vetco Gray and Starmark Offshore Inc., David W. Hughes and Robert Sexton,
Technical Monitors.

References
Bishop, R.E.D., Hassan, A.Y., 1963, The lift and drag forces on a circular cylinder in a flowing fluid,
Proceedings of the Royal Society Series A 277, 32-50.

Chorin, A. J., 1997, A Numerical method for solving incompressible viscous flow problems, Journal of
Computational Physics 135, 118-125.

Currie, I. G., Turnball, D. H., 1987, Streamwise oscillations of cylinders near the critical Reynolds number,
Journal of Fluids and Structures 1, 185-196.

Hartlen, R. T., Currie, I. G., 1970, Lift-oscillator model of vortex-induced vibration, ASCE Journal of
Engineering Mechanics 96, 577-591.

Iwan, W.D., Blevins, R.D., 1974, A model for vortex-induced oscillation of structures, Journal of Applied
Mechanics 41 (3), 581-586.

Iwan, W.D., 1975, The vortex induced oscillation of elastic structures, Journal of Engineering for Industry
97, 1378-1382.

Iwan, W.D., 1981, The vortex-induced oscillation of non-uniform structural systems, Journal of Sound and
Vibration 79 (2), 291-301.

Krenk, S., Nielsen, S.R.K., 1999, Energy balanced double oscillator model for vortex-induced vibrations,
Journal of Engineering Mechanics 125 (3), 263-271.

Kim, W. J., Perkins, N. C., 2002, Two-dimensional vortex-induced vibration of cable suspensions, Journal
of Fluids and Structures 16 (2), 229-245.

17
Landl, R., 1975, A mathematical model for vortex-excited vibrations of bluff bodies, Journal of Sound and
Vibration 42 (2), 219-234.

Nayfeh, A. H., 1973, Perturbation Methods, Wiley, New York.

Nayfeh, A. H., 1981, Introduction to Perturbation Techniques, Wiley, New York.

Nayfeh, A. H., Owis, F., Hajj, M. R., 2003, A model for the coupled lift and drag on a circular cylinder,
Proceedings of DETC03, ASME 2003 Design Engineering Technical Conferences and Computers and
Information in Engineering Conference.

Nayfeh, A. H., Marzouk, O. A., Arafat, H. N., Akhtar, I., 2005, Modeling the Transient and Steady-State
Flow over a Stationary Cylinder,” ASME 20th Biennial Conference on Mechanical Vibration and Noise,
5th International Conference on Multibody Systems, Nonlinear Dynamics and Control, DETC2005-85376,
Long Beach, CA, Sep 2005.

Qin, L., 2004, Development of Reduced-Order Models for Lift and Drag on Oscillating Cylinders with
Higher-Order Spectral Moments, Ph.D. Dissertation, Virginia Tech, Blacksburg, VA.

Rogers, S. E., Kwak, D., Kaul, U., 1987, On the accuracy of the pseudocompressibility method in solving
the incompressible Navier-Stokes equations, Applied Mathematics Modelling 11, 35-44.

Skop, R. A., Griffin, O. M., 1973, A model for the vortex-excited resonant response of bluff cylinders, Journal
of Sound and Vibration 27 (2), 225-233.

Skop, R. A., Griffin, O. M., 1975, On a theory for the vortex-excited oscillations of flexible cylindrical
structures, Journal of Sound and Vibration 41 (3), 263-274.

Skop, R. A., Balasubramanian, S., 1997, A new twist on an old model for vortex-excited vibrations, Journal
of Fluids and Structures 11 (4), 395-412.

Spalart, P. R., and Allamras, S. R., 1992, “A one-equation turbulence model for aerodynamic flows,” In
Proceedings of the AIAA 30th Aerospace Sciences Meeting and Exhibit, AIAA paper 92-0439.

Williamson, C. H. K., Govardhan, R., 2004, vortex-induced vibrations, Annual Review of Fluid Mechanics,
36 (4) 13-55.

18