Sports Engineering (2024) 27:34

[Link]

ORIGINAL ARTICLE

Practical computational fluid dynamic predictions of a cyclist in a time

trial position
Morgan Taylor1 · Daniel Butcher1 · Conor Crickmore1 · A. Duncan Walker1

Accepted: 16 September 2024 / Published online: 30 October 2024

© The Author(s) 2024

Abstract
On a flat road, at race speeds, aerodynamic drag is the main resistive force a cyclist must overcome. Computational fluid
dynamics (CFD) can be a useful tool to predict and understand the complex flow and, therefore, drive developments to
reduce drag. However, cycling aerodynamics is complex. The effects of Reynolds number, surface roughness, boundary
layer transition, flow separation, and turbulent wakes are challenging to accurately predict. High fidelity time-resolved com-
putations, such as Large eddy simulations (LES), require high-performance computing and lengthy simulation times. This
paper examines whether lower fidelity CFD, such as Reynolds averaged approaches, can predict the drag of a cyclist with
sufficient accuracy and within practical timescales on a desktop PC. Wind tunnel tests of a rider model (without bicycle) were
conducted at Reynolds numbers equivalent to speeds of ~ 20–70 km/h. Measured drag showed a notable Reynolds number
dependency with the drag coefficient reducing almost linearly by ~ 20% from 0.88 to 0.71. The computational accurately
replicated this relationship but only when employing a boundary layer transition model. The steady computations underpre-
dicted the magnitude of the measured drag coefficient by ~ 3% but the unsteady computations were within ~ 2%. Examina-
tion of the predicted flow field revealed variations in boundary layer transition, separation, and wake formation from each
body part which combine in a complex wake system. Overall, the data confirm validity and suitable accuracy of the CFD,
and therefore this provides a practical time and cost-effective tool for further examination of drag reduction within cycling.

Keywords Cycling · Drag reduction · Aerodynamics · Reynolds number · Computational Fluid Dynamics · Wind tunnel

1 Introduction drag mechanisms without the use of complex measurement

techniques (such as Particle Image Velocimetry [5]).
At speeds typical of a pro-cyclist, on a flat road, aerody- CFD has become a popular tool used by several authors
namic drag accounts for about 90% of the resistance a cyclist [6–17]) as it predicts the entire flow and is better for knowl-
must overcome [1], with 60–80% of this coming from the edge-based development. However, cycling aerodynamics is
cyclist themselves [2]. Consequently, reducing drag is cru- complex. It is challenging to accurately predict the effects of
cial to maximize speed for a given rider power. There are surface roughness, boundary layer transition, flow separation,
several methods to determine cyclist drag, including meas- and turbulent wakes. The expectation is that high fidelity com-
urements in wind tunnels, on the track or on the road testing. putations such as large eddy simulations (LES) or Detached
Alternatively, Computational Fluid Dynamics (CFD) can eddy simulations (DES) would be the most accurate (see for
also be used to simulate tests. The advantage of each method example, Fintelman et al. [9]). However, due to requirements
is discussed by Crouch et al. [2] and Malizia and Blocken for increased mesh density and small time steps, this level of
[3, 4]. Wind tunnels are the most accurate for measuring fidelity could require as much as 50–80 more CPU hours than
drag, but do not provide information on flow topology and steady Reynolds-Averaged Navier–Stokes (RANS) predictions
[9]. The time taken depends on the number of CPUs available
and LES/DES compute times could be reduced using high-
* A. Duncan Walker performance computing. For product development or rider
[Link]@[Link] optimization, this cost and time implication of LES/DES
are impractical. Hence, this paper examines if lower fidel-
1
Department of Aeronautical and Automotive Engineering, ity CFD can be used to predict, with sufficient accuracy, the
Loughborough University, Loughborough LE11 3TU, UK

Vol.:(0123456789)
34 Page 2 of 9 M. Taylor et al.

