2020 AE - Typical Hydrodynamic Model For Netting
2020 AE - Typical Hydrodynamic Model For Netting
2020 AE - Typical Hydrodynamic Model For Netting
Aquacultural Engineering
journal homepage: www.elsevier.com/locate/aque
Keywords: The net is regarded as the most critical component in marine aquaculture facilities as it is the only barrier which
Aquaculture net protects the environment from fish escapes. Accurate predictions of the net cage deformation and drag force on
Hydrodynamic model the nets are needed, both for ensuring fish welfare and for dimensioning of the mooring system. Thus, an
Morison model appropriate hydrodynamic model is essential. In practice, two types of hydrodynamic force models, i.e., the
Screen model
Morison type and the Screen type, are commonly used to calculate the hydrodynamic forces on nets. Application
Wake effect
of the models depends on the underlying structural model and the availability of data. A systematic review of
hydrodynamic models is therefore undertaken to compare the models and various parameterisations, in aid of
model selection during the design. In this study, eleven commonly used hydrodynamic models, i.e., five Morison
models and six Screen models, are reviewed comprehensively, and implemented into a general finite element
(FE) solver for dynamic simulations. Sensitivity studies on different current velocities, inflow angles and so-
lidities of the nets are carried out. Moreover, different wake effects are also considered in numerical simulations.
The numerical results from different models are compared against existing experimental data under pure current
conditions. Suggestions for selection of suitable hydrodynamic models are provided, based on the model com-
parison.
1. Introduction loads on nets account for more than 85 % of total loads on a conven-
tional fish cage (Cheng, 2017), accurate predictions on the hydro-
According to the Food and Agriculture Organization of the United dynamic responses of nets are essential in the structural design.
Nations, aquaculture has been the world’s fastest-growing food pro- The hydrodynamic forces on aquaculture nets depend on both the
duction method in the past 40 years (FAO, 2018). The development of experienced current velocity and the hydrodynamic characteristics of
high-value seafood such as Atlantic Salmon (Salmo salar) and Rainbow the net. The hydrodynamic characteristics depend on the material,
Trout (Oncorhynchus mykiss) has led to significant investments in the geometrical parameters (including mesh shape, mesh size, twine dia-
aquaculture industry. According to the Norwegian Seafood Research meter), and production methods, i.e., twine weaving method (twisted
Fund, the seafood industry has invested more than NOK 115 billion in or braided) and net weaving method (knotless or knotted). As shown in
Norway since 2000 (Blomgre et al., 2019). The investment in the Fig. 1, the four nets which are commonly used in marine aquaculture
aquaculture industry upgraded conventional farming facilities and have different hydrodynamic characteristics due to their different ma-
generated novel aquaculture structures such as Ocean Farm 1 and terials and production methods. Different twine materials and twine
Havfarm. These innovative structures are designed to operate in the weaving methods make the surface roughness different. Higher surface
open sea and aiming to minimise the environmental impacts of aqua- roughness will generate larger turbulence regions; and thus, higher
culture. drag force (Balash et al., 2009). According to experimental data from
Moving aquaculture to the open sea can benefit the fish welfare and Tsukrov et al. (2011), copper nets (smooth) exhibit significantly lower
the ecosystem through better water exchange and dispersal of waste drag resistance in steady currents than nylon nets (rough) of the similar
over a larger area (Cardia and Lovatelli, 2015). The more exposed solidity. The experimental data from Lader et al. (2014) indicates that
setting implies larger waves and currents, which can increase the en- the drag force on the knotted net is up to 10 % higher than that of the
vironmental loads on aquaculture structures. As the environmental knotless net, given the same environmental condition. Thus, in order to
⁎
Corresponding author.
E-mail address: hui.cheng@uis.no (H. Cheng).
https://doi.org/10.1016/j.aquaeng.2020.102070
Received 27 November 2019; Received in revised form 9 March 2020; Accepted 10 March 2020
Available online 17 March 2020
0144-8609/ © 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
predict the hydrodynamic forces precisely, all the related parameters Solidity is the other key parameter for hydrodynamic character-
should be considered in calculations. istics. For a net panel, the drag force is mainly dependent on the value
However, considering all the related parameters complicates the of solidity without obvious effects of twine diameter and mesh size,
force prediction and make it impractical in numerical simulations. To which both define solidity itself (Klebert et al., 2013). By definition,
make it feasible, one has to focus on key parameters while ignoring the solidity is the ratio between the projected net area Ap (area of the dark
secondary parameters. Through a large number of experiments lines in Fig. 2(b)) and the total area of the net panel At (area of the
(Tsukrov et al., 2011; Tang et al., 2018; Lader et al., 2014; Balash et al., dashed box in Fig. 2(b)). For an ideal knotless square net, a mathe-
2009) researchers found that hydrodynamic characteristics are mainly matical expression for Sn can be formulated as:
dependent on two dimensionless variables, Reynolds number (Re) and
solidity (Sn). The Reynolds number is defined as the following: d w (2L d w )
Sn =
L2 (2)
U dw
Re = where dw is the twine diameter, L is the half mesh size. Since knotless
(1)
nets are not always “mathematically perfect” and the solidity can
where U is the undisturbed fluid velocity, is the kinematic viscosity of change when the nets are submerged, some use a simplified expression
the fluid, dw is the twine diameter. For a typical aquaculture net, the for the solidity (Berstad et al., 2013):
Reynolds number is in the range of 100-10000. In some research 2d w
(Kristiansen and Faltinsen, 2012; Balash et al., 2009), the Reynolds Sn =
(3)
L
number is defined with local velocity. Models that define the Reynolds
number locally lead to increased Re, although the modelled flow field The deviation between Eqs. (2) and (3) is less than 10 % for a typical
does not change. In addition, the effect of the speed-up velocity on the net in aquaculture where the solidity is in the range of 0.1-0.4. These
hydrodynamic loads may need further investigations (Moe-Føre et al., two equations are obtained using the assumption of a perfect net geo-
2015). Thus, the Reynolds number in the present study is calculated metry. Besides, digital image processing (DIP) techniques can also be
based on Eq. (1). used to estimate the solidity of an aquaculture net. The net solidity is
Fig. 1. Different nets: (a) Knotless nylon net with rhombic mesh (Tang et al., 2018), (b) Knotted nylon net with rhombic mesh and Single English knot (Tang et al.,
2018), (c) Welded silicon-bronze net (Tsukrov et al., 2011) and (d) Knotless nylon net with square mesh.
Fig. 2. Definitions of half mesh size (L) and twine diameter (dw). (a) A sample knotless net (Kristiansen and Faltinsen, 2012). (b) An ideal knotless net.
2
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
processed as the ratio between pixels in different colours in the DIP booming of aquaculture, researchers have extended these models for
method (Yu, 2017). In general, the estimated solidities based on the DIP aquaculture nets. In general, the hydrodynamic forces on net panels in
evaluations are less than 4% discrepancy compared with Eq. (2) for an oscillatory flow can be written in Eq. (4).
typical aquaculture nets (Tsukrov et al., 2011). Thus, the solidity of the
u u v 1
net panel in the present study is calculated based on Eq. (2). F= + Ca Ca + Cd A|u v|(u v)
(4)
t t t 2
To acquire acceptable force predictions on aquaculture nets, re-
searchers have conducted considerable amounts of both experiments where is the fluid density, ∇ is the submerged volume of the net
and theoretical analyses. Based on the experimental results, hydro- structure, A is the reference area (the projected area of a net twine or a
dynamic models are proposed to calculate the forces on net panels or net panel), u is the fluid velocity vector, v is the structure velocity
net twines for numerical simulations. Some of them are implemented vector. The equation contains two empirical hydrodynamic coeffi-
into in-house codes or commercial software (Table 1) for the fish cage cients—the added mass coefficient, Ca and the drag coefficient, Cd . The
simulation. In general, the hydrodynamic models can be classified as coefficients are determined from experimental data and dependent on
two types, the Morison model and the Screen model. Tsukrov et al. the Keulegan–Carpenter number, Reynolds number, twine’s surface
(2000) applied the Morison model extensively from fishing gear mod- roughness and the solidity of the net (Kristiansen and Faltinsen, 2012).
elling to fish cage simulations and developed software, Aqua-FE. Lee The first three terms in the right side of Eq. (4) are Froud-Krylov force,
et al. (2005) proposed a Screen model and implemented it into soft- diffraction force and radiation force, respectively. Combining these
ware, MPSL, for numerical simulations of both fishing gear and fish three terms will get the inertial force, FI in Eq. (5). The last term in Eq.
cage. In SIMA, the hydrodynamic forces on the net panels are calculated (4) is also called viscous force. Thus, the total hydrodynamic forces on
based on the Screen model proposed by Løland (1991). From the the net are the sum of the inertial force and the viscous force. The in-
published articles (references in Table 1), the simulation results based ertial force can be rewritten to the following format:
on these software and codes agreed well with the experimental data u v
when the velocities were smaller than 0.5 m/s. However, the agreement FI = Cm (Cm 1)
t t (5)
with the experimental data is weaker when the velocities were higher
than 0.5 m/s and/or the solidities of the net panels were higher than where Cm is the inertia coefficient (Cm = Ca +1). For aquaculture nets,
0.35. a typical value of Cm = 2 is used for the inertia coefficient (Lader et al.,
Besides the hydrodynamic models for force prediction, the wake 2007). The inertial force is experienced either during the transition to
effect behind upstream net panels is also essential in the dynamic steady-state or when the net is experiencing wave loads. According to
analysis of fish cages. According to the previous research (Faltinsen and the experimental data and the dimensional analysis by Zhao et al.
