Ocean Engineering 210 (2020) 107387

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Automatic CFD analysis of planing hulls by means of a new web-based

application: Usage, experimental data comparison and opportunities
Raffaele Ponzini a, *, Francesco Salvadore b, Ermina Begovic c, Carlo Bertorello c
a
CINECA, HPC Department, Milan Office, Via R. Sanzio 4, Segrate, MI, Italy
b
CINECA, HPC Department, Cineca, Rome Office, Via dei Tizii 6/B, Roma, Italy
c
University of Naples Federico II, Department of Industrial Engineering, Via Claudio 21, Naples, Italy

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

Keywords: Computational fluid dynamics (CFD) modelling is a powerful tool to analyze complex hydrodynamics problems
LincoSim like flow around bluff bodies and their interactions including flow separation and reattachments where analytical
Planing hull hydrodynamics formulas are not applicable nor reliable. Nevertheless, a relevant limitation on the systematic adoption of
Computational fluid dynamics (CFD)
automated CFD workflows for a given typology of problems is still present. In this work, the results of an
Experimental fluid dynamics (EFD)
OpenFOAM®
automatic and web-based application, named LincoSim, specifically designed to perform automatic CFD
High performance computing (HPC) modelling of planing hulls are presented. The tool uses only open-source software and deploys High Perfor­
mances Computing infrastructures to take advantage of well-established yet innovative technological bricks. A
detailed comparison for planing hull hydrodynamics analyses is performed in an “automated fashion” on a
systematic series of four planing hulls previously studied experimentally over a wide range of Froude numbers.
The numerical outcomes and the systematic series trends of total resistance, sinkage, dynamic trim, and wetted
surface area are compared with the measured experimental results. This paper shows that today CFD automated
workflows can be used to study planing hull hydrodynamics highlighting the fact that several innovative aspects
involved in LincoSim could have a strong impact on the way CFD tools are adopted in modern marine industrial
design.

1. Introduction limitation driven by technical costs, there is a limitation on the effective

number of CFD models analyzed in the preliminary hull design phase. As
Thanks to the evolving technologies in numerical libraries and confirmed by Reichel and Minchev (2014), small design offices seldom
hardware computational power, the possibility of achieving high- rely on CFD, and only a fraction of the world’s fleet uses CFD-generated
quality computational fluid dynamics (CFD) modelling for complex trim optimization databases to save fuel in daily operations. Further­
hydrodynamics problems not only at the research level but also during more, for complex hydrodynamic phenomena including sharp flow
ordinary day-by-day industrial product design is today a matter of fact. separation at the chine, dry transom at design speed, relatively high
Nevertheless, the standard way of using CFD applications remains running trim and large pressure gradients at the stagnation line as for
somehow linked to an old-style “single-user” approach in which also planing hulls, correct numerical setup for Reynolds Averaged
routine and repetitive activities, involved in a standardized CFD model Navier-Stokes (RANS) simulations requires experienced users.
development process, are performed ex-novo each time. Adopting a Therefore, the use of simple empirical methods like Savitsky (1964)
well-established workflow automation paradigm, based on a-priori or ones based on regression analysis or artificial neural network (ANN) is
knowledge of the physics involved in the solution of the specific prob­ still a very common design practice, as stated in Dumortier et al. (2019).
lem, many time-consuming tasks can be avoided. Moreover, there are In that work, the authors monitored for 11 months a sample of 35
several limiting factors on a wider diffusion of 3D computational models SAR-Military boats “Plascoa 1650”, tested in real conditions in the
to perform virtual towing tank data acquiring related also to software Arcachon bay and compared with the Power Prediction Tool (PPT)
licensing and hardware infrastructure costs. As a result, due to the based on Savitsky theory with some modifications for the aerodynamic
necessary human time required to perform such activities and to the drag determination. The average of the differences for speed under 9

* Corresponding author.
E-mail address: [email protected] (R. Ponzini).

https://doi.org/10.1016/j.oceaneng.2020.107387
Received 18 November 2019; Received in revised form 27 February 2020; Accepted 12 April 2020
Available online 7 June 2020
0029-8018/© 2020 Elsevier Ltd. All rights reserved.
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 1. Systematic hull cad details.

the user and reducing the need of highly skilled CFD engineer.
Table 1
Nevertheless, there is not yet, to our knowledge, a vertical applica­
Systematic hull series properties details. L: length over all; B: beam on chine; LA-
tion that includes the automation of the overall CFD workflow allowing
B: length of the prismatic part of hull; TAP: draft at the transom stern.
for interactive navigation of inputs and outputs taking advantage of web
Mono W1 W2 W3
technologies, high performance computing (HPC) servers and open-
L [m] 1.900 1.900 1.900 1.900 source libraries except for LincoSim as we already presented in Salva­
B [m] 0.424 0.424 0.424 0.424 dore and Ponzini (2019).
TAP [m] 0.096 0.106 0.110 0.108
The LincoSim web application is part of the LINCOLN project, see
0.25LA-B [m] 0.375 0.375 0.375 0.375
Water temperature [K] 297.75 298.35 297.75 297.25 LINCOLN Consortium (see LINCOLN Consortium, 2020,) and is a ver­
LCG [m] 0.637 0.660 0.609 0.586 tical solution that includes a complete CFD RANS computational
VCG [m] 0.143 0.152 0.155 0.156 workflow using a fully open-source-based software stack providing the
Mass [kg] 32.6 32.6 32.6 32.3
possibility to transparently take advantage of state-of-the-art HPC fa­
Ixx; Iyy; Izz [kg m2] 0.8; 10; 10 0.8; 10; 10 0.8; 10; 10 0.8; 10; 10
Max deadrise angle 17 24 30 36
cilities. In some way, LincoSim embodies the philosophy of Workflow as
[deg] a Service (WfaaS), as presented in several works (Korambath et al.,
2016; Haocheng et al., 2011; Zill et al., 2013) for different fields of
application. In other words, LincoSim is a platform, which fully au­
knots is 0.24%, while in the whole speed range the differences are tomatizes the complex workflow of hydrodynamic simulation, allowing
included in a more widespread range [-13.63%; 22.3%]. A more a “user designer” to perform and manage state-of-the-art 3D RANS CFD
comprehensive work by Radojcic (2019) presents a detailed overview of studies without the necessity of having CFD software competences, nor
the latest experimental data and available regression analysis based having HPC skills to execute codes in feasible computing times.
methods for planing hull resistance and propulsion prediction, high­ The evaluation of such a transversal tool requires the necessity of
lighting the limits of applicability of each systematic series or relative defining and ranking a set of very different parameters related to a very
regression formula. different field of application:
Starting from the works of Caponnetto (2001) and Brizzolara and
Serra (2007), and accordingly to more recent ones (see Broglia and a) technical (developer side): platform accessibility and stability,
Durante (2017); Stern et al., 2015; Brizzolara and Vernengo, 2016; components integration and user group management;
Broglia and Durante (2017); Sukas et al., 2017; Khazaee et al., 2019; b) usability (end-user side): end-user experience in terms of usability;
Judge et al., 2020; Niazmand Bilandi et al., 2020, He et al., 2019, Huang c) quality of results: application to a given experimental case or set of
et al., 2019) it has been shown and accepted that CFD is potentially a cases of the computational tool;
powerful tool to provide a robust and accurate virtualized measurement d) computational costs and time to results: evaluation of costs for the
technique for planing hull problems. end-user in terms of time to result and hardware requirements.
As for the experimental fluid dynamics (EFD) approaches, when CFD
is used to reproduce the experiment, the limited evaluation of the key Points (a) and (b) presenting the LincoSim tool and including usage
hull parameters impact on the knowledge of the hydrodynamic perfor­ statistics matured over 18 months of usage by three different designer
mances in the early design stage can bring a final sub-optimal design groups working within the LINCOLN project have been described in
involving unexpected power engine consumption or undesired high trim Salvadore et al. (2019); in appendix A the main details about the plat­
angles at a given velocity condition. form implementation are given for sake of completeness.
As underlined in Gatin et al. (2019) automation in a naval design The remaining points are still open and are the objective of the
application is of interest for CFD to effectively contribute to the shipping present paper.
industry on a larger scale. The example of “automation of the entire Concerning point (c), the application of the automated CFD solver,
process” has been proposed for Kriso container ship KCS, JBC car carrier applied to the case of bare planing hulls analysis, is presented in detail. A
and naval ship DTMB 5512, leaving CFD-specific parameters away from systematic series made of four planing hulls previously experimentally

2
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 2. Automatic workflow: a. High level overview; b. detailed information for the three central steps; c. details of the automation steps and tests performed to
evolve from one step to the next one.

