Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

Received: Added at production Revised: Added at production Accepted: Added at production

DOI: xxx/xxxx

RESEARCH ARTICLE

Development, Verification, and Validation of Comprehensive


Acoustic Fluid-Structure Interaction Capabilities in an
Open-Source Computational Platform
Somayajulu L. N. Dhulipala*1 | Chandrakanth Bolisetti2 | Lynn B. Munday1 | William M.
Hoffman2 | Ching-Ching Yu3 | Faizan Ul Haq Mir3 | Fande Kong4 | Alexander D.
Lindsay4 | Andrew S. Whittaker3

1 Computational Mechanics and Materials,


Idaho National Laboratory, Idaho Falls, ID Summary
83402, USA
2 Advanced Reactor Technology and Design, The acoustic fluid-structure interaction (FSI) formulation is a practical numerical
Idaho National Laboratory, Idaho Falls, ID approach for the seismic analysis of fluid-filled tanks. However, there are no veri-
83402, USA fication and validation studies reported in the literature that demonstrate the ability
3 Department of Civil, Structural, and

Environmental Engineering, University at of an acoustic FSI numerical model to predict responses important to structural
Buffalo, Buffalo, NY 14260, USA and mechanical design for intense translational and rotational earthquake inputs.
4 Computational Frameworks, Idaho National
Herein, an acoustic FSI formulation is implemented in the open-source Multiphysics
Laboratory, Idaho Falls, ID 83402, USA
Object-Oriented Simulation Environment (MOOSE), and is formally verified and
Correspondence
validated using analytical solutions and code-to-code verification, and experimen-
*Dhulipala, S.L.N., Computational Scientist
in Uncertainty Quantification, tal data, respectively. The analytical solutions are for small amplitude, unidirectional
Computational Mechanics and Materials seismic inputs. The code-to-code verification utilizes a previously verified and val-
Department, Idaho National Laboratory,
idated Arbitrary Lagrangian-Eulerian (ALE) numerical model in the commercial
Idaho Falls, ID 83402, USA. Email:
Som.Dhulipala@inl.gov finite element code LS-DYNA. The validation studies utilize a comprehensive data
set assembled from results of 3D earthquake-simulator tests of a fluid-filled vessel.
The acoustic numerical model in MOOSE is verified and validated for hydrodynamic
pressures and support reactions except for cases that involve significant convective
response. For small amplitude inputs, numerically predicted wave heights match
those of the analytical solutions. The numerical model is not verified and vali-
dated for wave height calculations under intense 3D seismic inputs. The run times
for the acoustic FSI simulations in MOOSE are an order of magnitude, or more,
shorter than for the corresponding ALE simulations in LS-DYNA. The utility of
the MOOSE acoustic FSI implementation is demonstrated by seismic analysis of a
building equipped with a fluid-filled, advanced nuclear reactor.

KEYWORDS:
Fluid-structure interaction, Multiphysics simulations, Experimental validation, Advanced nuclear reac-
tors, Seismic analysis, Numerical model verification
2 Dhulipala ET AL

1 INTRODUCTION

Fluid-structure interaction (FSI) is a phenomenon in which, a fluid and a structure with interdependent responses share an
interface 1 . Seismic FSI analysis is important when fluid and structure have a shared boundary and are subjected to earthquake
shaking 2,3 . In such cases, FSI modeling and simulation become an integral part of the design and risk assessment of facilities,
especially, nuclear facilities as they usually include many liquid-filled containers 4 , some of them being critical to the safety of
the facility. Additionally, verification and validation (V&V) of numerical models is a critical step toward understanding their
benefits and limitations 5 . In the context of nuclear power plants V&V is imperative for regulatory compliance, which is required
for plant licensing. Over the last few decades, various technical disciplines have proposed and used different definitions for
V&V. Oberkampf et al. 6 and Oberkampf and Trucano 7 summarize the evolution of these definitions and their applicability to
engineering modeling and simulation. They use the following definitions, originally proposed by the Department of Defense’s
Defense Modeling and Simulation Office (DMSO) and the American Institute of Aeronautics and Astronautics (AIAA): “ver-
ification involves determining that a model implementation accurately represents the developer’s conceptual description” and
“validation involves determining the degree to which a model is an accurate representation of the real world” 7,6 . In the con-
text of modeling and simulation of physical systems, such as seismic FSI in this study, verification determines the accuracy of
a computational solution against an analytical solution or a highly accurate numerical solution, and validation determines how
accurately the computational solution compares with experimental data 7 . This study involves the V&V of an acoustic FSI for-
mulation, with an emphasis on seismic loading. An acoustic FSI formulation simulates, under a linearity assumption, both the
impulsive and convective response components in the fluid. Whereas impulsive response is generated by the fluid accelerating
with the tank, convective response is generated by the fluid moving vertically to form surface waves 8 . This study also applies
acoustic FSI to analyze the response of a molten-salt nuclear reactor vessel housed in a nuclear power plant building subjected
to multi-directional earthquake input.
Analytical solutions provide an efficient means to predicting FSI responses under earthquake shaking for simple geome-
tries. Jacobsen 9 , Housner 10 , Veletsos and Yang 11 , and Chalhoub and Kelly 12 proposed analytical solutions for liquid-filled,
base-supported cylindrical tanks under uni-directional shaking. These analytical solutions assume that the fluid is ideal (i.e.,
incompressible, inviscid, and irrotational) and behaves linearly. Yu and Whittaker 13 provide a review of analytical solutions
to seismic FSI in base-supported cylindrical tanks; these efforts also include correcting calculation errors in some previously
proposed analytical solutions. Yu and Whittaker 8 developed analytical solutions for head-supported tanks based on prior work
on base-supported tanks. Finite element analysis is capable of considering complex geometries, multi-directional input ground
motions, as well as the nonlinearity of the fluid behavior, which may be important when the fluid convective response is domi-
nant 8 . Aribitrary Lagrangian-Eulerian (ALE) 14 and mixed Eulerian-Lagrangian 15 finite-element formulations can account for
the separation at the fluid-structure boundaries. Computational fluid dynamics (CFD) 14,15 numerically solves the Navier-Stokes
equations in the fluid domain, thus capturing the fluid nonlinearity represented by the nonlinear convective term in the equation.
In addition to numerical studies, studies such as Pal et al. 16 , Goudarzi et al. 17 , Park et al. 18 , and Compagnoni and Curadelli 19
included experiments for the validation of numerical models. More recently, Mir et al. 5 conducted experiments on liquid-filled
cylindrical tanks subjected to multi-directional earthquake shaking.
An acoustic FSI formulation is an efficient numerical approach for predicting the seismic response of liquid-filled vessels or
tanks. This formulation involves a structural domain and a fluid domain described by an Eulerian mesh and separated by an
interface where equilibrium between the two domains is enforced. The structural domain is modeled using solid mechanics and
the fluid domain is modeled using acoustics, described by temporal and spatial changes in pressure 20 . Acoustic FSI assumes that
the fluid pressure and density are linearly related, the fluid undergoes small deformations, and the shape of the fluid domain does
not change significantly. Due to its linear formulation in an Eulerian description, acoustic FSI can be orders of magnitude faster
than ALE or CFD, while also considering complex fluid/structure geometries and multi-directional seismic inputs. Therefore, it is
well suited for applications that have a negligible fluid flow and small particle displacements, such as acoustic wave propagation,
for example, from seismic or impact loading 21,22,23 . There have also been V&V studies of acoustic FSI for both non-seismic
and seismic applications. For non-seismic applications, Bunting et al. 24 and Bunting and Miller 25 present V&V of acoustic FSI.
For seismic applications, Rawat et al. 26 and Rawat et al. 27 verified and validated the wave heights from acoustic FSI for both
cylindrical and rectangular tanks with numerical and experimental results from Chen et al. 28 under uni-directional shaking.
Furthermore, Rawat et al. 26 verified the pressures in a cylindrical tank with analytical solutions under uni-directional shaking.
Dhulipala ET AL 3

Phan and Paolacci 29 validated the pressure and wave height time series from acoustic FSI with experimental results using one
uni-directional earthquake input.
To the authors’ knowledge, no comprehensive V&V studies exist for acoustic FSI that consider multi-directional seismic
inputs with peak ground accelerations (PGA) ranging up to 1.0g, and systematically study multiple response quantities such
as pressures and wave heights at multiple locations in the domain and the support reactions. These response parameters are
especially important for the design of the tanks (or vessels) and their supports. Additionally, for such multi-directional seismic
inputs, a relative performance evaluation of acoustic FSI and more sophisticated formulations such as ALE is lacking. The study
presented in this paper aims to develop and demonstrate a comprehensively V&V’d tool for the seismic analysis of fluid-filled
tanks, especially for (but not limited to) advanced nuclear applications such as molten-salt and molten-fuel reactors. This tool
is developed in the Multiphysics Object-Oriented Simulation Environment 30 (MOOSE), which is an open-source multiphysics
finite element/volume platform for various science and engineering problems. MOOSE includes a wide range of physics capa-
bilities that can be easily coupled and therefore, development in MOOSE can enable acoustic FSI to be coupled with any of these
physics and efficiently perform multiphysics simulations for challenging applications (e.g., dams 31 ). As such, MOOSE is widely
used for computational modeling and simulations 30 and the development of monolithic acoustic FSI capabilities in an open-
source framework for seismic analysis can greatly benefit the earthquake engineering and other natural hazards communities,
in addition to the nuclear engineering community. Specific contributions of this paper are:

• Acoustic FSI capabilities, which previously did not exist in MOOSE, were implemented in the software framework.
This implementation performs monolithic FSI analysis, where differential equations for the acoustic fluid domain and
the structural domain are solved as a single, monolithic system, while ensuring equilibrium at the interface. Monolithic
solutions offer superior efficiency and robustness as they do not require iterations between the two domains or introduce
any domain splitting errors. The implementation also circumvents a common shortcoming of the monolithic method: large
and poorly-conditioned matrices due to the differences in the magnitudes of the stiffnesses and the variables in the two
domains. This paper also documents and describes the syntax for using acoustic FSI in MOOSE to support not only the
reproducibility of results in this paper, but also, its use for problems beyond those discussed herein.

• A comprehensive verification of the acoustic FSI tool against analytical solutions from simple 2D problems to more com-
plex 3D problems, and benchmarking the acoustic FSI formulation with ALE solutions calculated using the commercial
finite-element software LS-DYNA.