Shen, 2018), the anchor force increases up to 22 % if the net-to-net (2008), the inertial force on commonly used nets is in the order of O [dw
wake effect was neglected in the numerical analysis. /H] times the viscous force, where the twine diameter dw is far smaller
The purpose of the present study is to summarise and compare the than the wave height H. Since the twine diameter dw is far smaller than
result of the commonly used hydrodynamic models under pure current the wave height H, the inertial force can be negligible compared to the
conditions. Reviews of eleven hydrodynamic models are given in viscous force, e.g. the inertial force is approximately 0.1 % of the vis-
Section 2. Then, the present program for fish cage simulation, which cous force when dw = 1 mm and H = 1 m. Thus, it is reasonable to ig-
includes the eleven hydrodynamic models, is introduced in Section 3. In nore the inertial force when one calculates the hydrodynamic forces on
Section 4, extensive validation cases are performed against experi- an aquaculture net. Due to this reason, the inertial force on aquaculture
mental results. Series of experiments conducted by Lee et al. (2005); nets is not discussed in this study.
Tang et al. (2018); Tsukrov et al. (2011), and Moe-Føre et al. (2016) are
reproduced by the program to validate the code and investigate the 2.1. Morison model
performances of different hydrodynamic models. Finally, the results of
this study are discussed with concluding remarks. In the Morison model, the forces on nets are calculated based on
individual twines and knots. The twines are taken as cylindrical ele-
2. Hydrodynamic models ments, and knots are taken as spheres. In practice, the viscous force (the
last term in Eq. (4)) is usually decomposed into the two components:
According to the summary in Table 1, there are two types of hy- normal drag force (Fn ) and tangential drag force (Ft ):
drodynamic models for aquaculture nets: Morison model and Screen 1
Fn = Cn Ld w |ur n | ur n
model. In Morison models, the forces on the net panel are treated as the 2 (6)
sum of forces on individual twines. In Screen models, the forces are
1
calculated based on considering the net as a panel. These models were Ft = Ct Ld w |ur t | ur t
2 (7)
initially proposed to calculate the hydrodynamic forces on fishing nets,
especially trawl nets. With the slowdown of the fishing industry and where L is the twine length, dw is the twine diameter, is the fluid
Table 1
A summary of the main numerical programs and codes can be used for dynamic analysis of fish cages.
Software or code Hydrodynamic model Structural model Reference
ANSYS Morison type Truss, pipe, beam (Cheng et al., 2018; Cui et al., 2014)
ABAQUS Morison type Truss, beam (Moe-Føre et al., 2016, 2015; Li et al., 2013)
Aqua-FE Morison type Truss (Shainee et al., 2013; Decew, 2011; DeCew et al., 2010; Tsukrov et al., 2003)
SIMA Screen type Truss (Li and Ong, 2017; Faltinsen and Shen, 2018; Li et al., 2018)
FhSim Morison type / Screen type Triangles / (Reite et al., 2014; Endresen et al., 2013)
Mass-spring
AquaSim Morison type Truss (Berstad and Aarsnes, 2018; Berstad and Heimstad, 2017; Reichert, 1994)
Orcaflex Morison type Mass-spring (Cifuentes and Kim, 2017b; Cifuentes and Kim, 2017a; Cifuentes and Kim, 2015)
ProteusDS Morison type Mass-spring (Turner et al., 2017)
MPSL Screen type Mass-spring (Lee et al., 2005)
NaLA Morison type Mass-spring (Takagi et al., 2004; Tsukrov et al., 2003)
3
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Table 2
Hydrodynamic force coefficients in Morison models for twines, when 100 < Re < 10 000.
Model Cn Ct Reference
4
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 4. Normal drag coefficients (Cn ) versus Reynolds number (Re) according to different hydrodynamic models. Because Cn in M5 model changes with different
solidities (Sn), the filled area represents its variation range for its applicable solidities (0.172 < Sn < 0.208). The normal drag coefficient of the twine in S6 is a
polynomial function according to its formulae (Kristiansen et al. 2012). The normal drag coefficients for a smooth cylinder and the typical Re region (100-10 000) for
twines are also shown in the figure.
applied in MPSL (Table 1) did not disclose its formulation in the pub-
lished article (Lee et al., 2005), the coefficients are assumed linearly
proportional to Sn and independent of Re.
As shown in Fig. 6, the values of CD decay as expected with in-
creasing inflow angle. The first two models (S1 and S2) have a similar
shape and follow the dashed line, which indicates their CD decay fol-
lowing the cosine function. The values of CD in S3, S4 and S6 decay
faster than the cosine function with the increasing inflow angle. Ac-
cording to the expressions of S3 and S6, the drag coefficient is zero
when θ = 90˚. It means that the drag force is zero when the flow is
parallel to a net panel, which is incorrect as there must be a drag force,
although very small. If one used these two models to design a squared
fish cage, the drag force can be underestimated when half of the nets
are parallel to the ambient flow. Among the six Screen model, only the
Fig. 5. Illustration of the hydrodynamic forces on a net panel. FR is the re- value of CD in S5 decays slower than the cosine function with the in-
sultant force which can be decomposed to drag force FD and lift force FL , or creasing inflow angle. A detailed discussion on the drag force on a net
normal drag force FN and tangential drag force FT . The inflow angle θ of a net panel with experimental data is given in Section 4.2.2.
panel is the angle between the normal vector of the net panel and the current According to Fig. 7, the values of CL first increase, and then decrease
direction (θ=π/2- α). with the increasing inflow angle, similar to an airfoil. The value of CL is
zero when the flow is parallel or perpendicular to the net panel. The
where At is the area of the net panel element, ur is the fluid velocity curves of S1 and S2 are similar to the sine function of 2 . While for S3,
relative to the net panel, iD and iL are unit force vectors to indicate the S4 and S5, the crests of curves are located between 30° and 45°. Com-
directions of drag and lift forces. Procedures to calculate At and unit pared to the values of CD, the values of CL is relatively small when
force vectors (iD and iL ) are presented in Appendix A. CD and CL are the θ < 30˚. It means the drag force is the dominated force when the inflow
drag and lift force coefficients, respectively. The coefficients are de- angle is small. However, the ratio between lift and drag forces is
termined from experimental data and depend on the Reynolds number, changing with different inflow angles. The appropriate hydrodynamic
solidity and inflow angle (Fig. 5). The relationships of FN , FT , FD , FL model should represent the observed ratio of lift/drag in experiments.
and the inflow angle (θ) are shown in Fig. 5. The relationships of CD , CL , Solidity is an essential parameter in Screen models. Through a large
CN , CT are given in Eqs. (12) and (13). number of experiments (Klebert et al., 2013; Zhou et al., 2015; Tang
et al., 2018), researchers found that the hydrodynamic coefficients are
CD = CN cos cos2 + CT sin sin2 (12) highly dependent on the solidity of the net panel. Thus, all the Screen
models in this paper take the solidity, Sn, into their expressions (the
CL = CN sin cos2 CT cos sin2 (13)
detailed expressions and its applicable region are given in Appendix C).
Six Screen models which are the most commonly used in fish cage In general, the values of CD and CL increase with increasing solidity,
simulations are compared in this section and CN and CT are converted to which indicates that nets with larger solidity have higher hydro-
CD and CL using Eqs. (12) and (13) when required. Figs. 6 and 7 present dynamic forces when the other conditions are the same. Fig. 8 shows
the drag and lift force coefficients of the six Screen models within the the drag coefficients of knotless nylon net panels when θ = 0° with
applicable solidity range of corresponding models. S6 (Balash et al., different solidities from the available experimental data (Zhou et al.,
2009) is not shown in Fig. 7 since no expression for the lift coefficient is 2015; Tsukrov et al., 2011; Gansel et al., 2015). The regression curves
given. The Reynolds number is assumed as 1000 for S3, S4 and in the figure are fitted using the ordinary least squares methods. The
S6 models in Figs. 6 and 7. For S3, the harmonic terms (a3 and b4) coefficients of determination (R2 ) show that the cubic regression fits the
should increase with the increasing solidity, but no quantitative re- data better than the simple linear regression. This observation complies
lationship is given in Kristiansen and Faltinsen (2012). Thus, the har- with the expressions in S1 (Aarsnes et al., 1990) and S2 (Løland, 1991).
monic terms are set according to the experimental data reproduced by The flow patterns around nets should change with the Reynolds
Shimizu et al. (2018), whereas Sn = 0.29 and a3 = 0.15. Since S5 numbers, and thus influence the hydrodynamic coefficients. The
5
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 6. Drag force coefficients in different Screen models. The dashed lines in each subplot are CD ( = 0°) cos .
hydrodynamic coefficients in S1, S2 and S5 are constant with changing Screen models are seldom used in the commercial FE solvers (see
Re, as they do not include the Reynolds number in the expressions. Table 1) for fish cage simulations, due to the complexity of im-
While in S3, S4 and S6, the hydrodynamic coefficients are changing plementation. The structural solver usually calculates the motion and
with different Reynolds numbers. According to Fig. 1, the effect of Re deformation of aquaculture nets based on the line-type elements (bar,
might be negligible since the drag coefficient of a net twine is almost pipe or beam). In order to implement Screen models into the existing FE
unchanged when 100 < Re < 10 000. A detailed comparison of the solver, other types of element (shell or plane) must be introduced to
drag force on net panels with experimental data under different Rey- calculate the hydrodynamic forces and extra steps are required to map
nolds numbers is presented in Section 4.2.1. the hydrodynamic forces to the line-type structural elements. From a
Fig. 7. Lift force coefficients for different Screen models. The dashed lines in each subplot are max (CL) sin2 . S6 is omited due to the lack of formulas for the lift
coefficient.