aerodynamic drag of a cyclist within practical timescales of The effect of Re may be explained if the cyclist’s body
1–2 days on a desktop PC. parts are approximated as cylinders. The wake behind a
The complexity of the flow topology around a cyclist is cylinder is characterized by the point at which flow sepa-
illustrated in Fig. 1 via flow visualization [16] and PIV meas- rates from the surface. Achenbach [19, 20] showed this is
urements [5]. The flow is characterized by vortical structures dependent on Re. At low values (Re < 1 × ­104), the boundary
shed from various body parts, combining to form spatially layer on a smooth cylinder remains laminar and separates at
and temporally unsteady wake structures. This turbulent wake θ < 90° (Fig. 2). Above a critical Re value (­ Recrit ~ 5 × ­105),
has a low static pressure, and the difference (Δp) between this the boundary layer transitions to turbulent, and the enhanced
and the high pressure generated as flow impinges on the front momentum exchange of the turbulent eddies means the
of the cyclist produces a pressure drag force ( Fd ), Eq. (1). boundary layer remains attached longer. At these higher
This pressure (or wake) drag is much larger than skin friction Re, separation is delayed to θ > 120°, with the result that
drag and is the dominant resistive force that the cyclist must the wake reduces in size and the pressure drag reduces. Cd
overcome. drops from ~ 1.2 to ~ 0.2. Notably a cyclist operates across
this transitional range of Re such that Cd can vary substan-
Fd = ΔpA (1) tially with speed. The local Re of each body part means
Drag can be reduced by minimizing frontal area ( A) or by this critical behavior can happen at different speeds. Early
improving aerodynamic efficiency (streamlining) to minimize boundary layer transition can be forced by increased free
the size and strength of the wake. This is quantified in Eq. (2) stream turbulence or enhanced surface roughness; the latter
by the drag coefficient (Cd ) which relates the drag force to the effect is the reason textured fabrics are used to reduce drag.
velocity (U∞) and frontal area ( A). This illustrates why drag CFD has become popular as a design tool across all
area (Cd A) is often used to quantify drag but it is crucial to aerodynamic applications as it predicts quantitatively the
realize that Cd and A work in tandem. According to Malizia entire flow field. It is more attractive than wind tunnel force
and Blocken [4], a 2% variation in Cd A can result in about measurements or simple qualitative flow visualization such
a 0.5s time difference per kilometer for a cyclist traveling at as smoke as the three-dimensional quantitative data it pro-
57–60 km/h. duces can be more thoroughly analyzed and used to develop
a deeper understanding of the pertinent flow mechanisms.
Fd =
1 2
𝜌U C A (2) However, the complex flow around a cyclist is challenging to
2 ∞ d accurately model and, if not considered carefully, may result
The aerodynamic flow regime also plays a role, and this is in inaccurate results. The ability to predict trends, such as
characterized by the Reynolds number (Re), Eq. (3). Re is the the Re dependency shown in Fig. 2, is crucial if CFD is to
ratio of inertial to viscous forces and determines whether the be used to help reduce the drag of a cyclist.
flow is laminar (low Re) or turbulent (high Re). van Druenen and Blocken [10] provides a review of the
effect of relevant computational parameters on the prediction
of the drag of a cyclist. In summary, results are seen to be
√
𝜌U∞ A
Re = (3) sensitive to computational mesh resolution and the choice
𝜇
of turbulence model. They recommend that a well-refined
where A is the frotal area, U∞ the velocity, ρ the air density, near-wall region (y +  < 1) should be employed in conjunc-
and 𝜇 the molecular viscosity. tion to a laminar-to-turbulent transition model to capture the

Fig. 1  Complex flow topol-

ogy generated by a cyclist (left
Crouch et al. [16], right Terra
et al. [5]). Reproduced under
Creative Commons Attribu-
tion 4.0 International License
([Link]/licenses/
by/4.0/)
Practical computational fluid dynamic predictions of a cyclist in a time trial position Page 3 of 9 34