• A comprehensive validation of the acoustic FSI formulation under complex, high-intensity seismic inputs consider-
ing multiple response quantities of interest. The verification, benchmarking, and validation studies are used to provide
preliminary guidance on the usage of acoustic FSI for seismic analysis of fluid-filled tanks.

• An application of the acoustic FSI tool for the seismic analysis of a nuclear power plant building containing a molten-salt
nuclear reactor.

This paper is organized as follows. Section 2 presents the governing equations for acoustic FSI and the free-surface boundary
condition in the fluid. Section 3 briefly discusses the acoustic FSI implementation in MOOSE. Section 4.1 presents test cases
chosen for the V&V of the acoustic FSI formulation. Section 4 discusses the results of the V&V, with analytical solutions from
Yu and Whittaker 8 , and ALE and experimental results from Yu et al. 32 used as a reference for comparison. Section 4.4.3 briefly
discusses the computational efficiency of different numerical techniques for FSI, and makes suggestions for future research.
Section 5 applies the acoustic FSI tool developed in this study in a comprehensive seismic analysis of a nuclear power plant
building equipped with a molten-salt reactor vessel. Section 6 summarizes the study and presents key conclusions.

2 GOVERNING EQUATIONS FOR FSI USING ACOUSTICS

This section provides an overview of the governing equations for the acoustic fluid domain, the structural domain, coupling
between the fluid and the structure, and the free-surface boundary condition.
4 Dhulipala ET AL

2.1 Acoustic fluid domain


The acoustic fluid domain is assumed to be inviscid, irrotational (i.e., the curl of fluid velocity is zero everywhere), and subjected
to small displacements. The momentum equation is given by 33,1,34 :

𝜕 2 𝐮𝑓
𝜌𝑜 + ∇𝑝 = 0 (1)
𝜕𝑡2
where 𝜌𝑜 is the static fluid density, 𝐮𝑓 is the fluid displacement vector, and 𝑝 is the pressure. The continuity equation, assuming
no net mass inflow or outflow, is given by 33,1,34 :
𝜕𝜌𝑓 𝜕𝐮𝑓
+ 𝜌𝑜 ∇ ⋅
=0 (2)
𝜕𝑡 𝜕𝑡
where 𝜌𝑓 is the variable fluid density that is a function of time. The constitutive law for the fluid is given by:

𝑝 = 𝑐𝑜2 𝜌𝑓 (3)
where 𝑐𝑜 is the speed of sound. Substituting the constitutive law in Equation (2) and applying a time derivative gives:

1 𝜕2𝑝 𝜕 2 𝐮𝑓
+ 𝜌 𝑜 ∇ ⋅ =0 (4)
𝑐𝑜2 𝜕𝑡2 𝜕𝑡2
Substituting Equation (1) in Equation (4):
( )
1 𝜕2𝑝 1
+ 𝜌𝑜 ∇ ⋅ − ∇𝑝 = 0 (5)
𝑐𝑜2 𝜕𝑡2 𝜌𝑜
The resulting governing equation is:

1 𝜕2𝑝
− ∇2 𝑝 = 0 (6)
𝑐𝑜2 𝜕𝑡2
Multiplying this governing equation by a scalar test function, 𝜙𝑓 , integrating it over the fluid domain Ω𝑓 , and using the Green’s
formula 35 for the second term, the weak form is calculated as:

1 𝜕2𝑝
𝜙 𝑓 𝑑𝑉 + ∇𝜙𝑓 ⋅ ∇𝑝 𝑑𝑉 = 𝜙 ∇𝑝 ⋅ 𝐧𝑓 𝑑𝐴 (7)
𝑐𝑜2 ∫ 𝜕𝑡2 ∫ ∫ 𝑓
Ω𝑓 Ω𝑓 Γ𝑓
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
Acoustic inertia Diffusion Forcing function
where Γ𝑓 is the fluid boundary and 𝐧𝑓 is a vector normal to the fluid boundary. The three terms (from left to right) are referred
to in this study as “acoustic inertia”, “diffusion”, and the “boundary forcing function”, respectively. The governing equation
for the acoustic fluid domain (Equation (6)) is such that pressure is the only primary variable and this equation is a linearized
version of the Navier-Stokes equations (i.e., the nonlinear convective term is ignored). This linearization entails that, during
the acoustic FSI seismic simulations, there is no fluid flow and the displacement of the fluid particles is small. The transient
displacements and accelerations of the fluid particles can be derived from the pressure histories. Wave heights at the undisturbed
fluid surface, for example, can be calculated from the pressure at the surface of the fluid mesh as discussed in Section 2.4.
However, a linearized fluid domain may mis-represent the wave heights over the fluid under high-intensity seismic motions with
multi-directional inputs, as discussed in the V&V section 4.4. However, the linear acoustic model has several advantages from
a computational viewpoint, such as enabling an efficient Newton solve, rapid model convergence at each time step, and full
coupling of the fluid and structural domains in one solver.

2.2 Linear elastic structural domain


The conservation of linear momentum for the structural domain is given by 1 :

𝜕 2 𝐮𝑠
𝜎 𝑠 = 𝜌𝑠
∇𝜎 (8)
𝜕𝑡2
Dhulipala ET AL 5

where 𝜎 𝑠 is the Cauchy stress tensor and 𝐮𝑠 is the displacement vector. The relationship between stress and displacement is given
by the following constitutive law:

𝜎 𝑠 = 𝐃𝑠 𝜀 𝑠 (9)
where 𝐃𝑠 is the elasticity tensor and 𝜀 𝑠 is the small strain tensor defined as 𝜀 𝑠 = ∇𝐮𝑠 . Multiplying the governing equation by a
vector test function 𝜙𝑠 , integrating it over the structural domain Ω𝑠 , and using Green’s formula, the weak form is given by:

𝜕 2 𝐮𝑠
𝜙 𝑠 ⋅ 𝜌𝑠 𝑑𝑉 + 𝜙𝑠 ) ⋅ 𝜎 𝑠 𝑑𝑉 =
(∇𝜙 𝜙 𝑠 ⋅ 𝜎 𝑠 ⋅ 𝐧𝑠 𝑑𝐴 (10)
∫ 𝜕𝑡2 ∫ ∫
Ω𝑠 Ω𝑠 Γ𝑠
⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟
Inertia Stress divergence Neumann BC
where Γ𝑠 is the portion of the boundary where force or Neumann type boundary conditions are applied with normal vector 𝐧𝑠 . The
three terms (from left to right) are referred to as inertia, stress divergence, and the Neumann boundary condition, respectively.

2.3 Fluid-structure interface coupling


Assuming that the fluid and structure meshes are always attached, continuity of displacement (or acceleration) and traction is
enforced along the interface between the fluid and the structure through the equations 1,36,37 :

𝜕 2 𝐮𝑠 𝜕 2 𝐮𝑓
= , on Γ𝑠𝑓
𝜕𝑡2 𝜕𝑡2 (11)
𝜎 𝑠 ⋅ 𝐧 = −𝑝𝐧, on Γ𝑠𝑓
where Γ𝑠𝑓 is the structural-acoustic interface. The continuity conditions are applied in a weak sense in our finite element imple-
mentation 20,24,25 , as illustrated in Figure 1. Specifically, the acceleration from the structural finite element nodes on the FSI
interface, Γ𝑠𝑓 , are converted to an acoustic pressure flux using Equation (1) and applied as a Neumann boundary condition in
Equation (7) to the faces of the acoustic elements on Γ𝑠𝑓 . Acoustic pressures (i.e., the solution variable in Equation (6)) calculated
at the interface are converted to a traction in the direction of the normal of the interface, and applied as the Neumann bound-
ary condition in Equation (10) to the structural element faces on Γ𝑠𝑓 . The AcousticStructureInterface object in MOOSE
developed in this study and described in Section 3 automates these conversions and enforces the continuity and equilibrium
described in Equation (11) and Figure 1.

2.4 Free-surface boundary condition


When shaken, the fluid domain in a tank experiences convective modes that result in waves on its surface. A free-surface
boundary condition, which is not included in most acoustic FSI formulations for non-seismic applications, is required to simulate
these convective modes and to predict wave heights at the surface. The pressure at the free surface of the fluid due to these waves
is given by:

𝑝 = 𝜌𝑜 𝑔 𝑑𝑤 (12)
where 𝑑𝑤 is the wave height with reference to the initial free surface before applying the dynamic action, and 𝑔 is the acceleration
due to gravity. Wave height can be further expressed as:

𝑑𝑤 = 𝐧𝑓 𝑓 ⋅ 𝐮𝑓 = 𝑢𝑓 ,𝑧
( ) 𝜕𝑢𝑓 ,𝑧 (13)
∇𝑑𝑤 = ∇ 𝐧𝑓 𝑓 ⋅ 𝐮𝑓 =
𝜕𝑧
where 𝐧𝑓 𝑓 is the normal to the fluid free surface, 𝑧 is the vertical coordinate and 𝑢𝑓 ,𝑧 is the normal component of fluid
displacement at the free surface. Taking the first derivative of 𝑝 with respect to 𝑧 in Equation (12) gives:

𝜕𝑝 𝜕𝑢𝑓 ,𝑧
= 𝜌𝑜 𝑔 (14)
𝜕𝑧 𝜕𝑧
6 Dhulipala ET AL

Nodal Boundary Conditions:


! ! u" ! ! u#
= ! on Γ!
!" ! !"
Solid nodal acceleration to
fluid as Neumann pressure
flux BC on element face
Acoustic Structure
fluid 𝜞𝒇 𝜞𝒔
Integrated Boundary
Conditions:
𝜎" # n = −𝑝 # n on Γ"
Fluid nodal pressure to solid as
Neumann normal traction
boundary condition on
element face.

FIGURE 1 Schematic of the transfer of forces at the interface Γ𝑠𝑓 between acoustic fluid and structure. This schematic
summarizes Equation (11), which enforces the acoustic fluid and structure coupling.