3
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Table 2
Boundary conditions.
Patch name p U alpha k/omega

Hull fixedFluxPressure movingWallVelocity interfaceCompression wallFunction

Inlet fixedFluxPressure fixedValue fixedValue fixedValue
Outlet zeroGradient outletPhaseMeanVelocity variableHeightFlowRate inletOutlet
Top totalPressure pressureInletOutletVelocity inletOutlet inletOutlet
Bottom/Sides symmetryPlane symmetryPlane symmetryPlane symmetryPlane

Fig. 3. CAD positioning, domain definition, mesh topology scheme.

studied (Begovic and Bertorello, 2012) has been selected analyzing the 2. Materials and Methods
hydrodynamic outcomes of two Degrees of Freedom (DoF) solver
available in LincoSim spanning the same Froude number range as per­ 2.1. Experimental reference dataset
formed during the experimental campaign.
Concerning point (d), to have a complete understanding of the During the last decade, much research efforts have been invested in
computational performances and costs of LincoSim, for a given reference the understanding of possible strategies useful to obtain resistance
case hull at a given velocity value, we analyzed the best parallel per­ reduction for achieving the highest speed of fast planing hulls. New
formances achievable by the existing HPC platform considering three systematic series of planing hull models have been tested experimentally
different types of solver setup by playing with the number of degree of in calm water in Metcalf and Faul (2005) and Kowalyshyn and Metcalf
freedom from zero to two. (2006), Begovic and Bertorello (2012), Begovic et al. (2013), Kim et al.
The forthcoming of the paper is as follows: (2013),De Luca and Pensa (2012), de Luca and Pensa (2017). Also
stepped hulls have been tested by Taunton et al. (2010), Garland (2011),
� In section Materials and Methods the reference experimental dataset White et al. (2012), Lee et al. (2014), De Marco et al. (2017), Najafi et al.
and the automatic computational workflow are presented; (2019). Furthermore, some applications of the interceptors (Mansoori
� Results section contains the workflow numerical outcomes compared et al., 2017, Zou et al., 2019), spray rails (Seo et al., 2016; Molchanov
to the experimental measures and the application time to result and et al., 2019), tunnels (Yousefi et al., 2014) or forced ventilation on the
costs; bottom as in Cucinotta et al. (2017) and Yang et al. (2019), have been
� the Discussion section provides a deep qualitative comparison be­ revisited both experimentally and numerically as a possible improve­
tween EFD and CFD data analyzing the possible interplay between ment of hydrodynamic performances.
towing tank outcomes and numerical ones by including possible All new experimental data provides improved quality of measured
limitations of the overall approach; data due to better instrumentations, known uncertainty and contem­
� the Conclusions section provide a well-rounded closure of the paper porary hull form design assuring better references for the designers as
considering possible new perspectives for naval architects and well as for the validation of numerical tools. Among them, for the Lin­
designers; cosim application validation, the experimental data from the Begovic
� Appendix A is dedicated to the description of technical details of the and Bertorello (2012) are chosen.
web application; In the systematic series, the only parameter is a longitudinal varia­
� Appendix B contains the verification analysis of the CFD model and tion of the deadrise angle for four models: all models have the same
the mesh sensitivity analysis of the proposed meshing procedure; deadrise angle of 17� at 0.25L from the stern. In Fig. 1 the hulls body
� Appendix C is dedicated to the technical details of the computational plans are shown. One model has the constant deadrise angle all along the
costs, of the possible variations on the number of degrees of freedom length – so called monohedral model (Mono) and the three other have a
used by the solver and of the scalability performances of the solver. linear variation of the deadrise angle along the model length, obtaining
only linear inclination of the chine line (Wi, i ¼ 1,2,3). The peculiarity of
the models is the transparent bottom that enables the photographing of
the wetted surface during the experimental campaign and allows for
advanced analysis in “post-processing” as shown in the forthcoming of

4
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 4. a: Yplus value impact at different Froude number value for W3 hull; b: Imposed dynamic yPlus values over Froude number in the automatic mesh­
ing procedure.

the paper. Tests in calm water have been performed in the range of � Pre-processing: the geometry description of the hull (hull CAD) and
Froude numbers (Fr) from 0.266 to 1.73. the fluid dynamics conditions at which the hull has to be tested are
To perform meaningful CFD analysis also some relevant information used to define the computational grid in terms of extension and cells
of each hull studied in the series is necessary. These sets of values as size;
measured during the experimental campaign are reported in Table 1. � CFD model setup: the proper fluid properties, initial and boundary
conditions and rigid body dynamics constants are defined according
to user inputs. This stage defines also what kind of physics has to be
2.2. CFD modelling and automation
solved and how. In other words, the model setup is the element of the
workflow that defines what kind of ‘experiment’ we want to virtu­
The CFD workflow for ship hydrodynamics problems is made of four
alize. In hull hydrodynamics studies there are three main types of
main elements, schematically shown in Fig. 2.a:

5
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Python function, named evolve.py, containing a set of rules, identified by

the authors and based on their experience. These rules are adopted to
discern positive step completion, sufficient to evolve to the next step,
from negative ones related to an early workflow stop with information
reporting to the end user about the issue occurred. The presence of this
function is necessary due to the numbers of possible causes that can
prevent the computational workflow to end with success. On top of this
list, we have meshing issues. The meshing step is by far the most critical
step of the workflow since from this step depends most of our workflow
success. For this reason, as discussed in the next sections of the paper
and in Appendix B we designed the meshing procedure to ensure the
most robust outcome as possible.
In LincoSim the computational back-end of all these four workflow
steps are designed using only open-source libraries. More precisely, the
CFD undergoing engine is developed using the OpenFOAM toolbox
(version 6.0, see Weller et al., 1998). OpenFOAM is a state-of the-art
open-source toolbox that enables to solve the 3D Navier-Stokes equation
including dynamic mesh motion (N-DoF, N ¼ 0,1,2) in a cost effective
way. The solution of hull hydrodynamics involves the solving procedure
of a set of differential equations for a full 3D domain. The incompressible
Navier Stokes for two phases (air/water) with interface tracking and
Fig. 5. Sample of a resulting mesh. Top: overall domain boxes. Bottom Left: capturing and the rigid body dynamics of the hull in equilibrium under
zoom on hull. Bottom Right: zoom on hull boundary layer treatment. the effect of gravity (hull mass), hydrodynamic forces (drag, lift) and
moment must be solved to get information about the hull hydrody­
tests that can be performed for calm waters conditions: so called namics values. Proper numerical schemes and solver setup of all these
‘captive’ or zero degrees of freedom runs (0-DoF), or free sink runs equations are already included and available for direct usage in the
(1-DoF) or free sink and trim runs (2-DoF). OpenFOAM as the standard solver. In our automated workflow, a fixed
� CFD model computing: the CFD model is resolved using a desktop PC numerical solver setup for all our computations was kept as standard. In
and or distributed parallel computers depending on mesh model size particular, the standard interDyMFoam RANS solver equipped with
and desired time to solution. kOmegaSST turbulence model (see Menter, 1994) was used and wall
� Post-processing: the CFD outcomes are processed to get an arbitrary functions selecting very standard boundary conditions for the numerical
set of Key Parameter Indices (KPI) values, tables, plots and diagrams patches are defined in Table 2. For the rigid body dynamics solver, we
that are necessary to support a ranking of the hull performances and used the Newmark solver selecting the velocity integration coefficient
eventually support decision-making designer’s activity. equal to 0.5 and the position integration coefficient equal to 0.25.
Therefore, in LincoSim the CFD modelling parametrization refers
Standardization and parametrization of the user’s inputs are the two only to:
main bricks of the LincoSim way of working.
In Fig. 2.b we represented the technical steps involved in the three � the domain definition;
main automated blocks that are: � the meshing procedure;
� the rigid body dynamics settings;
� Domain Definition and Meshing; � the turbulence and the initial conditions definition.
� CFD Modelling and Equation Solving;
� Data Processing and Visualization. In this way, a sort of standardized setup is available for a given type
of numerical-experiment able to scale up and down with the hull size
From Fig. 2.c it is possible to understand how complex the solutions and adjust other parameters with hull hydrodynamics working
path can be and how, within each intermediate step, a set of necessary conditions.
tests are made before the workflow can be considered eligible to evolve More in detail, the user input used to automate the preprocessing
from one step to the next one. This set of tests is synthesized into a single step is the CAD geometry input file, the water temperature, the hull

Fig. 6. Sample plots of available post-processed free surface. Top: 3D plot of free-surface and hull; bottom left: 2D plot of free surface wave height; bottom right: 1D
plot of free-surface wave height.