Fig. 2  Typical variation of Cd v

Re for a cylinder

development of the boundary layer. Approximations used can be successful in predicting cyclist drag. However, Fintel-
by boundary-layer wall functions are insufficient. Similarly, man et al. [9] reported an increased factor of 50 and 80 CPU
wake regions must have sufficiently refined mesh sizes to hours, respectively, for DES and LES compared to RANS.
capture the gradients driving the turbulent mixing. This This paper’s objective is to examine if practical CFD
leads to mesh sizes for a bicycle and rider typically > 30 (RANS or uRANS) can be used in a timely manner to
million elements which represents a high computational help reduce drag in competitive cycling. Van Druenen and
demand. The choice of turbulence model is, therefore, a Blocken [10] found good agreement with wind tunnel data
balance between accuracy and computational cost. Broadly at one speed using both SAS and steady RANS. However,
speaking this balance depends on how much of the turbu- the aerodynamics is known to be Re dependent [5] and, as
lence is directly computed or modeled. such, this paper will also examine if CFD can replicate this
Direct numerical solution of the Navier–Stokes equa- behavior.
tions compute all scales of turbulence, but this is not com-
putationally feasible. Large Eddy Simulation (LES) directly
computes the large-scale turbulent structures temporally and 2 Methodology
spatially, modeling only the smallest sub grid eddies. How-
ever, LES requires expertise to obtain accurate results and, 2.1 Experimental
with highly refined meshes, is computationally expensive.
The most economic option is to solve the RANS using a Validation data were provided using Loughborough Univer-
simpler model to predict the turbulence without resolving sity’s large open-circuit wind tunnel [19]. This wind tunnel
all temporal and spatial scales. The k-ω SST (shear stress has a closed working section of 1920 × 1320 × 3600 mm
transport) two-equation turbulence model is generally the (W × H × L) and a maximum tunnel speed of 40 m/s at which
preferred option for RANS [10]. Given that RANS assumes the turbulence intensity is 0.15% [19]. For simplicity, only
the turbulence is statistically steady and homogeneous, it the cyclist’s body was considered, as done by Defraeye
will not capture unsteady effects. Unsteady (u)RANS can et al. [13] and Giljarhus et al. [14, 15]. This simplification
include some unsteady effects, but it uses the same turbu- is acceptable for the aims of this study as the cyclist’s body
lence closure as steady RANS and is still limited by the accounts for 60–80% [2] of the total drag. Inclusion of the
same modeling assumptions. Scale-Adaptive Simulations bicycle will increase the number of cells, and increase com-
(SAS) resolves larger scale eddies, similar way to LES, in putational power, making CFD simulations less practical.
regions of high unsteadiness while modeling smaller scales With reference to Fig. 1 the cyclist’s body generates a com-
with a RANS approach elsewhere. Detached Eddy Simu- plex flow and studying this in isolation from the bicycle will
lations (DES) uses RANS near walls and LES away from yield useful information.
walls with the switch based on local grid resolution. Hybrid A high-resolution 3D-printed model of an elite male
RANS-LES provides an attractive alternative. Javadi [11, cyclist was mounted directly to the underfloor six-axis bal-
12] showed that unsteady RANS and hybrid RANS-LES ance on a cylindrical sting (∅25 mm) as shown in Fig. 3.
34 Page 4 of 9 M. Taylor et al.