6
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 9. Illustration of different wake effects. (1) Twine-to-twine wake effect, where a grid of i+1 cylinders (cross-section of a net panel) are exposed to an incident
current velocity U. The Ui (i = 0, 1 …) denotes the velocity experienced by cylinder i, which is modified due to the presence of upstream cylinders. (2) Net-to-net
wake effect, where the upstream (left) net panel is exposed to an incoming current velocity U. The net-to-net wake effects from the upstream net panel result in a
reduced flow (rU) at the downstream net. (3) Cage-to-cage wake effect, where the incoming flow for the downstream (right) fish cage is anisotropic and might be
smaller than the incoming flow for the upstream (left) fish cage.
7
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
external and internal forces of the structure and neglects the fluid
mechanics. Therefore, the hydrodynamic forces on the downstream
nets can be overestimated if no special precautions are taken to include
the wake effect. In general, there are two ways to include the wake
effect in the fish cage simulation. One can couple a structural solver
with a fluid solver to include the fluid-structure interaction (Chen and
Christensen, 2017). Alternatively, one can use the quasi-static as-
sumption to “register” a wake region in the structural solver (Endresen
et al., 2013). For the latter method, the wake effect for a conventional
fish farm can be subdivided into three scales (Fig. 9) in the FE solver for
implementation: (1) twine-to-twine wake effect; (2) net-to-net wake
effect; (3) cage-to-cage wake effect.
Fig. 10. Illustration of the method to identify the nets which experience the net-
2.3.1. Twine-to-twine wake effect to-net wake effect caused by upstream nets in a cylindrical fish cage. The fish
The twine-to-twine wake effect represents the interactions between cage is shown from the top, and the blue part is the rear half of a cage where the
net twines (the influence region is in the order of centimetres). In a net nets will experience a reduced flow. The inflow angle θ of a net panel is the
panel, the velocity of the downstream twine is smaller than that of the angle between the normal vector of the net panel and the incoming flow. (For
upstream twine when the inflow angle of the net plane is larger than interpretation of the references to colour in this figure legend, the reader is
70°. According to (Endresen et al., 2013), when the inflow angle is 90°, referred to the web version of this article).
the drag force on a net panel without twine-to-twine wake effect can be
maximumly eight times larger than that with twine-to-twine wake ef- blue part in Fig. 10). Table 3 shows flow reduction factors from theo-
fect. retical analyses and experimental results. According to Table 3, the flow
The Morison model has a natural drawback on the implementation reduction factor should be a function of Reynold number, solidity and
of twine-to-twine wake effect. To include this effect in the Morison inflow angle. The most commonly used flow reduction factor, r = 1-
model, one needs to make a function to describe the flow pattern be- 0.46 CD ( = 0 ) , is consistent with different inflow angles. That means all
hind a cylinder. For example, the wake shape around a 2D circular the downstream nets in the rear half of a fish cage experience the same
cylinder in an infinite fluid can be calculated based on Blevins formula reduced current velocity, which is unphysical and contrary to the ex-
(Eq. (14)): perimental results reported by Bi et al. (2013).
Fig. 11 shows the flow reduction factor with respect to the inflow
Cd (y / d w )2
Udownstream = Uupstream (1 1.02 exp ) angle for downstream net panels in a cylindrical fish cage. In this figure,
6 + x /d w 0.0767Cd (6 + x / d w )
the data from experiment by Bi et al. (2013) are presented and com-
(14) pared with theoretical values. The experimental data indicate that the
where Udownstream is the velocity for the downstream cylinder at co- flow reduction factor should reduce with increasing inflow angle.
ordinate (x, y). In Morison models, the wake effect experienced by a However, only the result from Endresen et al. (2013), where the flow
single twine depends on contributions from all other twines in the reduction factor depends on the inflow angle, agrees with the trend.
model. Excessive numerical calculations and sophisticated algorithms The other three methods in which the flow reduction factors are cal-
are required to determine the spatial relationships among a large culated by r = 1-0.46 CD ( = 0 ) , give constant values with different inflow
number of twines in the numerical model. On the other hand, the twine- angles and disagree with the experimental trend. The discrepancy in the
to-twine effect is naturally included in Screen models, since the hy- flow reduction factors can influence the hydrodynamic forces on
drodynamic coefficients of net panels consider the interactions between downstream net panels for dynamic simulations of fish cages.
twines implicitly. A detail demonstration on the twine-to-twine wake Fig. 12 shows the equivalent drag coefficients (r 2CD ) of the down-
effect is shown in Section 4.2.2. stream net panels with Sn = 0.243. Because the drag force is propor-
tional to the square of the ambient velocity, r 2CD can be used to re-
present the equivalent drag coefficient of a net panel in the wake. In
2.3.2. Net-to-net wake effect
this figure, the drag coefficients (CD ) are calculated based on the
The net-to-net wake effect is used to represent the interaction be-
S1 model for both curves. For the dashed line, r is calculated according
tween nets inside a single fish cage (the influence region is in the order
to Endresen et al. (2013). For the solid line, r is calculated by 1-0.46
of tens of metres). Approximately half of the nets in a cylindrical fish
CD ( = 0 ) according to Aarsnes et al. (1990). As shown in this figure, the
cage will experience the net-to-net wake effect. The mooring force can
equivalent drag coefficients of downstream net panels based on the two
be overestimated up to 22 % if the net-to-net wake effect is neglected in
flow reduction factors are similar when θ < 30°. However, with in-
fish cage dynamic analyses (Faltinsen and Shen, 2018). In practice, a
creasing inflow angle, the equivalent drag coefficient based on the
flow reduction factor (r) is adopted in software and codes to represent
constant flow reduction factor is larger than the one based on the
the net-to-net wake effect. Eq. (15) is a typical expression for the net-to-
variable flow reduction factor. It means around 2/3 of the downstream
net wake effect, where r is the flow reduction factor (0 < r < 1), U is
net panels in a cylindrical fish cage contribute larger hydrodynamic
the ambient current velocity. According to this equation, the down-
forces if the constant flow reduction factor is applied. A detailed study
stream nets experience smaller current velocity compared to the up-
of the two forms of flow reduction factors is given in Section 4.3.
stream nets.
Udownstream = rUupstream (15)
2.3.3. Cage-to-cage wake effect
In numerical simulations, whether a net is located in the wake can The cage-to-cage wake effect is used to represent the interaction
be determined based on its position, the centre of the fish cage and the between cages (the influence region is in the order of a few hundred
incoming current direction (see Fig. 10). The flow reduction factor can metres). In the marine aquaculture industry, fish cages are usually
be set as an attribute of the downstream nets to reduce the ambient grouped in arrays as a fish farm. Due to the block effect of the upstream
velocity numerically. cage, the flow for the downstream cage is affected by the existence of
An accurate flow reduction factor is critical for predicting the hy- the upstream cage. In the previous study, a theoretical expression for
drodynamic forces on and the deformation of the nets in the wake (the the velocity reduction behind a net panel (r = 1-0.46 CD , r is the
8
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Table 3
Comparison of flow reduction factors (r ).
Flow reduction factors Sn Re Reference
3. Numerical method
¨
Mx + Kx = Fs + Fh (16)
9
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
3.2. Structural element the element’s shape function (Stengel and Mehdianpour, 2014).
10
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
where λ is the ratio between the half mesh size of the numerical net
(Ln) and the half mesh size of the physical net (Lp).
In this section, four experiments (Lee et al., 2005; Tang et al., 2018;
Tsukrov et al., 2011; Moe-Føre et al., 2016) are selected to validate the
newly developed program and discuss the aforementioned hydro-
dynamic models. The first case demonstrates the feasibility of the FE
solver through a comparison with an experiment in Section 4.1. The
second case compares the discrepancies among the different hydro-
dynamic models for net panels with respect to different current velo-
Fig. 14. Comparison of the net shape between the experimental and the nu- cities, inflow angles and solidities. Based on the comparison in Section
merical results. 4.2, the appropriate hydrodynamic model is selected for the fish cage
simulations. In Section 4.3, a comparative study on wake effects is
carried out based on fish cages under pure current conditions by using
3.3.1. The hydrodynamic element for Morison models
the verified solver.