6
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 7. CFD vs EFD for each single hull at all the Froude number conditions tested including uncertainty bandwidth where available: a) drag; (b) sinkage; (c) dynamic
trim; (d) wetted surface area (WSA).

velocity, the estimated wave height on the hull and the free-surface and 300, accordingly to the type of wall functions used, we found that
position. The hull length (Lx) defines the computational domain using our solver setup we would be able to obtain better results using a
extension parametrization as shown in Fig. 3. dynamic value of yPlus linking its value to the Froude number as shown
The computational domain meshing is performed by parametrizing a in Fig. 4.
well-defined successive refinement procedure as given by the Open­ In practice, while setting up the meshing strategy, for W3 hull, the
FOAM developers (see CFD Direct Ltd, 2020) for marine hydrodynamics wall function performances have been tested for each Froude number
problems (see also Ponzini and Penza, 2015). Starting from the overall and compared against the experimental data as shown in Fig. 4.a. During
domain extension, a set of nested successive box refinements, together this set of tests, all the meshing parameters were kept constant except for
with a free-surface refinement region and with further hull surface re­ the yPlus value. We tested four meshes, three with fixed yPlus for all
finements, are performed taking advantage of the free-surface input Froude numbers with values of 30, 50 and 150 and one with a dynamic
value. The dimension of the reference external mesh region cell is yPlus value changing with the Froude number as shown in Fig. 4.b. From
defined to match a target mesh cell size on the hull surface necessary to the plots in Fig. 4.a it can be noticed that the variation of the drag dif­
well describe the physical boundary layer region. ferences with respect to the EFD values are quite limited in the dis­
The boundary layer region is instead defined taking into account placing and fully planing regions, while they are becoming more
water temperature and hull velocity information using the ITTC (Inter­ prominent in the semi-planing region. Using this meshing approach, we
national Towing Tank Conference) formula for boundary layer thickness were able to obtain a robust and general meshing strategy obtaining a
evaluation, see ITTC 2015. The number of mesh layers to be used to reasonable number of cells size. The final resulting mesh for a reference
discretize the desired boundary layer thickness is adjusted automatically case hull with an average yPlus value of 100 and a cell mesh size on the
to reach a target yPlus value. While in theory (see Menter et al., 1994 hull of about 0.05 m counts about 1.5 million cells for the semi-hull and
Menter, 1994) the optimal average yPlus value should be set between 30 it is shown in Fig. 5.

7
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 7. (continued).

In appendix B (Table B1), for sake of completeness, verification and � the initial conditions values and the time-step value (using hull ve­
validation set of data is given for two hulls, Mono and W3, at two locity, free-surface position inputs values) by means of standard
different Froude numbers using three meshes with a mesh cell length engineering formulas (see ITTC 2015).
reduction factor of 1.4. Moreover, to show the robustness of our meshing
strategy, in Table B2 also the sensitivity analysis to changes on the For the turbulence values, a standard approach setting all the model
background cell mesh size, keeping the yPlus value fixed, on the same parameters (omega, k and nut) corresponding to a very low value of inlet
set of hulls at the same Froude numbers is reported using five different turbulent intensity, according to the towing tank experiment conditions
meshes. has been adopted. Concerning the time step selection, after testing the
Coherently with the wall functions specification, as shown in two options, adaptive and fixed time step strategy, we noticed that for
Table B2, fixing the value of the yPlus and the number of layer used to sake of robustness, the latter option was preferable and we designed
discretize the physical boundary layer thickness, there is a very low therefore LincoSim with fixed time step option. To ensure Courant
difference in terms of quantitative integral fluid dynamics values in number below one we perform an a-priori analysis of the input, we es­
adopting a finer or coarser mesh cell size at the far field position. We timate and set a value of the integration time step to have Courant at 0.5.
selected to adopt a background mesh cell size that will ensure an average At the end of the simulation, we verify that the average value of the
mesh size between 1.5 and 1.75 million cells for the half hull being this Courant number along the simulation is not exceeding 1.0.
final mesh size coherent to a reasonable time to result at a given degree In order to enable automatic stopping of the simulation a standard­
of parallelism as shown in Table C2. ized criterion that ends the simulation when hull forces and/or rigid
Finally, the CFD model setup parametrization is performed using: body dynamics parameters are stabilized within a certain time window,
depending on the number of the degrees of freedom of the selected
� the input that defines the rigid body dynamics constants (using solver, has been adopted. In particular, for the 0-DoF solver, the vertical
inertia, mass and CoG position input values); and the axial forces components have been monitored while for the 1-
� the water rheological properties (using water temperature input DoF and 2-DoF solver the rigid body dynamics vertical and rotational
value); velocity components have been controlled.

8
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 7. (continued).

Once the automatic stopping criteria arrested the simulation, the 3. Results
post-processing step is also automatically performed on the resulting
final condition. At the very end of the overall workflow, a complete set In order to evaluate LincoSim results quality, a set of analyses per­
of data of the hull at the final position is sampled for post-processing formed using the experimental data of the systematic series of bare
including: 3D hull pressure distribution, 3D water and hull air/water planing hulls described in section 2.1 is given.
interface description, 1D forces and momentum time history, 1D rigid In particular, we evaluated in this section of the paper, as outcomes
body dynamics time history, final mesh distribution on three different of main interest concerning the CFD workflow accuracy:
orthogonal planes. Moreover, data at the final equilibrium status of the
hull need to be processed in order to extract relevant KPI for the ‘designer � the comparison of main hydrodynamics quantities (forces, sink, dy­
user’ that will be therefore available for further analysis and data nav­ namic trim, wetted surface area) as calculated by CFD and measured
igation. Thanks to the OpenFOAM function objects data saving using the by EFD for the four hulls analyzing the whole Froude range;
VTK file format is straightforward to perform and it has been imple­ � the comparison of derived hydrodynamics quantities such as the
mented in LincoSim. The numerical data processing procedure for all the analysis of the pressure area stagnation and whisker spray area an­
data is performed using Python programming language (see Python gles as calculated by CFD and measured by EFD for the four hulls
Software Foundation, 2020; Oliphant, 2006; Langtangen, 2009) and analyzing the available Froude numbers.
dedicated modules such as Numpy (see Oliphant, 2006) and VTK (see
Schroeder et al., 2006). In Fig. 6 a sample of some of the KPI available In Appendix C, for the sake of completeness and as outcomes of
interactive panels from the LincoSim GUI is shown. secondary interest:

� the N-DoF solvers comparison: using two hulls, Mono and W3, at two
velocity conditions (2.3 and 6.32 m/s), we compared the main

9
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 7. (continued).