Away from the wall, polyhedral cells were used with volu-
metric controls employed to increase grid resolution in key
areas as shown in Fig. 4 using a growth rate of 1.2. The region
immediately surrounding the cyclist (outside the prism layers)
had a cell size of ~ 10 mm and outside of that ~ 20 mm. Note
that Fig. 4 shows only the part of the domain containing these
two volumetric controls. A mesh sensitivity study was con-
ducted examining these settings, but it is beyond scope here to
report in full. The final mesh parameters were similar to those
of van Druenen and Blocken [10] who studied the impact of
these parameters. The final mesh had about 16 million cells.
Both steady RANS and unsteady (u)RANS were employed
using a k-𝜔 SST turbulence model and γ-transition model. The
γ-transition is a one-equation correlation-based model that
Fig. 3  3D-printed ½-scale model of a cyclist
solves a transport equation for intermittency, γ, which is then
used to modify the transport equation for the turbulent kinetic
The cyclist model was obtained from a 3D scan of a profes- energy [21] as an additional source term. Intermittency defines
sional cyclist provided as a CAD file by their professional if the flow is laminar (γ = 0) or turbulent (γ = 1). The model
team. The work was undertaken with the consent of the uses a trigger function, Fonset, and a model coefficient, Conset, to
scanned athlete and within the guidelines of the Loughbor- govern the production of γ and the onset of transition. Higher
ough University Ethical Policy Framework. The athlete had values of Fonset (lower values of Conset) lead earlier transition to
a height of 1.83 m and a weight of 68 kg. To keep the tunnel turbulence. As seen later, the value of model coefficient, Conset,
blockage < 3%, [20], the model was printed at ½-scale and has a notable effect. Siemens STAR-CCM + uses a default of
the tunnel run at 2 × speed to maintain the Re. The model Conset=2.2 [21] but does not state why; hence, the need to fur-
was printed in ABS (acrylonitrile butadiene styrene) using ther investigate the effect of this.
a Statysys F270 printer with a layer thickness of 0.254 mm. ( )
Measurements were taken for Re ~ 2 × ­105 – 8 × ­105 equating Re𝜈
Fonset = fn (4)
to full-scale velocity ranging from 20 to 70 km/h. The drag Conset Re𝜃c
force, Fd , was measured directly by the balance with a mean 2
where Re𝜈 = d𝜈S is the strain rate Reynolds number, with
value computed form 120s of data measured at 300 Hz.
d the distance to the wall, 𝜈 the kinematic viscosity, S the
Force measurements had an absolute accuracy of ± 0.012
N at full-scale deflection [19] meaning that drag coefficient,
Cd , was measured with an uncertainty <  ± 0.0015 with a
95% confidence. No attempt was made to subtract the drag
generated by the mount and, as part of the CFD validation
exercise, the mount was also included in the CFD model.
However, the area used in the calculations was the projected
frontal area of the cyclist only (A = 0.0705 ­mm2).

2.2 Numerical

CFD predictions were performed using Siemens STAR-

CCM + with the methodology based on proven techniques
described by Blocken et al. [6–8]. The domain and the bound-
ary conditions were set to replicate the wind tunnel experiment
(including the mount). All surfaces were modeled as smooth
non-slip walls. The surface cell length was ~ 1 mm and to
resolve the boundary layer on the cyclist, 30 prism layers with
a total thickness of 15 mm and a growth rate of 1.1 were used
to give a wall y + value < 1. Default values of curvature and
surface proximity functions were used. On the tunnel walls, 7
prism layers (thickness 40 mm, growth rate 1.3) were used to
ensure a y +  > 30–60 and the use of a standard wall function. Fig. 4  Computational mesh around the cyclist
Practical computational fluid dynamic predictions of a cyclist in a time trial position Page 5 of 9 34