The hydrodynamic element for the Morison models is a one-di-
mensional line. Since they are compatible with the existing finite ele-
ment codes, the mesh for the hydrodynamic model is inherited directly
4.1. Feasibility of the structural solver
from the structural model. In the external module, the hydrodynamic
forces on net structure are first calculated by Eqs. (6) and (7), and then
As it is the first time that the open-source FE solver, Code_Aster, is
mapped to their corresponding nodes in the structural elements.
used to simulate aquaculture nets, the feasibility of the solver should be
assessed at the very beginning. In order to assess the FE solver itself, the
3.3.2. The hydrodynamic element for Screen models external module (highlighted by the red dashed box in Fig. 13) and the
It is difficult to integrate Screen models to a general FE solver (Bore mesh grouping method are not applied to the numerical model for
et al., 2017), as there is no direct way to build the “hydrodynamic panel avoiding interferences.
element” from the nodes and line-elements in the structural model. In The feasibility is assessed through (1) examining iterative con-
this study, an automatic meshing function was developed to generate vergence and (2) examining consistency in the solution. The numerical
the two-dimensional virtual panel elements for Screen models. The model is set up based on the experiment by Lee et al. (2005). In the
functionalities in the automatic meshing module included: (1) create experiment, the net is 12 × 18 meshes in squared shape with twine
the “hydrodynamic panel element” automatically based on the position diameter dw = 0.4 mm and half mesh size L = 100 mm. The Young’s
of nodes; (2) calculate the normal vector and the area of the “hydro- modulus of the twine is 119.37 MPa. The four corners are fixed, and
dynamic panel element”; (3) calculate the hydrodynamic forces on the three sinkers, whose masses are 0.5 kg, 1.5 kg and 0.7 kg from left to
net panels based on the formulas of the different Screen models and right in Fig. 14(a), are hung in the middle of the net. In the numerical
map the forces to the nodes in corresponding structural elements. simulation, the characteristics and configuration of the net are the same
as those in the experiment. The net is modelled by 424 elements and
3.4. Mesh grouping method 213 nodes. The three hung sinkers are represented by three vertical
concentrated forces, which are 5 N, 15 N and 7 N from left to right. The
In a full-scale fish cage, the cage net is composed of thousands of density of the twine is assigned 1125 kg/m3 by assuming the material is
small twines. It is impractical to build a numerical model twine by Nylon (Moe et al., 2010). The Newton-Raphson method is adopted to
twine. A mesh grouping method is used in Section 4.3 to reduce the solve the FE equations iteratively.
computational effort. In the present method, the material properties of The final shape of the net from the numerical simulation is shown in
the numerical model are assumed consistent with that of the prototype Fig. 14 and compared with the experimental results. Regarding the
net. In order to acquire the right solutions, the M, K, Fs and Fh in Eq. iterative convergence, the criterion is that the maximum force residue is
(16) should be consistent between the physical and numerical nets. To less than 2e-5. The present simulation converges after 200 iterations by
satisfy the consistency of the aforementioned variables, three derived using 25.7 s. Regarding the consistency in the solution, the balance of
diameters, i.e., structural diameter (dws), elastic diameter (dwe) and forces in the vertical direction is checked. The total reaction force on
hydrodynamic diameter (dwh), are applied during the model building the four fixed nodes in the vertical direction is 27.06 N which is equal
through the external module (the red dashed box in Fig. 13). The de- and opposite with the to the sum of the wight of the net 0.06 N and the
tailed derivation is illustrated in Appendix B, and only the final re- three concentrated forces 27 N. Through the two examinations, the
lationships between the three numerical diameters and the physical present FE solver is proved feasible to simulate the flexible net.
Table 4
Parameters of the four net panels.
Net plane Twine diameter (mm) Half-mesh size (mm) Solidity Material Knots Mesh orientation
11
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
4.2. Comparative study on the hydrodynamic models general, the drag forces increase with the increasing current velocity
but with different increasing rates which depend on the expressions in
The second study is based on net panels. Four net panels with dif- hydrodynamic models.
ferent parameters which are wildly used in the aquaculture industry are Solidity is an important factor for force predictions. In general, the
selected from limited available experimental data (Tang et al., 2018; higher solidity can induce larger drag force. The predicted drag forces
Tsukrov et al., 2011) to study the applicability and accuracy of the for N4 (higher solidity net) using Morison models can fit the experi-
aforementioned hydrodynamic models with respect to net structures, mental data well, except for M5 when the velocity is 1 m/s. According
ambient velocities and inflow angles. The mesh grouping method is not to formulas in M5, the drag coefficient could be negative when
applied to the four net panels to ensure that the differences of the Re *Sn2 > 218. That means when the solidity is 0.3 and the Reynold
predicted forces only come from the hydrodynamic models themselves. number is higher than 2 400, the drag coefficient can be negative. Thus,
The parameters of the four net panels are given in Table 4, and the one should notice this strict applicable region when using this model.
structure of the four nets are shown in Fig. 1. The size of the net planes Knots can increase the hydrodynamic forces on nets. Compared to
in the numerical simulation is 1 m × 1 m, and the four edges of the net N4, N2 has smaller solidity, which means the drag force on N2 should
planes are fixed in the simulation. be smaller than N4 when both net panels were experiencing the same
current velocity. However, due to the additional drag force from the
4.2.1. Drag forces under different current velocities knots on N2, the two net panels have similar total drag force under the
Fig. 15 shows the drag forces on net planes for different current same environmental conditions. For the knotted net (N2), Morison
velocities with θ = 0˚ using the eleven hydrodynamic models. In models underestimate the drag forces as they ignore the effect from
Fig. 15. Drag forces on net panels for different current velocities when the incoming current is perpendicular to the net panels (θ = 0˚). The left subplots are the
simulation results using Morison models (solid lines). The right subplots are the simulation results using Screen models (dashed lines).
12
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
knots. These phenomena are consistent with the findings from Lader changing orientation. For Screen models, the drag forces are calculated
et al. (2014), in which the drag force on the knotless net is up to 10 % based on the total area of a net panel whose area is also unchangeable
less than that of the knotted net. For the knotted net (N2), the predicted with the changing orientation. Thus, the predicted forces based on both
forces based on Screen models fit the experimental data better than that types of hydrodynamic forces models are independent of orientations
based on Morison models. In particular, the predicted forces based on when the flow is perpendicular to the net panel. For instance, the two
S4 and S6 are very close to the experimental data, because these two net panels in Fig. 16 should have the same drag force when experien-
models have included the effect from knots. However, it is also ob- cing the same perpendicular flow. However, the drag forces on the two
served that S4 model always over predicts the drag force on knotless net net panels can be different when θ ≠ 0° (Balash et al., 2015). A nu-
panels (N1, N3 and N4). One should notice that this model was pro- merical study indicates that when θ > 45°, the drag force on net (b) is
posed more than 40 years ago. At that time, the aquaculture was only a larger than that on net (a) given the same flow velocity and direction
small industry compared to the fishing industry. The researchers used (Bi et al., 2017).
fishing nets, most likely knotted nets, to generate this hydrodynamic
model. It can also be proved by the fact that the predicted drag forces
based on S4 model fit the N2 best compared to the other models. Thus, 4.2.2. Drag and lift forces under different inflow angles
S4 is suitable to predict the hydrodynamic force for the knotted nets. In practice, most of the nets in a cylindrical fish cage are not per-
Different materials can make the twine surface roughness different, pendicular to the incoming current. Thus, it is essential to compare the
and the smooth surface can reduce the drag force. For the silicon-bronze hydrodynamic forces under different inflow angles. Due to limited ex-
net (N3), all the hydrodynamic models overestimate the drag force perimental data, only N1 and N2 have the experimental results under
compared to the experimental data, especially when the velocity is different inflow angles. Since most of the hydrodynamic models are not
higher than 0.5 m/s. When the velocity is 1 m/s, the discrepancies be- applicable to the knotted net (N2), we only discuss N1 in this section.
tween the experimental data and the predicted forces are varied from Fig. 17 shows the drag forces, lift forces, and lift to drag ratios of the net
43 % to 113 %. However, for the Nylon nets (N1 and N4), the dis- plane for different inflow angles θ. The current velocity in this figure is
crepancies between experimental data and the predicted forces can be fixed to 0.6 m/s to be consistent with the experiments reported by Tang
as low as 0.4 %. Since all the eleven hydrodynamic models in the et al. (2018).
present study were developed based on fibred nets whose surface is In general, the drag force decreases with the increasing inflow
rougher than that of metal nets, they are unsuitable for the smooth angle, but the decreasing ratios are various due to the different for-
metal nets. Moreover, the experimental results reported by Cha et al. mulas in hydrodynamic models. As shown in Fig. 17, the calculated
(2013) revealed that the coefficients of chain-link woven copper alloy drag forces based on the five Morison models have similar values, and
nets are smaller than that of the fabric nets with similar solidity only for all fit the experimental data well when the inflow angle is smaller than
inflow angles under 60°. Additional research work is necessary to have 70˚. While θ > 70˚, all the Morison models overestimate the drag force
a better understanding of the hydrodynamic differences between fabric due to the absence of the twine-to-twine wake effect. That means the
nets and copper alloy nets. drag force on 22 % nets of a cylindrical fish cage can be overestimated.