hydrodynamics quantities (forces, sink, dynamic trim) differences by behaviours experienced by the hull (displacing, semi-planing and
changing the number of degrees of freedom of the solver from zero to planing) are reported in Fig. 8.
two;
� the computational costs: using two hulls, Mono and W3, at two ve­
3.2. Solver accuracy: derived hydrodynamics quantities
locity conditions (2.3 and 6.32 m/s) and the three solvers configu­
rations we analyzed the time to results, the scalability efficiency and
In addition to the analysis of the “directly measured values” pre­
the computational costs are instead analyzed using a single hull
sented in the previous section, we also analyzed a set of derived quan­
model (Mono) at a single velocity condition (6.32 m/s).
tities such as the stagnation line angle (αSL) and whisker spray area
angle (αWS). These values are obtainable from experimental data by
performing additional post-processing of the acquired reference images
3.1. Solver accuracy: main hydrodynamics quantities during the experimental session as explained in Begovic and Bertorello
(2012).
In order to evaluate solver accuracy, main hydrodynamics quantities In this paper, the procedure used for the evaluation of the pressure
such as total drag, sink, dynamic trim and wetted surface area as area stagnation line angle and whisker spray area angle has been applied
calculated by CFD and measured by EFD for the four hulls analyzing the also to CFD images in order to obtain their numerical counterpart. In
whole Froude range analyzed in Begovic and Bertorello (2012) have Fig. 9, the main steps of the image processing procedure used to assess
been compared. The plots shown in Fig. 7 give the comparison for each the two angles of interest starting from EFD data (left column) and from
hull of numerically calculated and the respective measured values CFD data (right column) are shown. Both sets of photos and figures refer
including the experimental uncertainty bandwidth, when available, to the monohedral hull at speed 4.6 m/s and Froude number equal to
using the Froude number value as the x-axis. 1.06.
The plots of the selected quantity for all four hulls together, using the As shown in Fig. 9, to calculate the pressure area stagnation line
Froude number as x-axis and highlighting the three hydrodynamics angle and whisker spray area angle the Rhinoceros© software has been

10
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 8. CFD vs EFD for all the hulls at all the Froude number tested including the highlighting of displacing, semi-planing and planing areas. From top to bottom:
drag; dynamic trim; sinkage; wetted surface area (WSA).

11
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 9. EFD and CFD angles (αSL) and (αWS) evaluation.a) longitudinal view; b) zoomed spray separation; c) Top view; d) Post-processing; e) Projection to 3D model;
f) Projected angles definition.

used. For the purpose of this work only the projected angles αSL and to which extent an automated CFD workflow can be adopted today to
αWS are considered as shown in panel (f) of Fig. 9. perform a reliable hydrodynamic assessment of planing hulls. The
The obtained angle values for all hulls are plotted in Figs. 10 and 11 evaluation of an automated CFD workflow has been performed for a
while in Fig. 12 the ratio of the two angles is presented. systematic series made of four planing hulls previously studied in Froude
number range from 0.2 to 1.73 using standard EFD facilities. A complete
4. Discussion set of quantitative and qualitative hydrodynamic key parameters has
been compared to assess this main question. The results presented in
The main aim of this work was to evaluate the reliability of the sections 3.1 and 3.2 show that thanks to standardization and automa­
automatic workflow design in LincoSim or more in general to establish tion, the virtualized tool is able to correctly predict the trends of all the

12
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 10. Pressure area stagnation line angles plotted over Froude number: left CFD data; right EFD data.

Fig. 11. Whisker spray area stagnation line angles plotted over Froude number: left CFD data; right EFD data.

Fig. 12. Ratio of Whisker Spray and Pressure area stagnation line angles plotted over Froude number: left CFD data; right EFD data.

four hulls in the whole Froude number range. For some relevant values the hull attitude and of the wave patterns agree.
of the Froude number range the CFD outcomes are within the experi­ The main conclusion is that CFD and EFD techniques not only allow
mental uncertainty bound, see Figs. 8 and 9. to find similar trends in the global quantities of interest but also measure
In other words, this case study shows that by using LincoSim, a the underlying physical dynamics of the phenomenon in a similar way.
“designer user” will be able to correctly compare and rank the different In other words, the two measurement tools are nearly equivalent for the
hydrodynamics performances of the four planing hulls. The effect of the studied set of hulls and this is true in particular from a design stage
hull form changing on hydrodynamic performances measured in ex­ perspective where the most important aspect is correctly reproducing
periments (total drag, sinkage and dynamic trim) is correctly matched the trends of the dynamics and of the results.
by CFD data. Analyzing also the results on some selected derived Once agreed that CFD and EFD are two equivalent measurement
quantities, such as pressure area stagnation line angle and the whisker tools for this set of planing hulls we can take advantage of the possibility
spray area angle, shown in Figs. 10 and 11, we can highlight that these to obtain no matter which information from CFD and get a better insight
derived quantities are also well caught by CFD compared to EFD trends about the contribution of the drag components to total drag in the whole
within the complete systematic series. As shown in Fig. 12, for all hulls Froude number range.
and all investigated speeds the value of the ratio of the two analyzed In Fig. 13, on the top and bottom panels, the two drag contributions
angles is lower than 2.0 for both EFD and CFD datasets. It also can be for all four hulls along the whole Froude number range are plotted,
seen that the ratio of two stagnation lines angles for the monohedral hull highlighting displacement, semi-planing and hydrodynamic planing
and W1 are showing a small variation from 1.3 to 1.6, with respect to W2 regimes. It is clear that pressure drag changes are driven by the dynamic
and W3 hulls. More in general, we can state that this ratio, in both EFD trim value (central panel of Fig. 13) by means of a direct linear corre­
and CFD data, is decreasing with the increasing of the deadrise angle for lation and independently from the Froude number value. Instead, the
Froude number up to one. viscous drag is completely independent of the dynamic trim trends and is
Considering also this set of comparisons based on derived quantities, driven by the Froude number value by means of a power correlation. The
it can be appreciated that not only the integral values computed to assess viscous contribution in the whole Froude range is quite similar for all the
the quantitative indices agree but also the qualitative visual aspects of four hulls, and only within the fully planing regime, the differences

13
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 13. From op to bottom: CFD data of pressure drag, dynamic trim and viscous drag contributions over Froude number.

became appreciable showing that for larger deadrise angles this � a relevant amount of computational time saving, of about 50–70%,
contribution is larger. see Table C2, can be achieved when some degree of freedom can be
Another relevant point, as shown in Fig. 14, is that the percentage excluded;
contribution to total drag is similar for all the four hulls along the Froude � the possibility to use an optimal number of computational cores, see
number explored range. During the displacing phase, the hulls drag is Table C3, without bothering for licensing limitations (that for CFD
mainly due to pressure forces contribution (70–80%) while viscous commercial solver at the increased number of cores can be extremely
forces became significantly relevant (>60%) only during the fully severe), is one of the main advantages freely provided by the adop­
planing phase. During the semi-planing phase, there is a clear constant tion of our totally open-source software stack workflow.
shift of the balance between the two contributions from pressure to
viscous as the Froude number increases. There are several limitations on the proposed tool that prevent its use
Notably, the same standardized workflow is adopted for every hull at to study every kind of hulls before a complete validation on a wider
every velocity condition so that either the mesh is topologically stan­ group of hulls can be performed.
dardized either the stopping criteria and the solver setup are coherent Concerning the meshing strategy adopted with a yPlus value driven
allowing for safe data comparison of different hulls at different flow by the Froude number we underline that:
conditions.
Considering also the result presented in Appendix C concerning the � the bandwidth selected to define the three hull behaviour regions
performances and computational costs of the two other solvers, we can (displacing, semi-planing and planing) can be modified since there is
say that: no accordance in the literature on the unique definition of the Froude
number value at which the hull is becoming semi-planing or fully
� the 1-DoF (free of sink only) and the captive ones are in strong planing;
agreement with the 2-DoF solver, see Table C1;

14
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. 14. CFD data of the four hulls pressure and viscous drag percentage contribution to total drag over Froude. From top to bottom, left to right, Mono, W1, W2
and W3.