modulus of the strain rate tensor, and Re𝜃c the Reynolds

number based on the critical momentum thickness, 𝜃c , and
is obtained from a correlation [21].
For a RANS solution, convergence should be determined
by monitoring the drag force. However, as the flow is inher-
ently unsteady, this fluctuates, and convergence was instead
determined when the rolling mean was stationary (typically
requiring 5000 iterations). For the uRANS, temporal discre-
tization was achieved using a second-order implicit scheme
with the time-step set to achieve a Courant–Friedrichs–Lewy
(CFL) number below one in the refined wake region sur-
rounding the cyclist [10]. The CFL number (UΔt∕Δx )
relates the time-step size (Δt ) to the cell size (Δx ) and veloc-
ity (U ). This gave a Δt of 0.0005–0.0015s depending on Re.
However, in the prism layers, it was not always possible to
maintain CFL < 1 and values of up to 10 were observed.
Calculations were initialized at the mean inlet velocity and
run to obtain a stationary mean drag force (1s) and then Fig. 5  Measured and predicted drag coefficient, Cd , versus Reynolds
sampled for a further 2s. number, Re (Experimental Cd uncertainty ± 0.0015 with 95% confi-
All simulations were performed on a Linux-based desk- dence)
top workstation with 20 Intel Xeon 2.4 GHz processors and
128 GB RAM. Both RANS and uRANS solutions took a
similar time to compute (due to the need to average the and a similar gradient is observed in all cases. However,
RANS over 5000 iterations) requiring ~ 900 CPU hours apart from at the highest Re., the absolute Cd is generally
(~ 2 days). On the same hardware, it is estimated that LES underpredicted by up to 10% depending on the choice of
would take over 100 days. model coefficient, Conset . Equation (7) suggests that lower
values of Conset should lead to earlier transition, and therefore
lower drag. Yet, as Conset is increased, the predicted drag
3 Results rises, approaching the measured value. This is due to the
model implementation [21] which employs min/max argu-
Measured and predicted Cd values are plotted against Re in ments to maximize the γ production term. Overall, Conset =
Fig. 5 showing (as expected) a notable dependency, reduc- 2.2, the default value, gives the best results although the Cd is
ing almost linearly. Clearly, boundary layer transition effects still underpredicted. At a Re equivalent to 50 km/h, the pre-
are in play, but a drag minimum (as seen in Fig. 2) is not dicted Cd (0.757) is 3.8% lower than that measured (0.787).
observed despite Re reaching 8 × ­105. A linear relationship The Re dependency is equally captured by the uRANS and
is surprising if the cyclist is considered as a combination of the predicted Cd (0.790) is now only 0.4% higher than that
several cylinders. Brownlie et al. [22] also reported no drag measured (0.787). Given the Re dependency trend is repro-
crisis (but a moderate drag reduction) when increasing Re duced, it is the authors’ opinion that this is an acceptable
for a system of three connected cylinders resembling the level of accuracy.
body and arms of a cyclist. This is a clear example of where Boundary layer transition is indicated by the intermit-
experimental data, despite being numerically accurate, does tency, as plotted in Fig. 6, and this reveals differing behav-
not provide clear understanding. ior over different body parts. For example, on the helmet,
which is exposed to undisturbed air, transition occurs earlier
at higher Re (as expected), delaying separation. This is cor-
4 Discussion roborated by the higher values of surface shear stress where
reduction to a zero value indicates separation. At lower Re,
4.1 Validation of the CFD and transitional effects the right calf shows laminar separation, turbulent reattach-
ment, and formation of a laminar separation bubble. At the
The fact that boundary layer transitional effects are key higher Re, transition is almost immediate, and separation is
is confirmed by the CFD. Without a transition model, the delayed. Even at the highest Re, the local calf Re is lower
“fully turbulent” RANS shows almost no Re dependency. Cd than the typical critical value (Fig. 2) which suggests inter-
remains almost constant at the low drag, high Re value (Cd action with the turbulent wake of the upstream left leg is
~0.72). With the γ-transition model, Re dependency returns encouraging early transition.
34 Page 6 of 9 M. Taylor et al.