At any free boundary in the acoustic domain, the Neumann boundary condition is automatically satisfied (i.e., ∇𝑝 = 0). However,
due to gravity and the resulting surface waves, the surface of the fluid is no longer a free boundary and the vertical component
of ∇𝑝 (i.e., 𝜕𝑧
𝜕𝑝
) will be non-zero. Using the momentum equation, this component can be expressed by:

𝜕𝑝 𝜕 2 𝑢𝑓 ,𝑧
= −𝜌𝑜 (15)
𝜕𝑧 𝜕𝑡2
Equating (14) and (15) and using Equation (12) to express 𝑢𝑓 ,𝑧 in terms of pressure gives 29 :

1 𝜕 2 𝑝 𝜕𝑝
+ =0 (16)
𝑔 𝜕𝑡2 𝜕𝑧
The above equation is the free-surface boundary condition. If Γ𝑓 𝑓 (Γ𝑓 𝑓 ⊂ Γ𝑓 ) is the free-surface boundary of the fluid, another
contribution to the fluid forcing function at this boundary is given by:

𝜕𝑝 1 𝜕2𝑝
𝜙𝑓 𝑑𝐴 = − 𝜙𝑓 2 𝑑𝐴 (17)
∫ 𝜕𝑧 𝑔 ∫ 𝜕𝑡
Γ𝑓 𝑓 Γ𝑓 𝑓

3 IMPLEMENTATION AND USAGE OF THE ACOUSTIC FSI TOOL IN MOOSE

MOOSE is a massively parallel, open-source finite element analysis framework for running high-fidelity multiphysics simula-
tions 30 . MOOSE relies on libMesh 38 for finite elements and on PETSc 39 for numerical solvers. It enables analysts to easily
implement new physics and couple multiple physics in a finite-element framework. MOOSE comprises various modules that
provide basic functionalities across a variety of applications. For example, the TensorMechanics module provides mechanics
and structural dynamics capabilities, the StochasticTools module provides uncertainty quantification capabilities, and the
PorousFlow module provides flow through porous media capabilities. Different modules can be combined and further enhanced
to create applications for specific problems. Developers have built MOOSE-based “apps” that are used in a wide array of appli-
cations such as nuclear fuels performance assessment 40 , structural material degradation modeling 41 , and geothermal systems
modeling 42 . Multihazard Analysis of Stochastic Time-Domain Phenomena (MASTODON) 31 is one such application for seis-
mic analysis and risk assessment of critical infrastructure. In this work, we implemented the acoustic FSI physics in MOOSE,
and used the MOOSE-based app, MASTODON, to perform seismic FSI analysis.
Dhulipala ET AL 7

Detailed descriptions of the software design and applications of MOOSE and MASTODON are provided in Permann et al. 30
and Veeraraghavan et al. 31 , respectively. The design of the acoustic FSI tool implemented in MOOSE as a part of this study is
briefly described here. MOOSE comprises various classes of “systems” that enable applications like MASTODON to perform
different tasks in solving a differential equation using the finite-element method. Three of the most fundamental systems in
MOOSE are the Kernels system, which evaluates volume integrals, the Materials system, which evaluates the constitutive
relationships of the materials, and the BCs system, which evaluates boundary integrals and enforces boundary conditions. When
multiple physics across a boundary need to be coupled (e.g., across the fluid-structure interface in FSI) the InterfaceKernels
system can be used. These MOOSE systems are used to “build” and solve the governing equations presented above. Figure
2 illustrates the Kernels, Materials, and BCs developed in this study to build the acoustic FSI tool in MOOSE. The fluid
domain kernels Diffusion and AcousticInertia (see Equation (7)), along with StressDivergence and InertialForce
kernels in the structural domain (see Equation (10)) are the required kernels (i.e., volume integrals) for running any acoustic
FSI analysis. The speed of sound and density of the fluid domain, and the mechanical properties (e.g., Young’s modulus and
Poisson’s ratio for a linear elastic material) and density of the structural domain are the required material properties for acoustic
FSI analysis. The FreeSurfaceBC, applied at the free surface of the fluid domain, enables simulation of the convective modes in
the fluid. Seismic input can applied to the structural boundaries using the FunctionDirichletBC, which is a simple Dirichlet
boundary condition that can enforce displacement, velocity, or acceleration histories. The AcousticStructureInterface
object, which is an InterfaceKernel type, is developed to couple the fluid and structural domains at the FSI interface and
enforces the displacement and traction continuities described by Equation (11).
In a monolithic FSI system, there is usually an order (or more) of magnitude of difference between the stiffnesses (i.e.,
elements of the Jacobian matrix) of the fluid and structural domains. This results in ill-conditioned systems that lead to poor
model convergence. To remedy this issue, we used the automatic_scaling feature in MOOSE, which scales each variable’s
residual components and Jacobian rows by the inverse of the variable’s largest diagonal Jacobian entry (based on magnitude).
This brings disparate physics onto the same scale, and results in a Jacobian matrix with an acceptable condition number. An
overview of the input file for performing acoustic FSI simulations in MOOSE is presented in Appendix A. Further information
on the use of the FSI module in MOOSE and MASTODON, including documentation and examples, can be found on the
websites, https://mooseframework.inl.gov (MOOSE), and https://mooseframework.inl.gov/mastodon (MASTODON).

Notations
Fluid Structure 𝑝: Pressure
𝜌! : Static fluid density
❶ Kernels 𝒖" : Fluid displacement vector
❶ Kernels
Diffusion StressDivergence 𝑐! : Speed of sound
AcousticInertia InertialForce 𝜌" : Variable fluid density
Γ" : Fluid boundary
❷ Materials Γ"" : Fluid free-surface boundary
❷ Materials 𝐧" : Fluid boundary normal
SpeedOfSound
IsotropicElasticityTensor 𝐧"" : Fluid free-surface normal
Density
Density
𝝈# : Cauchy stress tensor
❸ Boundary cond. ❸ Boundary cond. 𝒖# : Structure displacement vector
FreeSurfaceBC FunctionDirichletBC 𝑫# : Elasticity tensor
𝜺# : Small strain tensor
Γ# : Structure boundary
𝐧# : Structure boundary normal
Interface
AcousticStructureInterface Γ#" : Structure-acoustic interface

FIGURE 2 Description of the kernels, materials, and boundary conditions required for running a seismic FSI analysis using
MOOSE/MASTODON.
8 Dhulipala ET AL

4 VERIFICATION AND VALIDATION OF THE ACOUSTIC FSI FORMULATION

The V&V process is a vital aspect of simulation software development in engineering. For nuclear energy applications, simula-
tion software is required to have a thorough and documented V&V before the software can be used for the design and licensing
of safety-related systems. This section describes a V&V study of the acoustic FSI formulation developed in this study for seismic
analysis of fluid-filled cylindrical tanks. These cylindrical tanks are safety-critical equipment in some advanced reactor designs
such as molten-fuel or molten-salt reactors that need to be designed and qualified for seismic loading.
The goal of any V&V study is to demonstrate credibility that the software accurately solves the mathematical equations
(verification), and can be used to reliably simulate a real physical system (validation). In this study, verification is performed by
simulating simple cases that have analytical solutions and ensuring that the acoustic FSI tool reproduces these solutions. When
analytical solutions are not available, benchmarking is performed, which involves comparing the results with those calculated
from other verified and validated software tools: a code-to-code verification. Validation is performed by simulating a real physical
system with recorded experimental data, comparing the simulation results with this data, and ensuring that software adequately
reproduces the experimental results. This study uses analytical solutions, benchmark solutions, and experimental data from
recent studies involving seismic excitation of fluid-filled tanks for advanced nuclear applications.
The input parameters and response parameters used in the V&V of this study are chosen based on (a) design parameters
for nuclear reactor vessels and their supports, and (b) availability of experimental data. Given that the designs of the reactor
vessels and the supports are governed by wall pressures and reactions, respectively, these response parameters are used herein.
In reactors that have a free fluid surface, wave heights and sloshing will be of importance, especially for the design of reactor
internal components. Although the acoustic FSI formulation may not be an ideal choice for predicting wave heights, the V&V
study of this paper also examines wave heights to investigate the efficiency and accuracy of this formulation in predicting wave
heights, and to develop some guidance on its usage in this context. The experimental program 5 that generated the data used in
this study was also intended for nuclear applications and therefore considers the above parameters. Additionally, the experiments
also include a wide range of seismic inputs (both in terms of acceleration amplitude and directionality, i.e., translational and
rotational) that are representative of nuclear applications.

4.1 Description of the test cases for V&V


We compared the results of acoustic FSI formulation implemented in MOOSE with three categories of reference results: (1)
analytical solutions of simple benchmark problems; (2) analytical solutions from Yu and Whittaker 8 for a head-supported tank;
and (3) results from ALE simulations and experimental results for a base-supported tank from Yu et al. 32 . Category 1 problems,
due to their simplicity, enable us to directly verify the enforcement of Equation (11) at the fluid-structure interface. Category
2 increases the complexity with a uni-directional earthquake acceleration input for which, analytical solutions are available.
Category 3 further increases the complexity to multi-directional earthquake inputs involving both translational and rotational
accelerations at the base of the tank. For category 1, we are interested in the continuity of pressure and stress at the interface. For
categories 2 and 3, we are interested in matching both the time series and peak values of pressures acting on the structure and
wave heights at the fluid surface. For category 3, in addition to the pressures and wave heights, we are interested in the reactions
at the tank support.
Table 1 describes the test cases for V&V across the three categories. Category 1 is a verification study and involves 1-D and
2-D domains comprised of homogeneous fluid and acoustic domains separated by an interface. A half-sine pressure wave is
introduced in the acoustic fluid domain that propagates toward the structure. Exact analytical solutions are available for pressures
and stresses in the domain at different times. Category 2 is a verification study and involves a 3-D head-supported cylindrical
tank with fluid, which is subjected to uni-directional acceleration. The reference solutions are the analytical solutions proposed
by Yu and Whittaker 8 . Category 3 is a validation study and involves a base-supported cylindrical tank partially filled with fluid
and subjected to a multi-directional earthquake acceleration. The reference solutions are experimental results from Yu et al. 32
and Doulgerakis et al. 43 ; comparisons are also made using ALE solutions 32 . Quantities of interest to be verified or validated are
presented in the right-most column in Table 1.
Dhulipala ET AL 9

4.2 Verification with simple benchmark problems


In the benchmark examples, we want to verify that the fluid and structural domains are properly coupled by comparing simulation
results to analytical solutions. The first problem considers wave reflection and refraction between two acoustic media with
different densities. As presented in Figure 3a, this problem consists of 1-D fluid and structural domains, each 100 m in length.
A half-sine 40 Hz sine wave with an amplitude of 0.01 GPa is passed from the fluid to the structure. The fluid domain has a
speed of sound (𝐶𝑜 ) of 1500 m/s and a density (𝜌) of 1000 kg/m3 . The structural domain has a bulk modulus (𝐾) of 2.25 GPa,
a shear modulus (𝐺) of 0 GPa, and a 𝜌 of 4000 kg/m3 (these elastic constants represent a fluid with a 𝐶𝑜 of 750 m/s). When
the pressure wave is passed from the fluid to the structure, there will be reflected and refracted components whose theoretical
amplitudes are given by 0.0033 GPa and −0.0133 GPa, respectively 44 . In the simulation, each domain is discretized into 2000
elements and pressure and stress are recorded at the midpoints in the fluid and structural domains, respectively. A time step of
0.0001 second was used for the analysis. Figure 3b presents results, with the solid blue line representing the fluid pressure and
the dashed orange line representing the stress in the solid. The first peak in the fluid domain is the input pressure wave; the
second is the reflected component with an amplitude of 0.0034 GPa (3% error). The stress amplitude in the solid is 0.0134 GPa
(0.8% error). The amplitude sign is reversed in the structural domain due to the negative sign in Equation (11), which assumes
that compressive stresses are negative.