Although the solidity has the same physical meaning, the formulae The overestimated drag forces could lead to inaccuracy in the predic-
to estimate solidity are different in the six Screen models. Table 5 tion of displacements and cultivation volumes. Thus, the accuracy of
compares the solidities of N1 and N4 based on the analytical formulae the simulations based on Morison models is low when the fish cage has
in S1-S4 Screen models with that from experimental data. The analy- large deformation (Moe-Føre et al., 2016). For Screen models, not all
tical estimations of N1′s solidity are within 5% of the experimental the models agree with the experimental results well. S3, S5 and S6
data. However, for the solidity of N4, the relative difference between underestimate the drag forces when θ > 30˚; S4 overestimates the drag
the estimated and experimental value can be as large as 10.1 %. That forces when θ < 30˚ and underestimates the drag forces when θ > 30˚;
large difference in the solidity can affect the accuracy of the predicted S1 and S2 agree with the experimental data quite well for all inflow
force. Thus, the predicted drag forces on N4 (high solidity net) have angles. According to the drag coefficients in Fig. 6, only S1 and S2
large deviations than that for N1 (low solidity net) when using the decrease along with the cosine function with the increasing inflow
Screen models. In addition, when attaching the net panels to the frame, angle. The drag coefficients of the rest Screen models decrease much
a pre-tension is usually needed to keep the net in the desired shape in faster than the cosine function. That is the reason why the drag forces
the experiments. The different pre-tensions can make the solidity of the are underestimated by S3-S6 when θ > 30˚. It is observed that the drag
net panel various. Thus, it would be better to use the digital image forces based on S3 and S6 models are zero when θ = 90˚. That un-
processing (DIP) technique to estimate the solidity when testing the physical value contradicts with the experiment data by Zhou et al.
hydrodynamic properties of nets. (2015). Therefore, S1 and S2 are more accurate than the other Screen
The mesh orientations have negligible effects on drag force in nu- models in the drag force prediction for N1 net panel with different
merical simulations when inflow angles θ = 0°. For Morison models, inflow angles. The relative differences between the two models and
the drag forces on a net panel are the sum of the force on each twine. experimental results are shown in Fig. 18 and discussed later.
The sum of the projected area of the twines does not change with the Regarding the lift force, it first increases and then decreases with the
increasing inflow angle according to the experimental data in Fig. 17. In
Table 5
Solidities of N1 and N4.
Experimental value Estimated value
Model – S1 S2 S3 S4
Formula – Sn =
2dw
Sn =
dw (2L + 0.5dw )
Sn =
dw (2L dw )
Sn =
2dw
L L2 L2 L
N1 Solidity 0.132 0.1352 0.1375 0.1307 0.1352
Relative differencea – 2.4 % 4.2 % −0.9% 2.4 %
N4 Solidity 0.2056 0.2203 0.2264 0.208 0.2203
Relative differencea – 7.2 % 10.1 % 1.3 % 7.2 %
a
Relative difference = (estimated estimation - experimental value)/experimental value × 100 %.
13
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 16. Illustration of the inflow angle θ (left) and mesh orientation (right). The two net panels have the same solidity and mesh size.
Fig. 17. Drag and lift forces for different inflow angles when the flow velocity is 0.6 m/s for N1 using different hydrodynamic models. The left subplots are the
simulation results using Morison models (solid lines). The right subplots are the simulation results using Screen models (dashed lines).
general, curves of the lift forces are similar to the shape of the sine underestimate the lift-to-drag ratio when θ > 30˚, because of the
function. For Morison models, the predicted lift forces are similar to overestimated drag forces and underestimated lift forces. For Screen
each other, and all of them are smaller than the experimental results models, the curves are distinct among the different models. Only S1 and
when θ > 30˚. The underestimations of lift force might lead to under- S2 fit the experimental results. It can be observed that when θ > 45˚,
estimations on fish cage deformations. For the Screen models, S3, S4 the lift-to-drag ratios based on S3 and S5 are larger than one, and are
and S5 overestimate the lift force when 15˚ < θ < 45˚ due to their large two times higher than the experimental results. That irrational re-
lift force coefficients; S1 and S2 slightly underestimate the lift force lationship could lead to incorrect simulations where N1 is used as the
when 30˚ < θ < 60˚; S6 has zero lift force based on its formulas. cage net.
The lift-to-drag ratio is a dimensionless parameter to express the Based on the aforementioned discussion on Fig. 17, four models, i.e.
relationship between lift and drag forces. The experimental data in- M4, M5, S1 and S2, are chosen to calculate the relative difference be-
dicate that the maximum lift-to-drag ratio is 0.5. For the Morison tween their predicted results to the experimental results. As shown in
models, the curves are close to each other, and all of them can fit the Fig. 18(a), the drag forces predicted by the four models are within 5%
experimental results well when θ < 30˚. All the Morison models of the experimental results when θ < 70˚. However, the drag forces
14
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 18. The relative difference between predicted and experimental results. The relative difference = (predicted results-experimental results) / experimental results
×100 %.
predicted by Morison models are more than twice of the experimental regression formula using the least-squares method based on the sum of
results when θ > 70˚, due to the lack of the twine-to-twine wake effect. velocity reductions after cylinders according to Eq. (14). Moreover, the
These overestimations in large inflow angle were also observed by new formula considers the solidity as well. According to the new for-
Endresen et al. (2013). mula, r = f2 (Sn, ) , the nets with larger inflow angle ( ) induce smaller
According to Fig. 18(b), the lift forces predicted by Morison models flow velocity in its wake region. A comparison between the two for-
are less than half of the experimental results when θ > 45˚. The un- mulas with experimental results by Bi et al. (2013) and Patursson
derestimated lift forces together with the overestimated drag forces (2008) are shown in Fig. 19. Based on the experimental results, the flow
might lead to incorrect results when θ > 45˚. The lift forces and lift-to- reduction factor decreases with the increasing inflow angle. Compared
drag ratios predicted by Screen models are better than those predicted to the commonly used formula, r = f1 (Sn) , the new formula,
by Morison models, especially when θ > 45˚. In particular, the relative r = f2 (Sn, ) , shows better agreement with the experimental results.
difference of S1 is as high as 10 % compared to the experimental results
in Fig. 18(c). 4.3.1. Numerical models and loads
According to the comparisons and discussions on the four net pa- Two fish cages are used in this section. The two fish cages have the
nels, S1 is more suitable than the others to predict the hydrodynamic same diameter and height, but different solidities for the nets. The
forces on net panels for different velocities and inflow angles. Thus, S1 parameters for the numerical models and corresponding experimental
is chosen to apply to a cylindrical fish cage in the nest section to study
the wake effect.
Eq. (21) is the most commonly used formula in the dynamic analysis
of fish cages (Løland, 1991; Aarsnes et al., 1990; Kristiansen and
Faltinsen, 2012; Moe-Føre et al., 2016), in which r is dependent on Fig. 19. Comparison of the two formulas of flow reduction factor with ex-
solidity and unchangeable for all the rear half nets. Eq. (22) is a new perimental data.
15
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
models are listed in Table 6. f2 (Sn , ) is applied. In addition, the deformations at the frontal half of
The nodes in the upper circumference of the numerical model are the fish cage are similar in the two numerical models, because the
restricted from translational motion, representing the rigid and fixed frontal nets experience the same current velocity in both models.
steel ring in the physical model. In the experiments by Moe-Føre et al. Fig. 24 shows the normalised height of fish cage in numerical si-
(2016), each sinker is a circular steel cylinder with a diameter of 4 cm, a mulations with the two net-to-net wake effects. The normalised height
length of 6 cm, and a submerged weight of 4.48 N, as given in Table 6. is calculated as the height of fish cages at a given current velocity di-
In the numerical model, the 16 sinkers are represented by 16 vertical vided by the initial height of the fish cage (1.53 m). Since the bottom
concentrated forces corresponding to the submerged weight. Fig. 20 nodes of the fish cage are not in a horizontal plane, the height of the fish
shows the physical and numerical fish cage models in still water. It can cage is calculated as the vertical distance between the lowest node and
be observed that both the physical and numerical fish cages are stret- highest node. It can be observed that the height decreases with in-
ched in Z-direction due to the weights. creasing current velocity. The height of the model using r = f2 (Sn, ) is
clearly larger than that using r = f1 (Sn) , and the distinctions become
4.3.2. Convergence studies significant with increasing current velocity. In particular, when the Sn
In order to demonstrate the reliability of the present numerical re- = 0.347, the normalised height of the fish cage is 0.26 for the model
sults, convergence studies on both computational mesh and time step using r = f1 (Sn) , and 0.45 for the model using r = f2 (Sn, ) . The dis-
are performed at first. tinction in the height of fish cage can influence the design of feeding
In the convergence study of computational meshes, five different system and on-site operations related to nets, as the height of fish cage
sets of computational meshes shown in Table 7, are created for the fish should be provided to make a precise decision.
cage with high solidity net in Table 6. Drag forces on the fish cage are
estimated by using the five sets of computational meshes under flow
4.3.4. Comparison of the drag force
velocity of 1 m/s. As shown in Fig. 21(a), the relative differences of drag
Fig. 25 shows the comparison of the drag forces using the two forms
forces are less than 3%, which demonstrates that the present mesh
of flow reduction factor with experimental data by Moe-Føre et al.