� in the present validation work we faced only hulls at a model scale. back of the hull. Therefore, the location of transition is quite sensitive to
With full-scale hulls, the meshing schema should work but it must be the choice of wall functions and the other wall boundary conditions
confirmed by performing dedicated validations; limiting robustness and generality, which makes tuning especially
� extreme cases in terms of hull velocity will affect both the Froude difficult from our generic point of view. Moreover, this family of models
number and the Reynolds number and they must be tested to un­ is a low-Re model requiring the near-wall cell to have a yPlus value
derstand the robustness of the proposed meshing topology schema. below 5, ideally below 1 which is very expensive for hull calculations
due to the very high Reynolds numbers involved. Another possibility, to
Concerning the usage of the kOmegaSST turbulence model and wall enhance the viscous drag contribution, when needed, would be to use
functions, the selection has been made in order to have the most general, higher levels of inlet turbulence instead. The main problem with inlet
robust and easy to define setup keeping valid the ability of the selected turbulence is it tends to decay in the region upstream of the hull. Again,
model to capture flow separations, recirculating zones and complex there are ways to alter the models to avoid the undesired decay but they
tridimensional flow patterns (see Menter, 1994). Unfortunately, it is are quite complex and currently not available in OpenFOAM.
known that the selected model, like all other one and two equations Concerning wall roughness treatment, we had a very low reference
turbulence models families, is unable to capture the transition of the value from the hull model scale builder that would be easily usable by
flow whenever it would occur along the hull (see Stern et al., 2015). On the CFD solver. Nevertheless, we decided to keep a zero roughness CFD
the other hand, more recent models, including a larger set of equations modelling (perfectly smooth hull surface) for this first validation. In the
and constants, like kklOmega for instance (see Fürst, 2013), even if able future, we might also add this experimental value when available into
to catch the flow transition, are very sensitive to the wall function the set of input and see if the CFD and EFD comparison will show any
arrangement and to internal model constant calibration. In particular, further improvement.
the omega wall function which tends to delay the transition until the very

15
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

5. Conclusions editing, Formal analysis. Francesco Salvadore: Writing - original draft,

Writing - review & editing, Formal analysis. Ermina Begovic: Writing -
In conclusion, we believe that the results presented herein support original draft, Writing - review & editing, Formal analysis, Data cura­
the concept that by means of standardization and automation of a well- tion. Carlo Bertorello: Writing - original draft, Writing - review &
defined workflow, it is possible to separate CFD knowledge-based editing, Data curation, Formal analysis.
complexities from CFD data benefits obtaining a valuable virtualized
version of more traditional measuring systems. In this perspective, Acknowledgments
LincoSim, by integrating CFD open-source libraries, web UI technolo­
gies, Database management paradigms and HPC infrastructure deploy, CINECA personnel is grateful to the LINCOLN project which received
presents an interesting level of innovation in the field of application funding from the European Union’s Horizon 2020 under grant agree­
even if each single technological brick is not new itself. We believe that ment N. 727982 and to the LINCOLN project partners that helped the
new applications like LincoSim will possibly provide new standards for development of the platform in the present form by means of suggestions
the next generation of engineers and designers. and requests.

CRediT authorship contribution statement

Raffaele Ponzini: Writing - original draft, Writing - review &

Appendix A. Web application front-end design and usage

LincoSim’s main concept is to allow the end user - i.e., the designer user - to focus almost exclusively on the physical case under examination, leaving
all other activities driven by an automatic workflow fully transparent to the user. A brief discussion of the basic elements of the platform architecture,
as sketched in Fig. A1, is presented in the forthcoming.

Fig. A1. Sketch of LincoSim architecture with main roles involved.

From a technological point of view, the automation of high accuracy CFD simulations is far from a simple task. In a typical scenario, it is a matter of
automatically reproducing a complex series of related tasks, in which normally the human contribution is considerable, involving in addition to the
physics of the hydrodynamics problem to be simulated, six basic ingredients:

� CAD positioning and Meshing

� CFD solver setup
� Data processing and visualization
� High performance computing clusters usage
� Database management
� Integration scripting for automation and job control

The management of these competence areas requires broad-spectrum skills, not always owned by the same person or group. This lack of com­
petences is probably a reason why often in typical human-based workflows, despite their usefulness, databases and integration scripts are not used at
all.

16
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

In order to achieve automation, the first step was the standardization of a “case to be simulated”, i.e. to define the set of data from which the
platform is able to build and execute the entire workflow. The standardized LincoSim data set is primarily composed of some physical quantities:

� hull mass
� hull center of gravity
� hull moments of inertia
� hull speed
� water temperature
� draft (initial/estimated)
� trim angle (initial/estimated)

This set joins the geometry of the hull to be loaded and the choice of the simulation setup among a series of predefined proposals. Each simulation
setup corresponds to a different workflow with particular features, e.g. captive, 1 degree of freedom, or 2 degrees of freedom, and potentially
employing different software stack for the fluid-dynamics simulations. At the moment, LincoSim works only with OpenFoam as a solver, but the
architecture makes it possible to connect other applications of this kind.
The only access point for the designer user is the web browser by means of the application front-end. This is a key benefit of the LincoSim platform
with respect to portability and usability since no software has to be installed on the designer user machine except for a very common web-browser.
The web application is organized into two main sections: geometries and simulations.
In the geometry section, it is possible to load the triangulated surface (STL, OBJ) of the hull to be simulated. Upon loading, automatic validation of
the surface is performed, and, if successful, the geometry is displayed and made available when creating a simulation. A navigable table allows the
convenient and orderly management of the geometry database.
In the simulation section it is possible to create a simulation by choosing:

� one of the valid geometry from a drop down list of validated geometries,
� the simulation setup from a drop down list of available setups,
� the HPC cluster to be used from a drop down list of available machines
� the set of the input physical parameters defined above.

Once the simulation request is created, it is possible to submit a single simulation or even a range of simulations in which one of the input pa­
rameters varies in a given range using a certain number of given steps. If the simulation ends successfully, otherwise, an intelligible error is displayed;
the results are accessible from the simulation web page, in the form of tables and several 1D, 2D and 3D charts that propose the results in a form of
direct interest for the designer user. In addition, there is a summary table that allows the user to search and access the database of the simulations,
according to a group-based authorization logic. It is also possible to request comparative graphs of the results of the simulations (e.g., resistance vs
speed of the hull or parallel coordinates comparisons for decision-making). Notably, all simulations data can be downloaded for further analysis, in
VTK file format, CSV file format or images depending on the nature of the data. The front-end communicates to the back-end web-services, which
manage the database, the search engine, and interact with the workload managers of the different HPC clusters configured to launch the simulations.
The submitted jobs perform the workflow defined by the selected simulation setup. A simulation setup is commonly driven by one or more Python
scripts that, starting from the input set of data, perform automatically pre-processing, meshing, solving and post-processing of the data. The building
logic of a simulation setup will be discussed in the next section. For any further technical detail on the LincoSim architecture and implementation, refer
to Salvadore and Ponzini (2019).

Appendix B. Mesh Sensitivity Analysis

When dealing with possible changes of the mesh cell sizes, keeping the mesh topology fixed, we can decide to act:

� on the external cells only;

� on the boundary layer cells only;
� on both of them.

The verification and validation methodology, for instance, the Richardson extrapolation method, suggests modifying both of them by reducing all
the cells length size by a constant factor (ideally 2) to quantify the changes of the relevant differential parameters to establish what is the convergence
ratio, if any, of each monitored quantity.
The meshing automation determines the positioning of the layers surrounding the hull in order to reach an estimated yPlus close to a prescribed
value. It is clear that this value will be achieved within some approximation range and in any case only on average, as the actual values of yPlus at the
different points of the hull depend on the actual conditions of the fluid solution of the simulation. By way of example, Fig. B1 a), b) and c) show the
yPlus values, as extracted from simulation results, corresponding to average values of 69, 42 and 10, respectively. These behaviours are shown as
trends against the axial abscissa for cuts at different distances from the plane of symmetry of the hull. All in all, the trends show important deviations
from the average yPlus values, mainly related to the points of greatest over/under-velocity of the flow with respect to the average speed of the hull.
Given this variability, it is expected that the turbulence model wall functions will work in a non-uniform fashion and this circumstance will produce
effects on the final solution that are very difficult to accurately predict.

17
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. B1. Plot over longitudinal lines at different location of the hull width for three W3 mesh at Froude 0.53: (a) yPlus averaged 69, (b) yPlus averaged 42,(c) yPlus
averaged 10.

18
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Fig. B1. (continued).