0.010
Fluid domain
Structural domain

(GPa)
a)
0.005

100 m 100 m Pressure/Stress(GP


Pressure/Stress 0.000
0.01 GPa half
sine pressure
wave
Fluid Structure °0.005
Co = 1500 m/s K = 2.25 GPa
ρ = 1000 kg/m3 G = 0 GPa
°0.010
ρ = 4000 kg/m3

0.00 0.05 0.10 0.15


Time(s)
Time (s)
(a)
(b)

FIGURE 3 (a) Domain of the 1-D benchmark problem consisting of fluid and structural domains, each 100 m in length. A 0.01
GPa half-sine pressure wave of frequency 40 Hz is passed from the fluid to the structure. The solid black circles are the points
where either pressure or stress is recorded. (b) Time series of fluid pressure and structural stress recorded at the monitoring
locations in (a).

The second benchmark problem involves modeling a 1-D pressure wave in the 2-D domain shown in Figure 4a to verify the
continuity of acoustic pressure and normal tractions at the interface. The 2-D fluid and structural domains are each 10 m in
length and 5 m in width. As with the previous problem, a half-sine pressure wave is passed from the fluid domain to the structural
domain. The material properties are also the same as in the first benchmark problem, except that the solid density is now 1000
kg/m3 . This density is lowered to show that when the structure and fluid have the same properties, a pressure wave travelling
from fluid to structure simply passes through the interface without reflection back into the fluid domain. The fluid pressure and
structural stress (compressive stress in the direction normal to the interface) are recorded at the black circle presented in Figure
4a. The simulation uses a time step of 0.0001s. Figure 4b presents results, showing that the pressure and stresses are equal and
opposite, and that in the structural domain (which has fluid-like properties with 𝐺 = 0 GPa), the stresses in both directions (X
and Y) are also equal, which represents a hydrostatic stress state.
10 Dhulipala ET AL

0.010 Pressure
Stress xx
10 m 10 m

(GPa)
a)
y
Stress yy
0.005

Pressure/Stress(GP
0.01 GPa half
sine pressure

Pressure/Stress
wave 0.000
Fluid Structure
Y Co = 1500 m/s K = 2.25 GPa °0.005

ρ = 1000 kg/m3 G = 0 GPa


⍴ = 1000 kg/m3 °0.010
X 0.000 0.005 0.010 0.015 0.020
Time (s)
Time (s)

(b)
(a)

FIGURE 4 (a) Domain of the 2-D benchmark problem consisting of fluid and structural domains, each 10 m in length and 1
m in width. A 0.01 GPa half-sine pressure wave with a frequency of 40 Hz is passed from the fluid to the structure. The solid
black circle represents the monitoring location at which the pressure and stress are recorded. (b) Time series of fluid pressure
and normal structural stresses (X and Y directions) recorded at the monitoring location.

4.3 Verification with analytical solutions under a uni-directional acceleration input


A head-supported, rigid cylindrical tank partially filled with water (𝜌𝑜 = 1000 kg/m3 and 𝐶𝑜 = 1500 m/s) is modeled, and
is subjected to an acceleration input in the X direction. Rigidity is achieved by assigning a Young’s modulus two orders of
magnitude greater than that of carbon steel. Figure 5a presents the dimensions of the tank and the depth of the water in the
tank; the tank thickness is 10 mm. Figure 5b presents the finite element mesh of the tank with fluid composed of 18,304 and
55,200 elements for the structural and fluid domains, respectively. A simulation time step of 0.0025 s is used. Here, the acoustic
FSI solutions and the analytical solutions (Equations (2)-(3) and (9)-(11) in Yu and Whittaker 8 ) for head-supported tanks are
compared. Fluid pressure and wave height time series are calculated at the locations presented in Figure 5a (i.e., P1, P2, and P3
for pressures and W1, W2, and W3 for wave heights). Although the time series are compared at only the specific points presented
in Figure 5a, pressures and wave heights were recorded at multiple locations in the fluid during the simulation for verifying
pressure profiles along the tank depth, and wave height profiles along the radial distance of the tank at certain time instances.
Two input accelerations are considered: (1) an Ormsby wavelet with a peak acceleration of 0.5 g and the four characteristic
frequencies 𝑓1 = 0 Hz, 𝑓2 = 5 Hz, 𝑓3 = 45 Hz, and 𝑓4 = 50 Hz, and (2) a 40 Hz sine wave with a peak acceleration of 0.2 g. (An
Ormsby wavelet 45 is a wavelet whose Fourier spectrum has a constant amplitude within a specified frequency range. It allows
the study of responses of complex systems by considering the contributions from all frequencies within a specified range).
The time series of the pressures on the tank wall and the wave heights at the fluid surface are presented in Figures 6a–6c and
6d–6f, respectively, for the Ormsby wavelet input. The acoustic FSI results calculated using MOOSE and analytical solutions
are almost identical. Figures 7a and 7b enable a comparison of the pressure profiles at 1 s, when the peak pressure occurs
near the left and right ends of the tank, respectively. Both the acoustic FSI and analytical pressure profiles are nearly identical.
Figures 7c and 7d present wave height profiles across the radial distance with respect to the center of the fluid surface at 1.25
and 1.875 s, respectively, when the peaks occur. The figures show that the wave height profiles calculated using the acoustic FSI
and analytical results are virtually identical.
The time series of the pressures on the tank wall and the wave heights at the fluid surface are presented in Figures 8a–8c
and 8d–8f, respectively, for the 40 Hz sine wave. Again, the acoustic FSI results are in excellent agreement with the analytical
solutions. Although there are some discrepancies in the pressure time series of Figure 8a, the peaks are still quite close, with
values of 0.058 and 0.055 kPa for acoustic FSI and the analytical solution, respectively. Figures 9a and 9b enable a comparison
of the pressure profiles at 0.2125 s (when the peak pressure occurs according to Figures 8b and 8c) near the left and right ends
of the tank, respectively. Both the acoustic FSI and analytical pressure profiles are nearly identical. Figures 9c and 9d present
wave height profiles across the radial distance with respect to the center of the fluid surface at 0.2 and 0.4 s, respectively, when
the peaks occur. The figures show that the acoustic FSI and analytical results predict very similar wave heights for the sinusoidal
input.
Dhulipala ET AL 11

Y
R = 0.79 m

W2
W1 W3
P3
Z W1: (-0.7, 0, 1.8)
W2: (-0.5, 0, 1.8)
H = 1.8 m W3: (-0.3, 0, 1.8)
2m

P1: (-0.79, 0, 0.0)


P2 P2: (-0.79, 0, 0.9)
P3: (-0.79, 0, 1.8)

P1
(b)
(a)

FIGURE 5 (a) A head-supported cylindrical tank for verifying the acoustic FSI formulation with analytical solutions proposed
by Yu and Whittaker 8 . The solid black circles and green triangles identify monitoring locations. (b) Finite element mesh of the
tank composed of 18,304 and 55,200 elements for the structural and fluid domains, respectively.

In addition, for the sinusoidal input, we explore the impact of the time step on the solution for wave heights and pressures at
points W1 and P1, respectively, in Figure 5a. We simulated the results considering three different time steps: 0.001s, 0.0025s, and
0.005s. Figures 10a and 10b present the results for the wave heights and pressures, respectively. The results are quite consistent
for all three time steps. This consistency is due to the three time steps having a Nyquist frequency greater than the frequency
of the input sinusoidal motion (i.e., 40 Hz). In general, the chosen time step should be such that the corresponding Nyquist
frequency should be greater than the highest frequency of importance to the analyst. In the subsequent validation analyses, we
chose a time step of 0.002s, which has a Nyquist frequency of 250 Hz. Therefore, the highest frequency that can be successfully
represented in our FSI validation analyses is 250 Hz.