grouping method, as discussed in Section 3.2 and Appendix B, is
(2016). According to the experimental results: (1) the drag force on the
workable for aquaculture nets. As shown in Table 7, with the increasing
low solidity (Sn = 0.194) fish cage is nearly proportional to the current
number of nodes (elements), the computer memory and computational
velocity; (2) the drag force on the high solidity (Sn = 0.347) fish cage
time are increased, and the difference of the drag force compared to the
increases slower with increasing current velocity when the velocities
finest computational mesh (Mesh 5) is reduced. In order to achieve the
are above 0.5 m/s than that at lower velocities.
results within 1% difference compared to the finest computational mesh
The calculated drag forces using both net-to-net wake models in-
(Mesh 5) and keep the computational costs low, Mesh 3 is chosen for
crease with increasing current velocity and are close to the experi-
the subsequent simulations. By using Mesh 3, the numerical model
mental results when the current velocity is less than 0.5 m/s. Compared
consists of 64 elements (64 nodes) with 85.90 mm length around the
to the experiments conducted by Moe-Føre et al. (2016), the model
circumference and 16 elements (17 nodes) with 93.75 mm length over
using r = f1 (Sn) overestimates the drag force, especially when the
the depth. The total numbers of elements and nodes are 2112 and 1088
current velocity is high, and the overestimations are more evident for
in the numerical model, respectively. In addition, by assigning different
the higher solidity fish cage. For the model using r = f2 (Sn, ) , the slope
diameters in the present solver according to the relationships in Eq.
of the drag force curve decreases when the current velocity exceeds
(21), the two physical fish cage models with different solidities in
0.5 m/s, and the predicted drag forces agree with the experimental
Table 6 are modelled by using the same computational mesh resolution,
results quite well. In particular, the maximum difference between the
i.e., Mesh 3.
numerical and experimental results is only 5% when using
In the convergence study of time steps, four different time steps
r = f2 (Sn, ) . And the drag force on the fish cage can be as large as 30
listed in Table 8 are applied in the simulations by using Mesh 3. Drag
% higher than the experimental results when applying r = f1 (Sn) . Ac-
forces on the fish cage under different time steps are calculated under
cording to the experimental photos in Fig. 23, the fish cage has large
flow velocity of 1 m/s. As shown in Fig. 21(b), the drag forces first
deformation, i.e., the nets have large inflow angles, when the current
increase then decay fast with oscillations as the time increases; After 6 s,
velocity is high. Together with the comparison in Fig. 19 which in-
all the simulations reach equilibrium. As shown in Table 8 and
dicates that f1 (Sn) highly overestimates the flow reduction factor when
Fig. 21(b), the drag forces on the fish cage calculated with the four > 70°, the drag force on the downstream nets can be overestimated
different time steps reach the same value at the end of simulations. when applying r = f1 (Sn) . Therefore, the total drag force on the fish
Increasing the time step can reduce the computational time, sig- cage is overestimated when using f1 (Sn) .
nificantly. Since the simulations are calculated under pure current The comparison of the two net-to-net wake effects indicates that the
conditions without any oscillating load, the studied time steps have
neglectable influences on the final results as long as the simulation is Table 6
converged. Therefore, the subsequent simulations are calculated by The parameters of the fish cage.
using Mesh 3 with a time step of 0.2 s and a duration of 10 s.
Experimental model Numerical model
16
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 20. The fish cage in still water shown from the side: (a) the physical fish cage model, (b) the numerical fish cage model. The numerical model is subjected to the
flow along the x-axis.
Table 7 Table 8
Mesh size, computational time and estimated drag force with time step = 0.1 s. Computational time and final drag force by using Mesh 3 with different time
steps.
Mesh Mesh 1 Mesh 2 Mesh 3 Mesh 4 Mesh 5
Time step Δt = 0.02s Δt = 0.05s Δt = 0.1 s Δt = 0.2s
Number of nodes 320 672 1088 1840 2592
Number of elements 608 1296 2112 3600 5088 Computational time (s) 9919.8 4005.8 2175.6 1201.8
Length of element (mm) 171.81 114.54 85.90 68.72 57.27 Drag force (N) 212.6 212.6 212.6 212.5
Computer memory (MB) 386.83 536.97 550.22 1007.80 1260.84
Computational time (s) 312.1 1169.8 2175.6 5189.2 10641.0
Drag force (N) 210.0 211.9 212.6 213.3 213.6
a flexible fish cage under pure current conditions with satisfactory
accuracy. In particular, the maximum difference between the nu-
commonly used expression, r = f1 (Sn) , is not sufficient to model the merical and experimental results is only 5% in drag force prediction.
interaction between the fluid flow and nets (hydro-elasticity). The 3 The new formula proposed in the present work can fix the evident
variable flow reduction factor, r = f2 (Sn, ) , is recommended for nu- defect in the previous formula for the net-to-net wake effect. With
merical simulations of the fish cage with the high solidity nets and the help of the new formula, the discrepancy between the predicted
subjected to high current velocities. and experimental drag force for a fish cage can be reduced from 30
% to 5%.
5. Conclusion
Besides, comparative studies on the hydrodynamic models and
In the present study, five Morison models and six Screen models for wake effects are conducted using the developed model and workflow.
calculating hydrodynamic forces on aquaculture nets are reviewed and The following conclusions are drawn from this study:
implemented to a general FE solver, Code_Aster. The main contribu-
tions of the present study are listed as follows: 1 When the incoming flow is perpendicular to the net panel (θ = 0°),
the drag force on knotless nylon nets can be well predicted by all the
1 It is the first time that Code_Aster, the open-source FE solver de- hydrodynamic models except for S4 which is originally for knotted
veloped by EDF R&D, has been used to simulate nets in marine nets. The discrepancies between experimental data and the pre-
aquaculture facilities. The successful application is fulfilled through dicted forces can be as low as 0.4 % when the current velocity is
the external module in the present work. 1 m/s.
2 Verifications based on computational mesh and time step con- 2 For the metal net with smooth surface, all the eleven hydrodynamic
vergences and validations with experimental results are achieved. It models overestimate the drag force. That is because all these models
is shown that by employing the newly developed external module in are initially developed for twisted or braided nets with rough sur-
Code_Aster, the present numerical model can predict the response of faces. Further studies are needed to develop a new hydrodynamic
model for metal nets.
Fig. 21. Convergence studies on both computational mesh and time step. In (a) Mesh convergence study, the relative difference is calculated by taking the drag force
in Mesh 5 as a reference value.
17
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Fig. 22. Comparison of experimental and numerical results subjected to different current velocities, Sn = 0.194. The red model uses r = f1 (Sn) for the net-to-net
wake, and the blue model uses r = f2 (Sn , ) for the net-to-net wake. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article).
Fig. 23. Comparison of experimental and numerical results subjected to different current velocities, Sn = 0.347. The red model uses r = f1 (Sn) for the net-to-net
wake, and the blue model uses r = f2 (Sn , ) for the net-to-net wake. (For interpretation of the references to colour in this figure legend, the reader is referred to the
web version of this article).
3 Knots can bring additional drag force on nets. Morison models un- predicted drag forces are two times higher than the experimental
derestimate the drag force on knotted nets if the effects from knots results when θ > 70˚, due to the lack of the twine-to-twine wake
are not considered. As for Screen models, the drag forces on the effect. As for Screen models which include the twine-to-twine wake
knotted net can only be well predicted by S4 and S6. effect implicitly, the predicted forces are within 10 % of the ex-
4 When the incoming flow is not perpendicular to the net panel (0° perimental results for all inflow angles.
< θ ≤ 90˚), drag forces predicted by the five Morison models are 5 The drag force on a single fish cage is overestimated by the existing
within 5% of the experimental results if θ < 70˚. However, the Screen models, especially when high-solidity nets experience large
Fig. 24. The normalised height of fish cages in numerical simulations with the two net-to-net wake effects.
18
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
deformation. This is due to the inappropriate net-to-net wake effect. Declaration of Competing Interest
The consistent flow reduction factor, which is commonly used in the
fish cage simulation, can overestimate the current velocities on The authors declare that they have no known competing financial
downstream nets. Thus, the total drag force on a fish cage can be as interests or personal relationships that could have appeared to influ-
large as 30 % higher than the experimental results. ence the work reported in this paper.
Appendix A. Net panel area, normal vector and unit force vector
In the present study, the “hydrodynamic panel element” is a triangle (e.g. P1-P2-P4 in Fig. A1). The triangular net panel elements are generated
automatically by a mesh-generating function. The initial area of the triangular net panel is approximately equal to L2/2, where L is the half mesh size.