For the sake of completeness, however, we have run a mesh sensitivity analysis. This analysis has been performed for two Hulls, Mono and W3, at
two Froude number values for three meshes by varying the size of the side of all mesh cells by a factor equal to 1.4. The data are now in Table B1.

Table B.1
V and V data table. Mono and W3 hull at Froude number 0.53 and 1.46 using three meshes with a mesh cell length reduction factor of 1.4. Mesh CC is coherent to the
mesh used for the average CFD model in LincoSim.
Hull U Froude MESH INFO CFD DATA EFD DATA
Name [m/ Number
Mesh Mesh Mesh BL BL # yPlus Total Pressure Viscous Dynamic Sink Total Pressure Viscous Dynamic Sink
s] [-]
ID [-] Size cells thickness of avg Drag Drag [N] Drag Trim [m] Drag Drag [N] Drag Trim [m]
[mln on [m] layers wet [N] [N] [Deg] [N] [N] [Deg]
cells] hull [-] value
[-] [-]

Mono 2,3 0,53 AA 0,297 11214 0,00940 4 86 31,072 24,002 7070 3720 0,133 31,254 – – 3758 0,135
BB 0,629 17562 0,00938 7 51 30,702 23,760 6944 3804 0,132
CC 1760 31305 0,00940 19 10 30,918 23,052 7860 3677 0,133
6,32 1,46 AA 0,297 11214 0,00372 4 92 54,054 23,178 30,876 3322 0,173 56,172 – – 3,72 0,177
BB 0,589 17562 0,00372 5 71 50,964 21,818 29,146 3495 0,175
CC 1764 39315 0,00377 5 64 53,256 20,572 32,682 3189 0,171
W3 2,3 0,53 AA 0,303 10446 0,00950 4 69 38,880 32,556 6326 3745 0,148 40,71 – – 4132 0,146
BB 0,647 16486 0,00952 7 42 38,992 32,676 6316 3782 0,144
CC 1770 29571 0,00952 19 10 41,872 35,000 6870 4284 0,171
6,32 1,46 AA 0,303 10446 0,00378 4 67 79,722 37,898 41,824 0,883 0190 62,282 – – 2186 0,195
BB 0,615 16586 0,00379 5 50 64,930 27,320 37,610 1267 0,193
CC 1813 37158 0,00380 5 54 61,754 23,334 38,420 1259 0,193

From this sensitivity analysis, we can argue that:2

� Mono hull shows an oscillatory convergence for drag and sink and a non-convergent pattern for trim;
� W3 hull shows a monotonous convergence for the drag and for the sink and an oscillatory convergence for the dynamic trim value.

There are several common traits in the two hulls mesh sensitivity data:

� the sink and the resistance are quite good;

� the trim is the more problematic quantity.

19
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

It is anyway worthwhile underlining that the trim is always the most critical quantity to assess numerically especially when large changes are
present as for planing hulls. In our view, for all quantities we see that the error between CFD and EFD is limited and included within the uncertainty of
the numerical data that is below 2% for the Mono hull and below 9% for W3.
Vice versa, we found it is more meaningful to show how variations of the external cell, i.e. keeping a fixed value of the yPlus set, are not related to
relevant variations in the accuracy of the solution. This last approach, summarized in Table B2, is in our view more meaningful to show how the RANS
CFD modeling proposed is robust in estimating the different KPIs being this last feature one of our main targets. In fact, LincoSim is devised to support
early performance hull evaluation, and our main target is to support a correct and standardized ranking between different hull designs under the same
flow conditions.

Table B.2
Mesh sensitivity to cell size at yPlus give table. Mono and W3 hull at Froude number 0.53 and 1.46 using five meshes with fixed yPlus value and changing the
background mesh cell size. Mesh C is coherent to the mesh used for the average CFD model in LincoSim.
Froude MESH INFO CFD DATA EFD DATA
Number
Mesh Mesh Mesh BL BL # of yPlus Total Pressure Viscous Dynamic Sink Total Pressure Viscous Dynamic Sink
[-]
ID [-] Size cells on thickness layers avg wet Drag Drag [N] Drag [N] Trim [m] Drag Drag [N] Drag [N] Trim [m]
[mln hull [-] [m] [-] value [N] [Deg] [N] [Deg]
cells] [-]

0,53 A 0,905 18253 0,00944 19 10 30,54 23 7,54 3,66 0,138 31,254 – – 3758 0,135
B 1,32 24785 33,178 25,26 7,9 4066 0,137
C 1,76 31305 30,918 23,052 7,86 3677 0,133
D 2,35 39315 31,794 23,692 7866 3789 0,132
E 2,87 45292 30,322 22,41 7,91 3577 0,135
1,46 A 0,637 18253 0,00376 5 65 52,856 20,936 31,922 3146 0,172 56,172 – – 3,72 0,177
B 0,943 24785 52,28 20,438 31,842 3132 0,172
C 1,32 31305 52,03 21,192 30,838 3431 0,175
D 1,76 39315 53,256 20,572 32,682 3189 0,171
E 2,22 45292 51,144 19,5 31,644 3188 0,173
0,53 A 0,887 17125 0,00952 19 10 43,76 37,254 6,5 4,56 0,158 40,71 – – 4132 0,146
B 1303 23058 42,122 35,522 6,59 4366 0,155
C 1,77 29571 41,872 35 6,87 4284 0,171
D 2,06 37158 41,32 34,11 7,21 4115 0,151
E 2,35 42665 41,136 32,364 7,23 4039 0,149
1,46 A 0,645 17125 0,00379 5 50 61,95 24,4 37,55 1,34 0,193 62,282 – – 2186 0,195
B 0,968 23058 60,6 22,9 37,67 1,35 0,193
C 1,35 29571 61,95 24,51 37,44 1,34 0,193
D 1,81 37158 61,75 23,33 38,42 1259 0,193
E 2,27 42665 60,2 22,99 37,21 1,35 0,193

Appendix C. Solvers Comparison computational costs and solver parallel performances

For the sake of completeness, we present in this section, two different hulls, Mono and W3, at two velocity conditions (2.3 and 6.32 m/s),
equivalent to Froude numbers of 0.53 and 1.46, and their main hydrodynamic quantities’ differences after changing the solver number of degree of
freedom from zero to two. The model solver main parameters information and the results of the different configurations are given in Table C1 with
their reference EFD data. In this set of tests, we have to compare:

� from 2-DoF to 1-Dof: drag and sinkage, since the dynamic final trim is given and fixed;
� from 2-DoF to 0-DoF: drag, since both the sinkage and the dynamic final trim are given and fixed.

Notably, the impact of changes of the three solver N-DoF options on the fluid dynamics parameters is very limited as expected. In fact, by playing
with the DoF solver options we are just limiting the number of the allowed degree of freedom that the rigid body dynamics solver have. All other CFD
model parameters, including the set of equations that we are solving, is unchanged. Therefore:

� moving from the 2-DoF solver to the 1-DoF solver option (only free of sink), if the user is able to provide a correct value for the trim, the final sink
value will be correctly found by the solver and all other CFD outcomes will be comparable to the outcomes obtained with a more general solver;
� moving from the 2-DoF solver to the 0-DoF solver option (hull is under captive condition), if the user is able to provide a correct value for the sink
and the trim, all the CFD outcomes will be comparable to the outcomes obtained with a more general solver.