4.4 Benchmarking with ALE solutions and validation with experimental results under a
multi-directional seismic input
Mir et al. 5 conducted experiments by subjecting a cylindrical tank with a fixed base to multi-directional seismic inputs. As
presented in Figure 11a, the cylindrical tank is filled with water that is dyed. The tank is constructed with carbon steel.
Multi-directional seismic inputs were applied to the cylindrical tank using a six-degree-of-freedom earthquake simulator at the
University at Buffalo. Figure 11b presents an image of this earthquake simulator. Figure 12a presents the dimensions of the tank
and the depth of the water; the tank thickness is 7.92 mm. During the experiments, pressures and wave heights were recorded
at the monitoring locations represented by the solid black circles and the green triangles, respectively, in Figure 12a. In Figure
12a, monitoring locations PW1, PW2, and PW3 are for pressures and W1 and W2 are for wave heights.
The base-supported cylindrical tank partially filled with water (𝜌𝑜 = 1000 kg/m3 and 𝐶𝑜 = 1500 m/s) is modeled in MOOSE,
and subjected to a multi-directional seismic input. The tank wall is assumed have a density of 7880 kg/m3 , an elastic modulus
of 200 GPa, and a Poisson’s ratio of 0.27. We are interested in comparing the acoustic FSI solutions with the ALE solutions of
Yu et al. 32 and experimental results of Mir et al. 5 for this problem. Fluid pressure and wave height time series are verified and
validated at the locations presented in Figure 12a. Figure 12b presents the finite element mesh of the tank with fluid composed
of 3917 and 2865 elements for the structural and fluid domains, respectively. We used a simulation time step of 0.002s. Three
cases of seismic inputs (referred to in this paper as NM-1, NM-2, and NM-3; NM here stands for numerical model input), which
have both translational and rotational accelerations, are considered and their peak values are presented in Table 1. Recordings
NM-1, NM-2, and NM-3 are from the El Centro, Hualien, and Chi-Chi earthquakes, respectively 32 , with time compression per
the length scale of the test tank. Figures 13 and 14, respectively, reproduce the time series and 2% damped response spectra of
12 Dhulipala ET AL

Point P3 Point P3
Input Ormsby: Point P2 Point P2
Input Ormsby: Point P1 Point P1
Input Ormsby:
4 4 4
Acoustic
3 Analytical 3
3

(kPa)
(kPa)

Pressure (KPa)
(kPa)
Pressure (KPa)

Pressure (KPa)
2 2 2

Pressure
Pressure

Pressure
1 1 1

0 0 0

0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Time
Time(s)
(s) Time
Time(s)
(s) Time
Time(s)
(s)

(a) (b) (c)


Point W3Point W3
Input Ormsby: Point W2Point W2
Input Ormsby: Point W1Point W1
Input Ormsby:
0.03 0.03 0.03

0.02 0.02 0.02


(m)

(m)

(m)
height(m)

height(m)

height(m)
0.01 0.01 0.01
Waveheight

Waveheight

Waveheight
0.00 0.00 0.00

°0.01 °0.01 °0.01


Wave

Wave

Wave
°0.02 °0.02 °0.02

°0.03 °0.03 °0.03


0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
Time
Time (s)
(s) Time
Time (s)
(s) Time
Time (s)
(s)

(d) (e) (f)

FIGURE 6 Pressure time series for the Ormsby wavelet input at monitoring locations: (a) P3, (b) P2, and (c) P1. Wave heights
time series at monitoring locations: (d) W3, (e) W2, and (f) W1.

the three seismic inputs obtained from Yu et al. 32 . Case NM-1 has the largest amplitudes both in translation and rotation. Cases
NM-2 and NM-3 are seen to have multiple translational and rotational components. The translational components of the input
motions are directly applied to the base of the tank. While applying the rotational components, the tank base is first stiffened by
two orders of magnitude, and vertical accelerations with amplitudes equal to zero at the center of the tank and linearly varying
from the center along x and/or y axes are applied to mimic the rotational motion.
In the following sections, a discussion on the time series results for pressures, wave heights, and moment, followed by a
comparison of the predicted peak response quantities with the experimental results and ALE results. The computational expense
of the different numerical solutions is also discussed.

4.4.1 Comparison of time series


Figure 15 enables a comparison of pressure time series, with the three rows representing points P3, P2, and P1 (see Figure 12a)
and the three columns corresponding to seismic inputs NM-1, NM-2, and NM-3. These pressures are hydrodynamic only. The
figure shows that although the acoustic formulation neglects the nonlinearity in the Navier-Stokes equation and the input motions
are quite complex, there is a good match between the acoustic FSI, ALE, and experimental results. For NM-3, which also has a
vertical translational component, the pressures calculated using the acoustic FSI analysis between 0.5–3 s differ at PW3 (Figure
15g). The differences between the results are less prominent at the tank base (i.e., location PW2: Figure 15h; and location PW1:
Figure 15i), where the pressures are dominated by the impulsive mode. As mentioned previously, the pressures acting on the
tank wall can be divided into convective and impulsive components 8 . Near the surface, the convective response modes, which
are responsible for the waves, contribute the most to the pressure and this contribution attenuates with fluid depth. As discussed
later, it can be difficult to accurately capture the convective modes under certain high intensity seismic inputs, and this may be
especially so under a linearity assumption of the acoustic FSI formulation.
Dhulipala ET AL 13

PressurePressure
Input Ormsby: profile left end left end
profile PressurePressure
Input Ormsby: profile right end right end
profile

1.5 1.5
(m)

(m)
(m)
Height (m)

(m)
Height(m)
1.0 1.0

Height
Height

Height
0.5 0.5
Acoustic
Analytical
0.0 0.0
0 1 2 3 4 °4 °3 °2 °1 0
Pressure
Pressure (kPa)
Pressure(kPa)
(KPa) Pressure
Pressure(kPa)
Pressure (kPa)
(KPa)

(a) (b)
Wave profile
Input Ormsby: Waveatprofile
1.25s at 1.25s Wave profile
Input Ormsby: Waveatprofile
1.875sat 1.875s
0.030 0.030

(m)
0.015 0.015
(m)

height (m)
height(m)

Wave height
Waveheight

0.000 0.000

Wave
Wave

°0.015 °0.015

°0.030 °0.030
°0.70 °0.35 0.00 0.35 0.70 °0.70 °0.35 0.00 0.35 0.70
Radial distance(m)
Radial distance (m) Radial distance(m)
Radial distance (m)
(c) (d)

FIGURE 7 Pressure profiles for the Ormsby wavelet input as a function of tank height at the (a) left and (b) right ends of the
tank at 1 s. Wave height as a function of radial distance at (c) 1.25 and (d) 1.875 s.

Figure 16 enables a comparison of wave height time series, with the two rows representing the points W1 and W2 (see
Figure 12a), and the three columns corresponding to the seismic inputs NM-1, NM-2, and NM-3. The acoustic FSI, ALE, and
experimental time series are similar for NM-2 and NM-3 (NM-2: Figures 16b and 16e; NM-3: Figures 16c and 16f). For NM-1
(Figures 16a and 16d), there is reasonable agreement between the time series between 0–2 s. However, after 2 s, the acoustic FSI
formulation exhibits surface response modes that are absent in the experimental results. The linearity assumption made by the
acoustic formulation coupled with the complexity of the input motion (i.e., high frequency content and multi-directional inputs)
may be the cause for disagreements between the results of acoustic FSI and those of the experiments. Results from the ALE
model are not in good agreement with the experimental results for NM-1 (see between 4–6 seconds in Figure 16a). For NM-3
(Figures 16c and 16f), both the acoustic FSI and ALE predictions differ from the experimental results. Corroborating previous
results, results from acoustic FSI do not match with the experimental pressures near the fluid surface when the convective modes
dominate (see Figure 15c).
The fluid pressures acting on the wall and base of the tank contribute to 𝑀𝑦 . Because all the three seismic inputs considered
include rotational accelerations, there will be additional contributions to the moment about y-axis (𝑀𝑦 ) from the fluid and tank.
As explained in Mir et al. 46 , these contributions are: (1) moment due to weight of the fluid in the rotated configuration of the
tank; (2) moment due to rotation of the tank; and (3) moment due to the tank weight caused by a shifting center of gravity.
Additionally, in the experimental setup, the tank is supported on a square base plate of side 2 m and thickness 0.045 m 32 .
Rotation of this base plate also contributes to 𝑀𝑦 . Figure 17 enables a comparison between 𝑀𝑦 time series, with the three
columns corresponding to NM-1, NM-2, and NM-3. Overall, the time series corresponding to acoustic FSI match satisfactorily
14 Dhulipala ET AL

Point P3
Input Sine: Point P3 Point P2
Input Sine: Point P2 Point P1
Input Sine: Point P1
Acoustic
1 Analytical 1 1

Pressure (KPa)
Pressure (KPa)

Pressure (KPa)

(kPa)
(kPa)

(kPa)
0 0 0

Pressure
Pressure

Pressure
°1 °1
°1

0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
Time
Time (s) Time
Time (s) Time
Time (s)

(a) (b) (c)


Input Point
Sine:W3
Point W3 Input Point
Sine:W2
Point W2 Input Point
Sine:W1
Point W1
0.006 0.006
0.006

(m)
(m)
(m)

0.004

height(m)
height(m)
height(m)

0.004 0.004

Waveheight
Waveheight
Waveheight

0.002

Wave
0.002
Wave

0.002
Wave

0.000 0.000 0.000


0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
Time
Time(s)
(s) Time
Time(s)
(s) Time
Time(s)
(s)

(d) (e) (f)

FIGURE 8 Pressure time series for the 40 Hz sine wave input at monitoring locations: (a) P3, (b) P2, and (c) P1. Wave heights
time series at monitoring locations: (d) W3, (e) W2, and (f) W1.

with experimental results. Although the acoustic FSI 𝑀𝑦 amplitudes are quite close to that of experiments for NM-1 and NM-
2, there are differences for NM-3. The complexity of the NM-3 seismic input, which includes translational accelerations in all
three directions and rotational accelerations about x and y axes, may be responsible for the differences in amplitudes.

4.4.2 Comparison of peak responses


Table 2 presents peak response values for the experiments and the ALE and acoustic FSI simulations. For the acoustic FSI
and ALE results, the absolute percentage error, rounded to the nearest integer with respect to the experiments, is presented
in parentheses. Considering an error threshold of 15%, both acoustic FSI and ALE have three and four cases, respectively, in
which, the error threshold is exceeded (for such cases, the % errors are bolded in the table). Thus, for the 15% error threshold,
the performance of acoustic FSI is close to that of ALE across the response cases considered. Also, the average % errors across
all the cases, and for those cases in which the 15% error threshold is met, respectively, are 11% and 8% for acoustic FSI and
10% and 5% for ALE. The acoustic FSI formulation is linear, and this assumption may lead to an over or under estimation of
fluid pressures and wave heights. Additionally, the fluid in the acoustic domain is assumed to be inviscid, which underestimates
energy dissipation.