Fig. 6  Mean intermittency and wall shear stress at different Reynolds numbers (uRANS, Conset = 2.2)

A breakdown of the drag force (Fig. 7) also shows that

there are non-linear changes in the drag contributions from
various body parts as Re changes. The legs account for just
under 50% of the drag, the body ~ 30%, the arms ~ 18%, and
the head and helmet ~ 2%. As Re increases, the general split
remains similar but subtle and differing changes of 1–2% for
each body part confirm a more complex interaction of local
Re and local flow features. However, this raises an important
point in that the legs are stationary in these simulations and
these data suggest that the dynamic nature of the legs could
play a key role in the generation of drag. Using a mannequin
with moveable legs, Crouch et al. [23] showed that direct
force-balance measurements of different static leg positions
varied by ~ 15% between the maximum at 75° and 255° and
minimums between 0–15° and 180–195° (where 0° and 180°
represented feet at the top and bottom of the pedal stroke). Fig. 7  Breakdown of the drag force at different Reynolds numbers
For moving legs, they found that pedal stroke average drag (uRANS, Conset = 2.2, drag of the mount omitted)
Practical computational fluid dynamic predictions of a cyclist in a time trial position Page 7 of 9 34

matched the average of the static positions, within experi- quantify but clearly these changes affect the overall drag.
mental uncertainty, but there were some distinct differences Changes to the flow are more easily observed in data plot-
in drag at specific pedal stoke locations. This is certainly an ted on planes. Figure 9 shows streamlines and streamwise
area for future investigation with CFD, but it would require velocity contours on a plane through the centerline of the
much higher computational resource taking it outside the cyclist for both RANS and uRANS. The contours confirm
scope of this paper. that the wake structure is a combination of flow from the
head, arms, body, legs (and in this validation case, the cylin-
4.2 Effect of Re on flow topology and Cd drical mount). The main differences with increasing Re can
be seen as a general reduction in the size of the wakes cor-
Isosurfaces of Q-criterion can be used to visualization of responding to a reducing Cd . This can be seen over the top
coherent vortical structures and, using this, Fig. 8 quali- of the body where the flow remains attached longer over the
tatively shows that the CFD predicts the same complex rear of the cyclist and the wake structure generated by the
topology reported by Crouch et al. [9] and Terra et al. [16] body is notably reduced at the higher Re. Comparing RANS
(Fig. 1). and uRANS shows that these trends are replicated but in
Several vortical structures are generated by the hands, general uRANS-predicted wakes are marginally larger (in
arms, and head which then combine with the flow over the line with the larger predicted Cd).
body and legs to create the complex wake structure. Subtle The mid-plane provides limited information on the flow
changes in this for different Re are difficult to observe and generated around the legs, but more detail is revealed in
Fig. 10 which shows streamlines and isosurfaces of zero
streamwise velocity. The wake generated by each body part
is observed to reduce with increasing Re (thereby reduc-
ing Cd ). This is evidenced by the smaller separated regions
depicted by the isosurfaces. Considering the legs, which are
responsible for about half of the drag, the right leg (which
is further back) has a large region of separation from over
the hip and thigh region which can be visibly seen to shrink
with increasing Re. Indeed, the separation on the right foot
Fig. 8  Example isosurface of non-dimensional Q-criterion appears to disappear completely. For the left leg, there is still
(QA∕U∞ 2
= 7) at Re = 5.45 × 10.5 a separation from the hip, but not the thigh, and a separation

Fig. 9  Mid-plane streamlines and streamwise velocity contours (Conset = 2.2)

34 Page 8 of 9 M. Taylor et al.

Fig. 10  Planes of streamwise velocity and isosurfaces of zero streamwise velocity at different Re (uRANS, Conset = 2.2)