20
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Table C1
NDoF solver analysis performed on Mono and W3 hull at Froude number equal to 0.53 and 1.46.
Hull U Froude CFD SOLVER MODEL INFO CFD DATA EFD DATA
Name [m/ Number
CASE ID [-] N- IMPOSED IMPOSED Total Pressure Viscous Dynamic Sink Total Pressure Viscous Dynamic Sink
s] [-]
DOF SINK TRIM Drag Drag [N] Drag Trim [m] Drag Drag [N] Drag Trim [m]
[N] [N] [Deg] [N] [N] [Deg]

Mono 2,3 0,53 CAPTIVE 0 YES YES 30,83 23,66 7,17 – – 31,254 – – 3758 0,135
FREESINK 1 NO YES 30,41 23,52 6,90 – 0,135
FREESINKANDTRIM 2 NO NO 30,61 23,59 7,10 3,77 0,132
6,32 1,46 CAPTIVE 0 YES YES 54,19 20,39 33,80 – – 56,172 – – 3,72 0,177
FREESINK 1 NO YES 52,91 20,15 32,76 – 0,171
FREESINKANDTRIM 2 NO NO 52,84 20,37 32,46 3,08 0,172
W3 2,3 0,53 CAPTIVE 0 YES YES 38,64 32,18 6,46 – – 40,71 – – 4132 0,146
FREESINK 1 NO YES 38,63 31,96 6,68 – 0,145
FREESINKANDTRIM 2 NO NO 38,33 31,84 6,48 3,65 0,145
6,32 1,46 CAPTIVE 0 YES YES 61.17 22,16 39,04 – – 62,282 – – 2186 0,195
FREESINK 1 NO YES 61,02 23,34 37.67 – 0.194
FREESINKANDTRIM 2 NO NO 61,15 23,03 38.12 1,29 0.194

The two solver options with N-DoF equal to 1 or 0 can be convenient for very early design option scanning. In fact, different solvers with different
degree of freedom and different input conditions can have very different values in time to result and therefore in final computational costs. To have an
idea, a set of reference comparative values are shown in Table C2 for the three solvers setup available in LincoSim.

Table C2
Computational costs of different solver DoF for the Mono and the W3 hull Froude number equal to 0.53 and 1.46.

Hull U [m/ Froude Number nDoF solver Computational Cores MeshDensity per Physical Time CPU Time [core Wall Time
Name s] [-] [-] [-] Computational Core [s] hours] [hours]

Mono 2,3 0,53 0 32 30.000 2,85 108 3,4

1 32 30.000 2,85 145 4,5
2 32 30.000 2,85 195 6,1
6,32 1,46 0 32 30.000 3,5 441 13,8
1 32 30.000 3,5 652 20,4
2 32 30.000 3,5 975 30,5
W3 2,3 0,53 0 32 40.000 3,5 122 3,8
1 32 40.000 3,5 167 5,2
2 32 40.000 3,5 205 6,4
6,32 1,46 0 32 40.000 3.0 250 7,8
1 32 40.000 3.0 375 11,7
2 32 40.000 3.0 785 24,5

Computational costs and solver parallel performances.

The solution of a ship or planing hull hydrodynamics performed with CFD RANS modelling is, in general, an expensive computational task. This
kind of application can take great advantage of an available HPC platform in many ways:

� the availability of a large number of computational cores interconnected with high performance networks (low latency/large bandwidth) allows
for a split in the computation of the single hull CFD on a set of computational cores (let’s say hundreds) and obtain a lower time to result. Specific
scalability tests can be performed in order to exploit platforms benefits at best by determining the optimal number of mesh cells per computational
core.
� Several simultaneous CFD runs can now be performed thanks to open-source license modelling of the selected solver. This technique can also be
performed using other third-party CFD software but requiring a potentially very large licensing budget.

Thanks to our fully open-source technology development policy, and to the availability of HPC platforms, it was possible to explore both the
possibilities explained above. In order to rank the optimal number of computational cores per mesh problem density, a set of scalability tests fixing the
given CFD problem has been performed, changing only the reference cell size used to build the overall meshing topology. The different problem size
test and the given best parallel performance as obtained on our HPC architecture (see Haocheng et al., 2011 for any technical detail) in terms of
efficiency, including the mesh density per computational core that could be used to evaluate similar numerical problem scale up/down performances
are reported for the CINECA hardware, see Cineca, in Table C3.

Table C3
Scalability tests performed on W3 hull at Froude number 1.46.

Mesh Size [mlnCells] Computational Cores [-] MeshDensity per Computational Core [kcells] Efficiency [%]

1.3 64 20 99
6.1 128 47 97
16.7 384 40 82

21
R. Ponzini et al. Ocean Engineering 210 (2020) 107387

Appendix D. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.oceaneng.2020.107387.