4.4.3 Comparison of the computational efficiency of different numerical techniques


In sections 4.4.1 and 4.4.2, the acoustic FSI formulation was validated with experimental results. In general, when predicting
peak response quantities, the absolute % errors with respect to experiments were similar for the acoustic FSI and the ALE
formulation. Prediction of the pressure and wave height time series was also satisfactory, except for more intense earthquakes,
where the prediction of the convective mode using a linear formulation such as acoustic FSI becomes challenging. A clear
benefit of acoustic FSI analysis is its computational expense. Table 3 presents the computational expense of different numerical
Dhulipala ET AL 15

Pressure
Input Sine: profileprofile
Pressure left endleft end Pressure
Input Sine: profileprofile
Pressure right end
right end

1.5 1.5
(m)

(m)
Height(m)

Height(m)
1.0 1.0
Height

Height
0.5 0.5
Acoustic
Analytical
0.0 0.0
0.0 0.5 1.0 1.5 °1.5 °1.0 °0.5 0.0
Pressure(kPa)
Pressure (KPa) Pressure(kPa)
Pressure (KPa)

(a) (b)
WaveWave
Input Sine: profileprofile
at 0.2sat 0.2s WaveWave
Input Sine: profileprofile
at 0.4sat 0.4s
0.0050 0.0050
(m)

(m)
0.0025 0.0025
height(m)

height(m)
Wave height

Wave height
0.0000 0.0000
Wave

Wave
°0.0025 °0.0025

°0.0050 °0.0050

°0.70 °0.35 0.00 0.35 0.70 °0.70 °0.35 0.00 0.35 0.70
Radial distance(m)
Radial distance (m) Radial distance(m)
Radial distance (m)
(c) (d)

FIGURE 9 Pressure profiles for the 40 Hz sine wave input as a function of tank height at the (a) left and (b) right ends of the
tank at 0.2125 s. Wave height as a function of radial distance at (c) 0.2 and (d) 0.4 s.

techniques, considering the three seismic inputs for validation. Although these numerical techniques were run on different
desktop machines of similar computing powers and using different software platforms, the run times are significantly different:
acoustic FSI in MOOSE is one to two orders of magnitude faster than ALE and ICFD, respectively. As discussed in Section
2.1, acoustic FSI assumes a linear fluid domain, while ALE and ICFD model the non-linearity in the fluid domain through a
convective term in the Navier-Stokes equations. Inclusion of this non-linearity will provide a better capability to model fluid
sloshing, but at the expense of much longer run times.
16 Dhulipala ET AL

0.006

1
(m)

(kPa)
height(m)

Pressure(KPa)
0.004
Wave height

Pressure
Wave

0.002 0.001s dt
0.0025s dt
°1
0.005s dt
Analytical
0.000
0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 0.4
Time (s)
(s) Time
Time(s)
(s)
Time
(a) (b)

FIGURE 10 40 Hz sine wave input (a) wave heights at W1 and (b) pressures at P1.

FIGURE 11 Experimental set up at University at Buffalo: (a) cylindrical tank on a stiff base plate, (b) six-degree-of-freedom
earthquake simulator; figures reproduced from Yu et al. 32 .

Y
R = 0.76 m

W1 Z W2

W1: (-0.709, 0, 1.62)


H = 1.62 m

PW3
2m

W2: (0.709, 0, 1.62)


PW2
PW1: (-0.76, 0, 0.305)
PW1 PW2: (-0.76, 0, 0.914)
PW3: (-0.76, 0, 1.524)

(a)
(b)

FIGURE 12 (a) A base-supported cylindrical tank for verifying and validating the acoustic FSI formulation with ALE solutions
and experimental results by Yu et al. 32 . (b) Finite element mesh of the tank and fluid.
Dhulipala ET AL 17

TABLE 1 Test cases for V&V of the acoustic FSI formulation.

Case # Model description Inputs Materials Quantities of interest


(1) Simple benchmark problems
(Two acoustic media)
1-D domain Half-sine pressure Match peak stress
Fluid-like structure
1 Structure: 𝑢𝑥 wave from fluid and pressure with
with 4× the
Fluid: 𝑝 to structure analytical values
density as fluid
(Two acoustic media) Stresses and pressure should
2-D domain Half-sine pressure
Fluid-like structure be equal and opposite
2 Structure: (𝑢𝑥 , 𝑢𝑦 ) wave from fluid
and fluid have at the fluid-structure
Fluid: 𝑝 to structure
identical properties interface
(2) Comparison with analytical solutions from Yu and Whittaker 8
3-D cylindrical tank Ormsby wavelet Pressures on tank wall
Structure is rigid,
3 Structure: (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) applied to tank head and fluid wave heights
fluid is water
Fluid: 𝑝 in X directon compared with 8
3-D cylindrical tank 40 Hz sine wave Pressures on tank wall
Structure is rigid,
4 Structure: (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) applied to tank head and fluid wave heights
fluid is water
Fluid: 𝑝 in X direction compared with 8
(3) Comparison with ALE solutions and experimental results from Yu et al. 32
Pressures on tank wall,
3-D cylindrical tank Input NM-1
Structure is steel, fluid wave heights, and
5 Structure: (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) 𝑎𝑥 (peak: 1.025g)
fluid is water base moment about y-axis
Fluid: 𝑝 𝑟𝑦 (peak: 13.15 rad/s2 )
compared with 32
Input NM-2
Pressures on tank wall,
3-D cylindrical tank 𝑎𝑥 (peak: 0.102g)
Structure is steel, fluid wave heights, and
6 Structure: (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ) 𝑎𝑦 (peak: 0.108g)
fluid is water base moment about y-axis
Fluid: 𝑝 𝑟𝑥 (peak: 0.624g)
compared with 32
𝑟𝑦 (peak: 0.66 rad/s2 )
Input NM-3
𝑎𝑥 (peak: 0.131g) Pressures on tank wall,
3-D cylindrical tank
𝑎𝑦 (peak: 0.145g) Structure is steel, fluid wave heights, and
7 Structure: (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 )
𝑎𝑧 (peak: 0.076g) fluid is water base moment about y-axis
Fluid: 𝑝
𝑟𝑥 (peak: 0.64g) compared with 32
𝑟𝑦 (peak: 0.489 rad/s2 )
Notations. NM: Numerical model; (𝑢𝑥 , 𝑢𝑦 , 𝑢𝑧 ): Displacements along the 𝑥, 𝑦, 𝑧 directions; (𝑎𝑥 , 𝑎𝑦 , 𝑎𝑧 ): Accelerations along
the 𝑥, 𝑦, 𝑧 directions; (𝑟𝑥 , 𝑟𝑦 ): Rotations about the 𝑥, 𝑦 directions; 𝑝: Fluid pressure.
18 Dhulipala ET AL

Input NM-1: translation Input NM-2: translation Input NM-3: translation


1 0.1 0.2
x x

(g)
(g)
0.1 y
Acc. (g)

Acc.(g)
Acc. (g)

Acc.(g)
0 0.0 z

Acc.
Acc.
0.0
x
y
°1 °0.1 °0.1
0 2 4 6 0 2 4 6 0 2 4 6
Time (s)
Time (s) Time(s)
Time (s) Time(s)
Time (s)

(a) (b) (c)


Input NM-1: rotation Input NM-2: rotation Input NM-3: rotation
10 0.5 0.5
Acc. (rad/s )
Acc. (rad/s*s)

(rad/s2)
(rad/s2)

Acc.(rad/s*s)
Acc.(rad/s*s)
2

0 0.0 0.0
rx

Acc.
rx
Acc.

°10 ry °0.5 ry °0.5 ry

0 2 4 6 0 2 4 6 0 2 4 6
Time (s)
Time (s) Time(s)
(s)
Time(s)
Time (s) Time

(d) (e) (f)

FIGURE 13 Translational and rotational seismic acceleration inputs for the experimental cases (a) and (d) NM-1; (b) and (e)
NM-2; and (c) and (f) NM-3.

Input NM-1: translation Input NM-2: translation Input NM-3: translation


x x 0.4 x
4 y y
SaSa(g)(g)
SaSa(g)(g)

SaSa(g)(g)

0.2 z
0.2
2

0 0.0 0.0
10°1 100 101 102 10°1 100 101 102 100 102
Frequency (Hz)
Frequency (Hz) Frequency (Hz)
Frequency (Hz) Frequency (Hz)
Frequency (Hz)

(a) (b) (c)


Input NM-1: rotation Input NM-2: rotation Input NM-3: rotation
100 ry 6 rx 4 rx
Sa (rads/s*s)
(rads/s*s)

Sa (rads/s*s)

ry ry
(rad/s2)

(rad/s2)
2)

4
Sa (rad/s

50 2
2
Sa

Sa
Sa

0 0 0
10°1 100 101 102 10°1 100 101 102 10°1 100 101 102
Frequency(Hz)
Frequency (Hz) Frequency (Hz)
Frequency (Hz) Frequency (Hz)
Frequency (Hz)

(d) (e) (f)

FIGURE 14 Translational and rotational 2% damped response spectra of acceleration inputs for the experimental cases (a) and
(d) NM-1; (b) and (e) NM-2; and (c) and (f) NM-3.
Dhulipala ET AL 19

Acoustic ALE Experiment

Input NM-1: Point PW3 Input NM-1: Point PW2 Input NM-1: Point PW1
10 10 10
(kPa)
(kPa)

Pressure (KPa)

(kPa)
Pressure (KPa)

Pressure (KPa)
5 Pressure 5 5
Pressure

Pressure
0 0

°5 °5 °5

°10 °10 °10


0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s)
Time (s) Time (s)
Time (s) Time (s)
Time (s)

(a) (b) (c)


Input NM-2: Point PW3 Input NM-2: Point PW2 Input NM-2: Point PW1
1 1 1
(kPa)

(kPa)
Pressure (KPa)

(kPa)

Pressure (KPa)
Pressure (KPa)
Pressure

Pressure
Pressure

0 0 0

°1 °1 °1
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s)
Time (s) Time (s)
Time (s) Time (s)
Time (s)

(d) (e) (f)


Input NM-3: Point PW3 Input NM-3: Point PW2 Input NM-3: Point PW1
1
1
1
(kPa)

(kPa)
Pressure (KPa)

(kPa)

Pressure (KPa)
Pressure (KPa)
Pressure

Pressure

0
Pressure

0 0

°1
°1 °1
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Time (s)
Time (s) Time (s) Time (s)
Time (s)
Time (s)

(g) (h) (i)

FIGURE 15 Pressure time series for acoustic FSI, ALE, and experiments at monitoring locations: (a), (d), and (g) PW3; (b),
(e), and (h) PW2; and (c), (f), and (i) PW1.
20 Dhulipala ET AL

Acoustic ALE Experiment

Input NM-1: Point W1 Input NM-2: Point W1 Input NM-3: Point W1


0.12 0.12
height (m)

height (m)
height (m)
(m)

(m)
(m)
0.12
Waveheight

Waveheight
Waveheight
0.00 0.00
0.00
Wave

Wave
Wave
°0.10 °0.10 °0.10
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7
Time(s)
Time (s) Time(s)
Time (s) Time(s)
Time (s)

(a) (b) (c)


Input NM-1: Point W2 Input NM-2: Point W2 Input NM-3: Point W2
0.12 0.12 0.12
height (m)
(m)

height (m)
(m)
height (m)
(m)

Waveheight
Waveheight

Waveheight
0.00 0.00 0.00
Wave
Wave

Wave
°0.10 °0.10
°0.10
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7
Time(s)
Time (s) Time(s)
Time (s) Time(s)
Time (s)

(d) (e) (f)

FIGURE 16 Wave height time series at monitoring locations: (a)–(c) W1 and (d)–(f) W2.