The area of the “hydrodynamic panel element” can change during the simulations and is dependent on the position of its vertexes. In the simulation,
the coordinates of every node are known. The area of the “hydrodynamic panel element” is calculated according to the following equation:
1
At = |P1 P2 × P1 P4|
2 (A.1)
The unit normal vector of the net panel is calculated by the following equation:
P1 P2 × P1 P4
en =
|P1 P2 × P1 P4| (A.2)
The unit force vectors (iD and iL ) which are used to indicate the directions of drag and lift forces are calculated as follows:
u v
iD =
|u v| (A.3)
(u v ) × e n × (u v)
iL =
|(u v ) × en × (u v )| (A.4)
This section gives an introduction on how to reduce the number of elements using the present mesh grouping method. As mentioned in Section
3.4, the principle is to keep M, K, Fs and Fh in Eq. (16) consistent between the physical net and numerical net. In the numerical model, the fluid
density ( fluid ), the fluid velocity (u ), the density of twine ( twine ) and Young’s modulus of twine (E) are consistent with the physical value, and λ is the
ratio between the half mesh size of the numerical net (Ln) and the half mesh size of the physical net (Lp). As shown in Fig. B1, the nets in the two blue
dashed boxes should have the same mass (M), stiffness (K) and environmental loads (Fs+Fh). To satisfy the consistency, three derived diameters, i.e.,
structural diameter (dws), elastic diameter (dwe) and hydrodynamic diameter (dwh), are used in the numerical model building. Below, the relationships
between the three diameters and the physical twine diameter (dw0) are derived.
Mass conservation (M)
As shown in Fig. B1, the mass of the physical net in the blue dashed box is:
Fig. 25. Measured and calculated drag force in different current velocities using the two net-to-net wake effects.
19
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
Mp = twine (2 d w2 0 Lp 2 d w2 0 d w 0)
4 4 (B.1)
And the mass of the numerical net in the blue dashed box is:
2 2
Mn = twine (2 d ws Ln d ws d ws )
4 4 (B.2)
Because Mp = Mn , the structural diameter d ws should satisfy:
2L p dw0
d ws = d w0
2Ln dws (B.3)
For typical aquaculture nets, 2Lp d w0 and 2Ln d ws . Thus, the square root term can be simplified as:
2L p d w0 1
2Ln dws (B.4)
Then the structural diameter can be obtained as:
d ws d w0 (B.5)
Stiffness equivalent (K)
Because typical aquaculture nets have a negligible bending stiffness and cannot carry any compression load, the stiffness of the physical net in the
blue dashed box can be written as:
4
d w2 0 E 1 0
Kp =
Lp 0 1 (B.6)
And the stiffness of the numerical net in the blue dashed box is:
20
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
2
4
d ws E 1 0
Kn =
Ln 0 1 (B.7)
Based on Kp = Kn , then the elastic diameter can be obtained as:
d we = d w0 (B.8)
Environmental loads conservation (Fs and Fh )
Gravity and buoyancy forces (Fs )
The Fs on the physical and numerical net are given in Eqs. (B.9) and (B.10):
2 2 2 2
Fsn = fluid g 2 d ws Ln d ws d ws twine g (2 d ws Ln d ws d ws )
4 4 4 4 (B.10)
Similar to the derivation for the mass conservation, the diameter for the gravity and buoyancy forces is the same as the one for the mass.
Therefore, it can use the same parameter, d ws , to calculate the gravity and buoyancy forces.
Hydrodynamic forces (Fh )
For both Morison and Screen models, the hydrodynamic forces are calculated based on the following equation:
1
Fh = Cd fluid A|u v|(u v)
2 (B.11)
In Morison models, the hydrodynamic coefficients depend on the physical twine diameters (d w0 ) and the reference area A is the projected area of
twines. The projected area of twines in the physical net is:
2d 2
Ap = 2 d w0 Lp w0 (B.12)
The projected area of the twines in the numerical net is:
An = 2d wh Ln 2
d wh (B.13)
Based on Ap = An , the hydrodynamic diameter should satisfy:
d wh = d w0 (B.14)
In Screen models, the hydrodynamic coefficients depend on the solidity or the twine diameters (d w0 ) of the physical net. The reference area A in
the numerical model is the net panel area, which is the same as the physical net. The solidity of the physical net is:
d w0 (2Lp dw0 )
Snp =
Lp 2 (B.15)
And the solidity of the numerical net is
d wh (2Ln d wh )
Snn =
Ln 2 (B.16)
Base on Snp = Snn , the derived hydrodynamic diameter satisfies the same relationship in Eq. (B.14).
In summary, Eqs. (B.5),(B.8) and (B.14) have been used in the present mesh grouping method to reduce the computational effort.
M3 100.7Re 0.3
Re < 200 – 10<Re<2 × 105
Cn =
1.2 Re > 200
Ct = 0.1; Cm = 1.0
M4 8 2) – 0<Re<107
(1 0.87s 0 < Re < 1
sRe
1.45 + 8.55Re 0.9 1 < Re < 30
Cn = 1.1 + 4Re 0.5 30 < Re < 2.33 × 105
3.41 × 10 (Re 5.78 × 10 ) 2.33 × 105 < Re < 4.92 × 105
6 5
Re × 105
0.401(1 e 5.99 ) 4.92 × 105 < Re < 107
21
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
dw u Du
Recyl = ; Resp = ; D = knot diameter
µ (1 Sn) µ (1 Sn)
References and currents using a Morison-force model. Ocean. Eng. 141, 283–294.
Cifuentes, C., Kim, M.H., 2017b. Numerical simulation of fish nets in currents using a
Morison force model. Ocean. Syst. Eng. 7 (2), 143–155.
Aarsnes, J.V., Løland, G., Rudi, H., 1990. Current Forces on Cage, Net Deflection. Decew, J., 2011. Development of Engineering Tools to Analyze and Design Flexible
Engineering for Offshore Fish Farming. pp. 891–921. Structures in Open Ocean Environments. Doctor of Philosophy. University of New
Antonutti, R., Peyrard, C., Incecik, A., Ingram, D., Johanning, L., 2018. Dynamic mooring Hampshire, Durham.
simulation with Code_Aster with application to a floating wind turbine. Ocean. Eng. DeCew, J., Tsukrov, I., Risso, A., Swift, M.R., Celikkol, B., 2010. Modeling of dynamic
151, 366–377. behavior of a single-point moored submersible fish cage under currents. Aquac. Eng.
Aubry, J.P., 2019. Beginning with Code_Aster –A Practical Introduction to Finite Element 43 (2), 38–45.
Method Using Code_Aster and Gmsh. Publié sous licence Paris, Framasoft (coll. Electricité de France (EDF), 1989. Finite Element Code_Aster, Analysis of Structures and
Framabook). Thermomechanics for Studies and Research. Open source on www.code-aster.org. .
Balash, C., Colbourne, B., Bose, N., Raman-Nair, W., 2009. Aquaculture net drag force and Endresen, P.C., Føre, M., Fredheim, A., Kristiansen, D., Enerhaug, B., 2013. Numerical
added mass. Aquac. Eng. 41 (1), 14–21. modeling of wake effect on aquaculture nets. ASME 2013 32nd International
Balash, C., Sterling, D., Binns, J., Thomas, G., Bose, N., 2015. The effect of mesh or- Conference on Ocean, Offshore and Arctic Engineering: American Society of
ientation on netting drag and its application to innovative prawn trawl design. Fish. Mechanical Engineers V003T05A027-V003T05A027.
Res. 164, 206–213. Faltinsen, O.M., Shen, Y.G., 2018. Wave and current effects on floating fish farms: key-
Berstad, A.J., Aarsnes, J.V., 2018. A case study for an offshore structure for aquaculture: note contribution for the international workshop on wave loads and motions of ships
comparison of analysis with model testing. ASME 2018 37th International Conference and offshore structures. J. Mar. Sci. Appl. 17 (3), 284–296.
on Ocean, Offshore and Arctic Engineering: American Society of Mechanical FAO, 2018. In: The State of World Fisheries and Aquaculture 2018 Meeting the sustain-
Engineers V11BT12A057-V11BT12A057. able development goals. Rome. Licence: CC BY-NC-SA 3.0 IGO.
Berstad, A.J., Heimstad, L.F., 2017. Experience from introduction of the design code NS Fridman, A.Lv., 1973. Theory and Design of Commercial Fishing Gear (Trans. from Russ.).
9415 to the aquaculture industry in Norway and expanding the scope to cover also Jerusalem. .
operations. ASME 2017 36th International Conference on Ocean, Offshore and Arctic Gansel, L.C., Plew, D.R., Endresen, P.C., Olsen, A.I., Misimi, E., Guenther, J., Jensen, O.,
Engineering: American Society of Mechanical Engineers V009T12A003- 2015. Drag of clean and fouled net panels - measurements and parameterization of
V009T12A003. fouling. PLoS One 10 (7), e0131051.
Berstad, A.J., Walaunet, J., Heimstad, L.F., 2013. Loads from currents and waves on net Klebert, P., Lader, P., Gansel, L., Oppedal, F., 2013. Hydrodynamic interactions on net
structures. ASME 2012 31st International Conference on Ocean, Offshore and Arctic panel and aquaculture fish cages: a review. Ocean. Eng. 58, 260–274.
Engineering: American Society of Mechanical Engineers 95–104. Kristiansen, T., Faltinsen, O.M., 2012. Modelling of current loads on aquaculture net
Bessonneau, J.S., Marichal, D., 1998. Study of the dynamics of submerged supple nets cages. J. Fluids Struct. 34, 218–235.