can now be seen along the calf. Again, these reduce in size, The data show that the legs contribute ~ 50% of the drag, the
reducing their Cd with increasing Re. Interestingly, however, body ~ 30%, the arms ~ 18%, and the head ~ 2%. However, as
with reference to Fig. 7, the contribution to the drag of each Reynolds number increases, the data reveal changes in the
leg switches. At the lower Re, the left leg produces 25% of relative contributions that confirm a more complex inter-
the drag force and the right 23%, but, at the higher Re, these action of local Reynolds number and local flow features.
change to 24% and 28%, respectively. This is further evi- Examination of the predicted flow field revealed variations
dence that the system is complex, non-linear, and contains in boundary layer transition, separation, and wake formation
interacting aerodynamic features. As mentioned earlier, it from each body part which combine in a complex system to
also alludes to the role that the dynamic nature of the legs produce the changes in drag coefficient with Reynolds num-
could play in the generation of drag. ber. Overall, the data confirm that RANS CFD can predict
trends in aerodynamic drag using a desktop PC. The numeri-
cal methodology has been validated, and thus provides a
5 Conclusion practical time- and cost-effective tool for further examina-
tion of drag reduction within cycling.
Wind tunnel tests of a rider model (with the bicycle omit-
ted) were conducted at Reynolds numbers equivalent to rider
Funding This work received no external funding and was completed
speeds of ~ 20–70 km/h (2 × ­105 < Re < 8 × ­105). Omission in support of a Masters’ Degree in Aeronautical Engineering at Lough-
of the bicycle, while a limitation of the study, was deemed borough University.
acceptable in the context of achieving practical CFD predic-
tions. The measured drag data showed a notable Reynolds Data availability The authors do not have permission to share data.
number dependency, with the drag reducing almost linearly
by ~ 20% from 0.88 to 0.71. Steady RANS and unsteady
Declarations
(u)RANS CFD predictions produced similar results. They Conflict of interests The authors declare they have no conflicts of in-
were both able to accurately capture the Reynolds num- terest.
ber dependency, but only when employing the γ-transition
Open Access This article is licensed under a Creative Commons Attri-
boundary layer model. Unsteady (u)RANS CFD performed bution 4.0 International License, which permits use, sharing, adapta-
better and predicted drag within 2% of the measured value tion, distribution and reproduction in any medium or format, as long
(across the range of Reynolds numbers). The steady RANS as you give appropriate credit to the original author(s) and the source,
CFD underpredicted the drag by approximately 3% with provide a link to the Creative Commons licence, and indicate if changes
were made. The images or other third party material in this article are
the unsteady (u)RANS CFD closer and within 2% of the included in the article’s Creative Commons licence, unless indicated
measured value (across the range of Reynolds numbers).
Practical computational fluid dynamic predictions of a cyclist in a time trial position Page 9 of 9 34