Acoustic ALE Experiment

Input NM-1 Input NM-2 Input NM-3


50 5.0
2.5
(KN-m)
(KN-m)
(KN-m)

2.5
y (kN-m)
y (kN-m)
y (kN-m)

0 0.0
0.0
MMy
MMy

MMy

°2.5 °2.5
°50
0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6
Time (s)
Time (s) Time (s) Time (s)
Time (s)
Time (s)
(a) (b) (c)

FIGURE 17 Moment about the y-axis (𝑀𝑦 ) time series at the tank base.
Dhulipala ET AL 21

TABLE 2 Peak values of the response quantities for the three seismic inputs. For ALE and acoustic FSI, the absolute % error is
also presented in parentheses, wherein, the bolded values represent cases with a 15 % error threshold.

Arbitrary
Response Experiment Lagrangian Acoustic FSI
Eulerian
Case 1: NM-1 seismic input
Pressure (PW3) 4.56 kPa 4.97 kPa (9%) 4.89 kPa (7%)
Pressure (PW2) 8.28 kPa 8.41 kPa (2%) 9.56 kPa (15%)
Pressure (PW1) 8.44 kPa 8.79 kPa (4%) 8.59 kPa (2%)
Wave height (W1) 0.094 m 0.15 m (60%
60%
60%) 0.107 m (14%)
Wave height (W2) 0.103 m 0.094 m (9%) 0.109 m (6%)
Moment (My) 44.69 kN-m 48.49 kN-m (9%) 45.72 kN-m (2%)
Case 2: NM-2 seismic input
Pressure (PW3) 0.74 kPa 0.57 kPa (23%
23%
23%) 0.7 kPa (5%)
Pressure (PW2) 0.84 kPa 0.87 kPa (4%) 0.93 kPa (11%)
Pressure (PW1) 0.85 kPa 0.92 kPa (8%) 0.96 kPa (13%)
Wave height (W1) 0.086 m 0.092 m (7%) 0.089 m (3%)
Wave height (W2) 0.073 m 0.071 m (3%) 0.089 m (22%
22%
22%)
Moment (My) 4.19 kN-m 3.98 kN-m (5%) 3.58 kN-m (15%)
Case 3: NM-3 seismic input
Pressure (PW3) 0.83 kPa 0.81 kPa (2%) 0.81 kPa (2%)
Pressure (PW2) 0.96 kPa 0.92 kPa (4%) 0.78 kPa (19%
19%
19%)
Pressure (PW1) 1.18 kPa 1.23 kPa (4%) 1.09 kPa (8%)
Wave height (W1) 0.098 m 0.096 m (2%) 0.09 m (8%)
Wave height (W2) 0.081 m 0.102 m (26%
26%
26%) 0.095 m (17%
17%
17%)
Moment (My) 4.67 kN-m 4.39 kN-m (6%) 3.12 kN-m (33%
33%
33%)

TABLE 3 Average computational time required by different numerical techniques for the three cases of Table 1.

Arbitrary Incompressible
Acoustic FSI
Lagrangian Computational
(4 processors;
Eulerian Fluid Dynamics
MOOSE/MASTODON)
(LS-DYNA 14 ) (LS-DYNA 14 )
Average
computational 1.36 hours 27 hours 32 192 hours 32
time
22 Dhulipala ET AL

5 APPLICATION: FLUID-STRUCTURE INTERACTION ANALYSIS OF A MOLTEN SALT


REACTOR NUCLEAR POWER PLANT BUILDING

The above sections present the implementation and V&V of the acoustic formulation in MOOSE for performing seismic FSI
analysis. The acoustic formulation provides a computationally efficient approach for performing seismic FSI analysis while
explicitly modeling the fluid, structures in and around the fluid, as well as coupling them. Together with the various other
capabilities of MOOSE, and MOOSE applications such as MASTODON, the acoustic FSI capabilities presented in this paper
can be used to develop detailed FE models of complex facilities like nuclear power plants.
The nuclear power plant analyzed in this section houses a molten chloride fast reactor (MCFR). The seismic analysis is
performed in MASTODON and involves tri-directional earthquake shaking of the building at its foundation. The building mesh
is developed in Cubit 47 and comprises a one-story shear-wall building containing a reactor vessel that is head supported from
a concrete slab. For simplicity, the internal structures of the reactor vessel are ignored and it is assumed to be partially filled
and the reactor vessel is modeled as a cylindrical tank with a spherical bottom. The building also houses four steam generators
supported on stiff, steel-framed bases. Figures 18a and 18b present the full building and building with roof and walls excluded,
respectively. Figure 18c presents its sectional view, where the reactor vessel with a spherical bottom (shown in grey) is supported
from a concrete slab (shown in black) and the steam generators (shown in pink) are supported on stiff steel framed bases (shown
in yellow). A detailed description of the building is provided in Bolisetti et al. 48 and Parsi et al. 49 .

(a) (b)

(c)

FIGURE 18 Reactor building: (a) full view; (b) view with roof and walls excluded; and (c) sectional view of the building. The
reactor vessel (shown in grey) is attached to the slab and has a rounded base.

Both the building and reactor vessel are assumed to be linearly elastic. The building is constructed from reinforced concrete,
and the reactor vessel is fabricated from steel. The assumed density, Young’s modulus and Poisson’s ratio used for concrete
(steel) are 2400 kg/m3 (7850 kg/m3 ), 24.8 GPa (170.0 GPa), and 0.2 (0.3), respectively. The building has plan dimensions of
90 × 60 m. The reactor vessel is 20 mm thick and its dimensions are shown in Figure 19. Since the demonstration here focuses
on the reactor vessel and seismic FSI, the steam generators are idealized as solid cylinders. Bolisetti et al. 48 and Parsi et al. 49
provides further details on the building and its components. Rayleigh damping of 2% of critical is assumed for both the building
and the reactor vessel. Additionally, for simplicity, the molten salt in the reactor vessel is assumed to have the same properties
Dhulipala ET AL 23

as water: density (𝜌𝑜 ) of 1000 kg/m3 and a speed of sound (𝐶𝑜 ) of 1500 m/s. Although the properties of molten salt are likely
to be very different from those of water, this assumption is deemed reasonable for this demonstration. This section presents the
pressure and wave height responses of the fluid in the reactor vessel as well as the acceleration responses at the top of the building.
Figure 19 identifies monitoring locations for pressures and wave heights. The base of the building is subjected to translational
acceleration histories from the 1999 Chi Chi earthquake recorded at station TCU052. Figure 20 presents the acceleration time
series and response spectra of the three components of this ground motion.

R = 2.5 m

W3 W4
W1W2 Z W5W6
PW3: (-2.5, 0, 5.0) W1: (-2.0, 0, 5.0)
PW3 PE3
PW2: (-2.5, 0, 2.455) W2: (-1.25, 0, 5.0)
PW1: (-2.5, 0, 1.2) W3: (-0.5, 0, 5.0)
6m

PW2 PE2
5m

PE3: (2.5, 0, 5.0) W4: (0.5, 0, 5.0)


PE2: (2.5, 0, 2.455) W5: (1.25, 0, 5.0)
PW1 PE1 PE1: (2.5, 0, 1.2) W6: (2.0, 0, 5.0)

West East

FIGURE 19 The dimensions of the reactor vessel and monitoring locations.

0.2
0.4
x
(rads/s*s)

0.1 y
(g)
Acc. (g)

(g)

z
Acc.

0.2
Sa Sa

0.0

°0.1 0.0
0 5 10 15 20 25 10°1 100 101 102
Time
Time(s)
(s) Frequency (Hz)
Frequency (Hz)
(a) (b)

FIGURE 20 Three components of ground motion (a) time series and (b) response spectra from the Chi Chi earthquake recorded
at station TCU052.

Figures 21a and 21b present the hydrodynamic pressure time series at the three monitoring interest on the west and east ends
of the vessel, respectively. Figures 22a and 22b present the peak pressure profiles over the depth of the tank at 7.5, 20, and
24.66 s on the west and east sides of the tank, respectively. Figure 22c presents the wave heights along the east-west diameter
of the surface of the fluid at the same three time instances. Note that at 7.5 s, the wave height profile does not have a near-zero
amplitude at zero radial distance, but it is almost zero for other time instances. This is because, at 7.5 s, there is a peak in the
24 Dhulipala ET AL

vertical acceleration, as seen in Figure 20, which causes vertically propagating pressure waves in the fluid due to the presence of
only one convective mode. This results in non-zero wave heights all across the fluid surface, unlike the horizontal accelerations,
which result in roughly equal and opposite wave heights on either sides of the center.

PW3 PW2 PW1 PE3 PE2 PE1

1.0 1.0

Pressure (KPa)
Pressure (KPa)

(kPa)
(kPa)

0.5 0.5

0.0 0.0

Pressure
Pressure

°0.5 °0.5

°1.0 °1.0
0 10 20 0 10 20
Time (s)
Time (s) Time (s)
Time (s)
(a) (b)
W1 W2 W3 W4 W5 W6
(m)

(m)
height (m)

height (m)
0.05 0.05
Wave height

Wave height
0.00 0.00
Wave

Wave
°0.05 °0.05

0 10 20 0 10 20
Time (s)
Time (s) Time (s)
Time (s)
(c) (d)

FIGURE 21 Pressure time series at (a) west and (b) east faces of the reactor vessel; (c) and (d) wave height time series on the
fluid surface.
Dhulipala ET AL 25

5 5
7.5
7.5ss
20
20 ss
4 4
24.6
24.66ss

Depth (m)
(m)

Depth (m)
(m)
Depth

Depth
3 3

2 2

1 1
°1.0 °0.5 0.0 0.5 1.0 °1.0 °0.5 0.0 0.5 1.0
Pressure (KPa)
Pressure (kPa) Pressure(kPa)
(KPa)
Pressure
(a) (b)

0.05
(m)
height(m)
Wave height

0.00
Wave

°0.05

°2 0 2
Radial distance (m)
Radial distance (m)
(c)

FIGURE 22 Pressure profiles as a function of depth on the (a) west and (b) east faces of the vessel; (c) wave height profiles as
a function of radial distance.