(applications to trawls). Ocean. Eng. 25 (7), 563–583. Lader, P., Jensen, A., Sveen, J.K., Fredheim, A., Enerhaug, B., Fredriksson, D., 2007.
Bi, C.W., Xu, T.J., 2018. Numerical study on the flow field around a fish farm in tidal Experimental investigation of wave forces on net structures. Appl. Ocean. Res. 29 (3),
current. Turk. J. Fish. Aquat. Sci. 18 (5), 705–716. 112–127.
Bi, C.W., Zhao, Y.P., Dong, G.H., Xu, T.J., Gui, F.K., 2013. Experimental investigation of Lader, P., Enerhaug, B., Fredheim, A., Klebert, P., Pettersen, B., 2014. Forces on a cru-
the reduction in flow velocity downstream from a fishing net. Aquac. Eng. 57, 71–81. ciform/sphere structure in uniform current. Ocean. Eng. 82, 180–190.
Bi, C.W., Balash, C., Matsubara, S., Zhao, Y.P., Dong, G.H., 2017. Effects of cylindrical Lee, C.W., Lee, J.H., Cha, B.J., Kim, H.Y., Lee, J.H., 2005. Physical modeling for under-
cruciform patterns on fluid flow and drag as determined by CFD models. Ocean. Eng. water flexible systems dynamic simulation. Ocean. Eng. 32 (3–4), 331–347.
135, 28–38. Li, L., Ong, M.C., 2017. A preliminary study of a rigid semi-submersible fish farm for open
Blomgre, A., Fjelldal, Ø.M., Quale, C., Misund, B., Tvetarås, R., Kårtveit, B.H., 2019. seas. ASME 2017 36th International Conference on Ocean, Offshore and Arctic
Kartlegging av investeringer i fiskeri og fangst. akvakultur og fiskeindustri, Engineering: American Society of Mechanical Engineers V009T12A044-
Stavanger, pp. 1970–2019. V009T12A044.
Cardia, F., Lovatelli, A., 2015. Aquaculture Operations in Floating HDPE Cages: A Field Li, L., Fu, S.X., Xu, Y.W., Wang, J.G., Yang, J.M., 2013. Dynamic responses of floating fish
Handbook. FAO and Ministry of Agriculture of the Kingdom of Saudi Arabia. cage in waves and current. Ocean. Eng. 72, 297–303.
Cha, B.J., Kim, H.Y., Bae, J.H., Yang, Y.S., Kim, D.H., 2013. Analysis of the hydrodynamic Li, L., Jiang, Z.Y., Hoiland, A.V., Ong, M.C., 2018. Numerical analysis of a vessel-shaped
characteristics of chain-link woven copper alloy nets for fish cages. Aquac. Eng. 56, offshore fish farm. J. Offshore Mech. Arctic Eng.-Trans. ASME 140 (4), 041201.
79–85. Løland, G., 1991. Current Forces on and Flow Through Fish Farms. Doctor. Norwegian
Chen, H., Christensen, E.D., 2017. Development of a numerical model for fluid-structure Institute of Technology.
interaction analysis of flow through and around an aquaculture net cage. Ocean. Eng. Moe, H., Fredheim, A., Hopperstad, O.S., 2010. Structural analysis of aquaculture net
142, 597–615. cages in current. J. Fluids Struct. 26 (3), 503–516.
Cheng, H., 2017. Study on the Anti-Current Characteristics of a New Type Gravity Fish Moe-Føre, H., Endresen, P.C., Aarsæther, K.G., Jensen, J., Føre, M., Kristiansen, D.,
Cage and Design Optimizing. Master, Ocean University of China, Qingdao, China. Fredheim, A., Lader, P., Reite, K.-J., 2015. Structural analysis of aquaculture nets:
Cifuentes, C., Kim, M.H., 2015. Dynamic analysis for the global performance of an SPM- comparison and validation of different numerical modeling approaches. J. Offshore
feeder-cage system under waves and currents. China Ocean. Eng. 29 (3), 415–430. Mech. Arct. Eng. 137 (4), 041201.
Cifuentes, C., Kim, M.H., 2017a. Hydrodynamic response of a cage system under waves Moe-Føre, H., Lader, P., Lien, E., Hopperstad, O., 2016. Structural response of high
22
H. Cheng, et al. Aquacultural Engineering 90 (2020) 102070
solidity net cage models in uniform flow. J. Fluids Struct. 65, 180–195. Turner, A.A., Steinke, D.M., Nicoll, R.S., 2017. Application of wake shielding effects with
Patursson, O., 2008. Flow Through and Around Fish Farming Nets. Doctoral. University of a finite element net model in determining hydrodynamic loading on aquaculture net
New Hampshire, Durham. pens’. ASME 2017 36th International Conference on Ocean, Offshore and Arctic
Reichert, P., 1994. Aquasim – a tool for simulation and data-analysis of aquatic systems. Engineering: American Society of Mechanical Engineers V006T05A001-
Water Sci. Technol. 30 (2), 21–30. V006T05A001.
Reite, K.-J., Føre, M., Aarsæther, K.G., Jensen, J., Rundtop, P., Kyllingstad, L.T., Endresen, Wan, R., Hu, F.X., Tokai, T., 2002. A static analysis of the tension and configuration of
P.C., Kristiansen, D., Johansen, V., Fredheim, A., 2014. Fhsim—time domain simu- submerged plane nets. Fish. Sci. 68 (4), 815–823.
lation of marine systems’. ASME 2014 33rd International Conference on Ocean, Yu, J., 2017. Experimental and Numerical Investigations of Loads on Aquaculture Nets.
Offshore and Arctic Engineering: American Society of Mechanical Engineers Master. NTNU, Trondheim.
V08AT06A014-V08AT06A014. Zdravkovich, Momchilo M., 2003. Flow Around Circular Cylinders, vol. 2 Applications.
Shainee, M., DeCew, J., Leira, B.J., Ellingsen, H., Fredheim, A., 2013. Numerical simu- Oxford university press.
lation of a self-submersible SPM cage system in regular waves with following cur- Zhan, J.M., Jia, X.P., Li, Y.S., Sun, M.G., Guo, G.X., Hu, Y.Z., 2006. Analytical and ex-
rents. Aquac. Eng. 54, 29–37. perimental investigation of drag on nets of fish cages. Aquac. Eng. 35 (1), 91–101.
Shimizu, H., Mizukami, Y., Kitazawa, D., 2018. Experimental study of the drag on fine- Zhao, Y., Li, Y., Dong, G., Gui, F., Teng, B., 2007a. Numerical simulation of the effects of
mesh netting. Aquac. Eng. 81, 101–106. structure size ratio and mesh type on three-dimensional deformation of the fishing-
Stengel, D., Mehdianpour, M., 2014. Finite element modelling of electrical overhead line net gravity cage in current. Aquac. Eng. 36 (3), 285–301.
cables under turbulent wind load. J. Struct. 2014. Zhao, Y., Li, Y., Dong, G., Gui, F., Teng, B., 2007b. A numerical study on dynamic
Takagi, T., Shimizu, T., Suzuki, K., Hiraishi, T., Yamamoto, K., 2004. Validity and layout properties of the gravity cage in combined wave-current flow. Ocean. Eng. 34
of "NaLA": a net configuration and loading analysis system. Fish. Res. 66 (2–3), (17–18), 2350–2363.
235–243. Zhao, Y., Li, Y., Dong, G., Gui, F., Wu, H., 2008. An experimental and numerical study of
Tang, H., Xu, L.X., Hu, F.X., 2018. Hydrodynamic characteristics of knotted and knotless hydrodynamic characteristics of submerged flexible plane nets in waves. Aquac. Eng.
purse seine netting panels as determined in a flume tank. PLoS One 13 (2), e0192206. 38 (1), 16–25.
Tsukrov, I., Eroshkin, O., Fredriksson, D., Swift, M.R., Celikkol, B., 2003. Finite element Zhao, Y., Bi, C., Dong, G., Gui, F., Cui, Y., Guan, C.-T., Xu, T.J., 2013a. Numerical si-
modeling of net panels using a consistent net element. Ocean. Eng. 30 (2), 251–270. mulation of the flow around fishing plane nets using the porous media model. Ocean.
Tsukrov, I., Drach, A., DeCew, J., Swift, M.R., Celikkol, B., 2011. Characterization of Eng. 62, 25–37.
geometry and normal drag coefficients of copper nets. Ocean. Eng. 38 (17–18), Zhao, Y., Bi, C., Dong, G., Gui, F., Cui, Y., Guan, C.-T., Xu, T.J., 2013b. Numerical si-
1979–1988. mulation of the flow around fishing plane nets using the porous media model. Ocean.
Turner, A.A., Jeans, T.L., Reid, G.K., 2016. Experimental investigation of fish farm hy- Eng. 62, 25–37.
drodynamics on 1: 15 scale model square aquaculture cages. J. Offshore Mechan. Zhou, C., Xu, L., Hu, F., Qu, X., 2015. Hydrodynamic characteristics of knotless nylon
Arctic Eng.-Trans. ASME 138 (6), 061201. netting normal to free stream and effect of inclination. Ocean. Eng. 110, 89–97.
23