otherwise in a credit line to the material. If material is not included in 12. Javadi A, Buckrell AJM, and Peterson SD (2020) Improving
the article’s Creative Commons licence and your intended use is not Numerical Estimation of Cyclist Drag Area in Static Conditions
permitted by statutory regulation or exceeds the permitted use, you will Using Unsteady RANS, Proceedings 49(1):139, [Link]
need to obtain permission directly from the copyright holder. To view a 10.​3390/​proce​eding​s2020​049139
copy of this licence, visit [Link] 13. Defraeye T, Blocken B, Koninckx E, Hespel P, Carmeliet J (2010)
Aerodynamic study of different cyclist positions: CFD analysis
and full-scale wind-tunnel tests. J Biomech 43(7):1262–1268.
[Link]
14. Giljarhus KET, Liland FF, Oggiano L (2023) Virtual skeleton
References methodology for athlete posture modification in CFD simulations.
Sports Eng 26(1):39. [Link] oi.o​ rg/1​ 0.1​ 007/s​ 12283-0​ 23-0​ 0430-8
1. Kyle CR, Burke E (1984) Improving the racing bicycle. Mech Eng 15. Giljarhus KET, Stave DÅ, Oggiano L (2020) Investigation of
106(9):34–45 influence of adjustments in cyclist arm position on aerodynamic
2. Crouch TN, Burton D, LaBry ZA, Blair KB (2017) Riding against drag using computational fluid dynamics. Proceedings 49(1):159.
the wind: a review of competition cycling aerodynamics. Sports [Link]
Eng 20:81–110. [Link] 16. Crouch TN, Burton D, Brown NAT, Thompson MC, Sheridan
3. Malizia F, Blocken B (2020) Bicycle aerodynamics: History, J (2014) Flow topology in the wake of a cyclist and its effect
state-of-the-art and future perspectives. J Wind Eng Ind Aerodyn on aerodynamic drag. J Fluid Mech 748:5–35. [Link]
200:104134. [Link] 1017/​jfm.​2013.​678
4. Malizia F, Blocken B (2021) Cyclist aerodynamics through time: 17. Achenbach E (1971) Influence of surface roughness on the flow
Better, faster, stronger. J Wind Eng Ind Aerodyn 214:104673. through a staggered tube bank. Wärme-und Stoffübertragung
[Link] 2:120–126
5. Terra W, Sciacchitano A, Scarano F (2020) Cyclist Reynolds num- 18. Achenbach E, Heinecke E (1981) On vortex shedding from
ber effects and drag crisis distribution. J Wind Eng Ind Aerodyn smooth and rough cylinders in the range of Reynolds numbers
200:104143. [Link] 6× ­103 to 5× 10. J Fluid Mech 109:239–251
6. Blocken B, van Druenen T, Toparlar Y, Andrianne T (2018) Aero- 19. Johl G, (2010) The design and performance of a 1.9 mx 1.3 m
dynamic analysis of different cyclist hill descent positions. J Wind indraft wind tunnel, PhD Thesis, Loughborough University, UK.
Eng Ind Aerodyn 181:27–45. [Link] oi.o​ rg/1​ 0.1​ 016/j.j​ weia.2​ 018.​ 20. Franke J, Hellsten A, Schlünzen H, and Carissimo B (2007)
08.​010 Best practice guideline for the CFD simulation of flows in the
7. Blocken B, Toparlar Y, van Druenen T, Andrianne T (2018) Aero- urban environment. 11th Conference on Harmonisation within
dynamic drag in cycling team time trials. J Wind Eng Ind Aerodyn Atmospheric Dispersion Modelling for Regulatory Purposes,
182:128–145. [Link] Cambridge, UK, July 2007. Cambridge Environmental Research
8. Blocken B, van Druene T, Toparlar Y, Andrianne T (2019) CFD Consultants.
analysis of an exceptional cyclist sprint position. Sports Eng 21. Siemens Industries Digital Software (2023), Simcenter STAR-
22:10. [Link] CCM+ User Guide v2306, Siemens 2023.
9. Fintelman DM, Hemida H, Sterling M, Li X (2015) CFD simula- 22. Brownlie LW, Gartshore I, Chapman A, Banister EW (1991) The
tions of the flow around a cyclist subjected to crosswinds. J Wind aerodynamics of cycling apparel. Cycl Sci 3(3–4):44–50
Eng Ind Aerodyn 144:31–41. [Link] oi.o​ rg/1​ 0.1​ 016/j.j​ weia.2​ 015.​ 23. Crouch TN, Burton D, Thompson MC, Brown NAT, Sheridan
05.​009 J (2016) Dynamic leg-motion and its effect on the aerodynamic
10. van Druenen T, Blocken B (2024) CFD simulations of cyclist performance of cyclists. J Fluids Struct 65:121–137. [Link] oi.​
aerodynamics: impact of computational parameters. J Wind Eng org/​10.​1016/j.​jflui​dstru​cts.​2016.​05.​007
Ind Aerodyn 29:105714. [Link]
105714 Publisher's Note Springer Nature remains neutral with regard to
11. Javadi A (2022) Aerodynamic study of the pedalling of a jurisdictional claims in published maps and institutional affiliations.
cyclist with a transitional hybrid rans–les turbulence model.
Flow Turbul Combust 108:717–738. [Link]
s10494-​021-​00297-4