References Mansoori, M., Fernandes, A., Ghassemi, H., 2017. Interceptor design for optimum trim
control and minimum resistance of planing boats. Appl. Ocean Res. 69, 100–115.
https://doi.org/10.1016/j.apor.2017.10.006.
Begovic, E., Bertorello, C., 2012. Resistance assessment of warped hullform. Ocean Eng.
Marco, A.D., Mancini, S., Miranda, S., Scognamiglio, R., Vitiello, L., 2017. Experimental
56, 28–42. https://doi.org/10.1016/j.oceaneng.2012.08.004.
and numerical hydrodynamic analysis of a stepped planing hull. Appl. Ocean Res.
Begovic, E., Bertorello, C., Prpic-Orsic, J., 2013. Roll decay assessment of warped hull
64, 135–154. https://doi.org/10.1016/j.apor.2017.02.004.
forms. In: ASME 2013 32nd International Conference on Ocean. Offshore and Arctic
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering
Engineering OMAE2013.
applications. AIAA J. 32, 1598–1605. https://doi.org/10.2514/3.12149.
Brizzolara, S., Serra, F., 2007. Accuracy of CFD Codes in the Prediction of Planing
Metcalf J, B, Faul, L, et al., 2005. Resistance tests of a systematic series of U.S. Coast
Surfaces Hydrodyamic Characteristics.
Guard planing hulls. Report No. NSWCCD-50-TR-2005/063.
Brizzolara, S., Vernengo, G., 2016. A three-dimensional vortex method for the
Molchanov, B., Lundmark, S., Fürth, M., Green, M., 2019. Experimental validation of
hydrodynamic solution of planing cambered dihedral surfaces. Eng. Anal. Bound.
spray deflectors for high speed craft. Ocean Eng. 191, 106482. https://doi.org/
Elem. 63, 15–29. https://doi.org/10.1016/j.enganabound.2015.10.008.
10.1016/j.oceaneng.2019.106482.
Broglia, R., Durante, D., 2017. Accurate prediction of complex free surface flow around a
Najafi, A., Nowruzi, H., Karami, M., Javanmardi, M., 2019. Experimental investigation of
high speed craft using a single-phase level set method. Comput. Mech. 62, 421–437.
the wetted surfaces of stepped planing hulls. Ocean Eng. 187 https://doi.org/
https://doi.org/10.1007/s00466-017-1505-1.
10.1016/j.oceaneng.2019.106164, 2019.
Caponnetto, M., 2001. In: Hamburg, V. Bertram (Ed.), Practical CFD Simulations for
Niazmand Bilandi, R., Dashtimanesh, A., Tavakoli, S., 2020. Hydrodynamic study of
Planing Hulls, Second International EuroConference on High Performance Marine
heeled double-stepped planing hulls using CFD and 2DþT method. Ocean Eng. 196
Vehicles, HIPER 01, pp. 128–138.
https://doi.org/10.1016/j.oceaneng.2019.106813.
CFD Direct Ltd. OpenFOAM documentation. Accessed: 2020-01-15. https://cfd.direct
Oliphant, T., 2006. Guide to NumPy.
/openfoam/documentation/.
Ponzini, R., Penza, A., 2015. An Italian twist on an America’s cup sailing boat hull
Cineca. Galileo HW description. Accessed: 2020-01-15. https://www.cineca.it/it/cont
design. In: Osseyran, A., Giles, M. (Eds.), Industrial Applications of High-
ent/galileo.
Performance Computing: Best Global Practices. CRC Press, pp. 279–284 (chapter
Cucinotta, F., Guglielmino, E., Sfravara, F., 2017. An experimental comparison between
20).
different artificial air cavity designs for a planing hull. Ocean Eng. 140, 233–243.
Python Software Foundation, 2020. Python language reference, 01-15. http://www.pyt
https://doi.org/10.1016/j.oceaneng.2017.05.028.
hon.org.
De Luca, F., Pensa, C., 2012. Experimental investigation on conventional and
Radoj�ci�c, D., 2019. Reflections on Power Prediction Modeling of Conventional High-
unconventional interceptors. Trans. Roy. Inst. Nav. Architect. Part B: Int. J. Small
Speed Craft. Springer International Publishing. https://doi.org/10.1007/978-3-319-
Craft Technol. 154, B65–B72. https://doi.org/10.3940/rina.ijsct.2012.b2.129.
94899-7.
De Luca, F.D., Pensa, C., 2017. The Naples warped hard chine hulls systematic series.
Reichel, M, Minchev, A, et al., 2014. Trim Optimisation - Theory and Practice. the
Ocean Eng. 139, 205–236. https://doi.org/10.1016/j.oceaneng.2017.04.038.
International Journal on Marine Navigation and Safety of Sea Transportation 8 (3).
Dumortier, C., Bonnet, J.F., R� egnier, N., Ousten, Y., 2019. Comparing results of a power
https://doi.org/10.12716/1001.08.03.09.
prediction tool with measured data from a series of 35 boats. Ocean Eng. 178,
Salvadore, F., Ponzini, R., 2019. LincoSim: a web based HPC-cloud platform for
501–516. https://doi.org/10.1016/j.oceaneng.2018.12.076.
automatic virtual towing tank analysis. J. Grid Comput. 17, 771–795. https://doi.
Fürst, J., 2013. Numerical simulation of transitional flows with laminar kinetic energy.
org/10.1007/s10723-019-09494-y.
Eng. Mech. 20, 379–388.
Salvadore, F., Ponzini, R., Arlandini, C., 2019. Improving the productivity of hull
Garland R, W, 2011. Stepped Planing Hull Investigation. Society of Naval Architects and
designers with HPC in the cloud: the LincoSim experience. In: 2019 IEEE
Marine Engineers 448–458.
International Conference on Systems, Man and Cybernetics (SMC). IEEE. https://doi.
Gatin, I., Vukcevic, V., Jasak, H., 2019. Fully automated CFD analysis for full-scale ship
org/10.1109/smc.2019.8914462.
hydrodynamics. In: 4th Hull Performance & Insight Conference. HPIC, Gubbio,
Savitsky, D., 1964. Hydrodynamic design of planing hulls. Mar. Technol. 1, 1.
2019.
Schroeder, W., Martin, K., Lorensen, B., 2006. Visualization Toolkit: an Object-Oriented
Haocheng, F., Mingqiang, L., Hu, L., Zhe, W., 2011. A web-based software framework for
Approach to 3D Graphics, fourth ed. (Kitware).
aircraft design modeling, analysis and multidisciplinary optimization. Procedia
Seo, J., Choi, H.K., Jeong, U.C., Lee, D.K., Rhee, S.H., Jung, C.M., Yoo, J., 2016. Model
Environ. Sci. 11, 291–296. https://doi.org/10.1016/j.proenv.2011.12.046.
tests on resistance and seakeeping performance of wave-piercing high-speed vessel
He, P., Filip, G., Martins, J.R., Maki, K.J., 2019. Design optimization for self-propulsion
with spray rails. Int. J. Nav. Architect. Ocean Eng. 8, 442–455. https://doi.org/
of a bulk carrier hull using a discrete adjoint method. Comput. Fluid 192, 104259.
10.1016/j.ijnaoe.2016.05.010.
https://doi.org/10.1016/j.compfluid.2019.104259.
Stern, F., Wang, Z., Yang, J., Sadat-Hosseini, H., Mousaviraad, M., Bhushan, S., Diez, M.,
Huang, L., Ren, K., Li, M., Tukovi�c, Z.,
� Cardiff, P., Thomas, G., 2019. Fluid-structure
Sung-Hwan, Y., Wu, P.C., Yeon, S.M., Dogan, T., Kim, D.H., Volpi, S., Conger, M.,
interaction of a large ice sheet in waves. Ocean Eng. 182, 102–111. https://doi.org/
Michael, T., Xing, T., Thodal, R.S., Grenestedt, J.L., 2015. Recent progress in CFD for
10.1016/j.oceaneng.2019.04.015.
naval architecture and ocean engineering. J. Hydrodyn. 27, 1–23. https://doi.org/
International Towing Tank Conference Association. ITCC friction formula. Accessed:
10.1016/s1001-6058(15)60452-8.
2020-01-15. https://ittc.info/media/2021/75-02-02-02.pdf.
Sukas, O.F., Kinaci, O.K., Cakici, F., Gokce, M.K., 2017. Hydrodynamic assessment of
Judge, C., Mousaviraad, M., Stern, F., Lee, E., Fullerton, A., Geiser, J., Schleicher, C.,
planing hulls using overset grids. Appl. Ocean Res. 65, 35–46. https://doi.org/
Merrill, C., Weil, C., Morin, J., Jiang, M., Ikeda, C., 2020. Experiments and CFD of a
10.1016/j.apor.2017.03.015.
high-speed deep-V planing hull––Part I: calm water. Appl. Ocean Res. 96 https://doi.
Taunton, D., Hudson, D., Shenoi, R., 2010. Characteristics of a series of high speed hard
org/10.1016/j.apor.2020.102060, 2020.
chine planing hulls - part 1: performance in calm water. Int. J. Small Craft Technol.
Khazaee, R., Rahmansetayesh, M.A., Hajizadeh, S., 2019. Hydrodynamic evaluation of a
152, 55–75. https://eprints.soton.ac.uk/172717/.
planing hull in calm water using RANS and Savitsky’s method. Ocean Eng. 187
Weller, H.G., Tabor, G., Jasak, H., Fureby, C., 1998. A tensorial approach to
https://doi.org/10.1016/j.oceaneng.2019.106221, 2019.
computational continuum mechanics using object-oriented techniques. Comput.
Kim, D.J., Kim, S.Y., You, Y.J., Rhee, K.P., Kim, S.H., Kim, Y.G., 2013. Design of high-
Phys. 12, 620. https://doi.org/10.1063/1.168744.
speed planing hulls for the improvement of resistance and seakeeping performance.
White, G., Beaver, W., Vann, D., 2012. An experimental analysis of the effects of steps on
Int. J. Nav. Architect. Ocean Eng. 5, 161–177. https://doi.org/10.2478/ijnaoe-2013-
high speed planing boats. In: 3rd Chesapeake Power Boat Symposium.
0124.
Yang, D., Sun, Z., Jiang, Y., Gao, Z., 2019. A study on the air cavity under a stepped
Korambath, P., Wang, J., Kumar, A., Davis, J., Graybill, R., Schott, B., Baldea, M., 2016.
planing hull. J. Mar. Sci. Eng. 7, 468. https://doi.org/10.3390/jmse7120468.
A smart manufacturing use case: furnace temperature balancing in steam methane
Yousefi, R., Shafaghat, R., Shakeri, M., 2014. High-speed planing hull drag reduction
reforming process via kepler workflows. Procedia Comput. Sci. 80, 680–689. https://
using tunnels. Ocean Eng. 84, 54–60. https://doi.org/10.1016/j.
doi.org/10.1016/j.procs.2016.05.357.
oceaneng.2014.03.033.
Kowalyshyn, D., Metcalf, B., 2007. A USCG systematic series of high speed planing hulls.
Zill, T., Ciampa, P.D., Nagel, B., 2013. A collaborative MDO approach for the flexible
Trans. - Soc. Nav. Archit. Mar. Eng. 114, 268–309.
aircraft. In: 54th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and
Langtangen, H.P. (Ed.), 2008. Python Scripting for Computational Science. Springer
Materials Conference. American Institute of Aeronautics and Astronautics. https://
Berlin Heidelberg. https://doi.org/10.1007/978-3-540-73916-6.
doi.org/10.2514/6.2013-1677.
Lee, E., Pavkov, M., McCue-Weil, L., 2014. The systematic variation of step configuration
Zou, Lu, Jiang, Sun, Li, 2019. Experimental and numerical research on the influence of
and displacement for a double-step planing craft. J. Ship Prod. Des. 30 https://doi.
stern flap mounting angle on double-stepped planing hull hydrodynamic
org/10.5957/JSPD.30.2.130040.
performance. J. Mar. Sci. Eng. 7, 346. https://doi.org/10.3390/jmse7100346.
LINCOLN Consortium, 2020. LINCOLN Project web site, 01-15. http://www.lincolnpr
oject.eu/.

22