6 SUMMARY AND CONCLUSIONS

FSI analysis is an essential component of the seismic design and risk assessment of various critical infrastructure facilities such as
advanced reactor nuclear power plants. Acoustic FSI offers a computationally efficient means of performing seismic FSI analysis
of fluid-filled tanks while also accounting for complex, multi-directional seismic inputs and complex geometries. However,
limited numerical verification with more sophisticated formulations such as ALE and validation studies using experiments exist
for acoustic FSI that consider multi-directional seismic inputs and multiple response quantities such as pressures, wave heights,
and support reactions. The main outcomes of this study are:

• An acoustic FSI formulation was implemented in MOOSE, an open source, highly-parallel, multiphysics, finite ele-
ment analysis framework, which is widely used for computational modeling and simulation, especially in the nuclear
engineering community.

• The acoustic FSI formulation was verified against benchmark problems, including analytical solutions of FSI in a head
supported cylindrical tank partially filled with water. The pressure and wave height time series predicted by acoustic FSI
satisfactorily matched with the analytical solutions.

• The acoustic FSI implementation was validated using experimental data from tests of a base-supported cylindrical tank
subjected to multi-directional seismic inputs with intensities ranging between 0.1 g - 1g. Pressures and base moments
matched satisfactorily. Wave heights under complex wave responses were not predicted accurately.

• Acoustic FSI performed similarly to that of an ALE formulation, with the exception of wave height.

• If pressures and support reactions are of interest, acoustic FSI is viable numerical approach.

• Acoustic FSI may only provide a first order approximation for wave heights and is best suited for low to moderate intensity
seismic inputs.
26 Dhulipala ET AL

• Acoustic FSI was at least one to two orders of magnitude faster than both ALE and ICFD numerical methods.

• The acoustic FSI implementation in MOOSE was used to demonstrate the seismic simulation of a detailed, full-scale
nuclear power plant building that housed a hypothetical advanced nuclear reactor.

ACKNOWLEDGEMENTS

The work presented herein was funded in part by the Advanced Research Projects Agency-Energy (ARPA-E), U.S. Department
of Energy, under Award Number DE-AR0000978. This research is also supported through the INL Laboratory Directed Research
and Development (LDRD) Program under DOE Idaho Operations Office Contract DE-AC07-05ID14517. This research used of
the resources of the High Performance Computing Center at INL, which is supported by the Office of Nuclear Energy of the
U.S. DOE and the Nuclear Science User Facilities under Contract No. DE-AC07-05ID14517. The views and opinions of the
authors expressed herein do not necessarily state or reflect those of the U.S. Government or any agency thereof. Two anonymous
reviewers are thanked for their valuable comments, which improved the quality of this paper.

How to cite this article: Dhulipala S.L.N., C. Bolisetti, L.B. Munday, W.M. Hoffman, C.C. Yu, F.U.H Mir, F. Kong, A.D.
Lindsay, and A.S. Whittaker (2021), Development, Verification, and Validation of Comprehensive Acoustic Fluid-Structure
Interaction Capabilities in an Open-Source Computational Platform, Earthquake Engineering & Structural Dynamics, –.

APPENDIX

A MOOSE INPUT FILE SYNTAX FOR ACOUSTIC FLUID-STRUCTURE INTERACTION


MODELING

An overview of setting up the MOOSE input file for running acoustic FSI simulations is presented in this appendix. Before
discussing acoustic FSI objects in MOOSE, there is a preliminary step. A finite element mesh for the acoustic FSI problem
is generated and read by setting up the MOOSE meshing system in the input file. This step is standard for all MOOSE input
files and https://mooseframework.inl.gov/syntax/Mesh/index.html, Permann et al. 30 , and Veeraraghavan et al. 31 provide more
information about generating and reading meshes.
The primary and auxiliary (or derived from primary) variables of the FSI problem are set up next. For the fluid domain, the
primary variable is pressure (p) and the auxiliary variable is wave height (Wave). For the structural domain, the primary variables
are displacements (disp_x, disp_y, disp_z) and the auxiliary variable are velocities and accelerations. Listing 1 presents
the Variables and AuxVariables blocks of the input file. Notice that these varibles are domain restricted by using the block
option.

Listing 1: Variables and AuxVariables blocks for the acoustic FSI input file.
[ Variables ]
[p]
block = ‘ Fluid ’
[]
[ disp_x ]
block = ‘ Tank ’
[]
[ disp_y ]
block = ‘ Tank ’
[]
[ disp_z ]
block = ‘ Tank ’
[]
[]

[ AuxVariables ]
[ Wave ]
block = ‘ Fluid ’
[]
Dhulipala ET AL 27

[ vel_x ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[ accel_x ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[ vel_y ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[ accel_y ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[ vel_z ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[ accel_z ]
order = FIRST
family = LAGRANGE
block = ‘ Tank ’
[]
[]

The kernels (or the components of the governing equations) for fluid and structural domains are specified. As discussed
in Figure 2, the fluid domain governing equation has diffusion and double time derivative of pressure components that are
simulated using the Diffusion and AcousticInertia kernels, respectively. The structural domain governing equations has
elastic stiffness and inertial components that are simulated using the DynamicTensorMechanics and InertialForce kernels,
respectively. All three inertial components are considered in the structural domain. Listing 2 presents the Kernels block of the
input file.

Listing 2: Kernels block for the acoustic FSI input file.


[ Kernels ]
[ diffusion ]
type = Diffusion
variable = ‘p ’
block = ‘ Fluid ’
[]
[ inertia ]
type = AcousticI n er ti a
variable = ‘p ’
block = ‘ Fluid ’
[]
[ DynamicTensorMechanics ]
displacements = ‘ disp_x disp_y disp_z ’
block = ‘ Tank ’
[]
[ inertia_x ]
type = InertialForce
variable = ‘ disp_x ’
block = ‘ Tank ’
[]
[ inertia_y ]
type = InertialForce
variable = ‘ disp_y ’
block = ‘ Tank ’
[]
[ inertia_z ]
type = InertialForce
variable = ‘ disp_z ’
block = ‘ Tank ’
[]
[]
28 Dhulipala ET AL

Material properties corresponding to the fluid and structural domains are specified. The material property for the
fluid domain is the speed of sound (𝑐𝑜 ) specified using the inverse of squared 𝑐𝑜 inv_co_sq. The material prop-
erties for the linear elastic structural domain are the Young’s modulus and Poisson’s ratio, and are specified within
ComputeIsotropicElasticityTensor. In addition, for the structural domain, linear elasticity is specified using the
ComputeSmallStrain and ComputeLinearElasticStress objects. Listing 3 presents the Materials block of the input file.

Listing 3: Materials block for the acoustic FSI input file.


[ Materials ]
[ co_fluid ]
type = G e n e r i c C o n s t a n t M a t e r i a l
prop_names = inv_co_sq
prop_values = 4.65 e -7
block = ‘ Fluid ’
[]
[ density_tank ]
type = G e n e r i c C o n s t a n t M a t e r i a l
block = ‘ Tank ’
prop_names = density
prop_values = 7.85 e -6
[]
[ elasticity_tank ]
type = C o m p u t e I s o t r o p i c E l a s t i c i t y T e n s o r
youngs_modulus = 200 # ( in Giga Pascals )
poissons_ratio = 0.27
block = ‘ Tank ’
[]
[ strain_tank ]
type = Co m pu t eS m a l l S t r a i n
block = ‘ Tank ’
displacements = ‘ disp_x disp_y disp_z ’
[]
[ stress_tank ]
type = C o m p u t e L i n e a r E l a s t i c S t r e s s
block = ‘ Tank ’
[]
[]

Boundary conditions (BCs) for the FSI model are specified. The fluid domain has a FluidFreeSurfaceBC to simulate the
convective modes and resulting gravity waves. The structural domain has input acceleration specified in the required direction
using PresetAcceleration that reads the acceleration values from the input function ‘accel_input’. If the displacements
corresponding to other directions are fixed, these are specified using the DirichletBC object. Listing 4 presents the BCs block
of the input file.

Listing 4: BCs block for the acoustic FSI input file.


[ BCs ]
[ free ]
type = Fl u id F re e S u r f a c e B C
variable = ‘p ’
boundary = ‘ Fluid_top ’
alpha = ‘0.1 ’ # ( inverse of acceleration due to gravity )
[]
[ accel_input ]
type = Pr e se t Ac c e l e r a t i o n
variable = ‘ disp_x ’
velocity = ‘ vel_x ’
acceleration = ‘ accel_x ’
beta = 0.25
function = ‘ accel_input ’
boundary = ‘Top ’
[]
[ disp_x2 ]
type = DirichletBC
variable = ‘ disp_y ’
boundary = ‘Top ’
value = 0.0
[]
[ disp_x3 ]
type = DirichletBC
variable = ‘ disp_z ’
Dhulipala ET AL 29

boundary = ‘Top ’
value = 0.0
[]
[]

Fluid and structure coupling is accomplished using InterfaceKernels, and, in particular, the
StructureAcousticInterface for acoustic FSI. Three interface kernels are specified for coupling the acoustic fluid behavior
to the three components of structural displacement. Listing 5 presents the InterfaceKernels block of the input file.

Listing 5: InterfaceKernels block for the acoustic FSI input file.


[ I n t erfaceKernels ]
[ interface1 ]
type = S t r u c t u r e A c o u s t i c I n t e r f a c e
variable = ‘p ’
neighbor_var = ‘ disp_x ’
boundary = ‘ Interface ’
D = 1e -6 # ( fluid density in Giga kg per m ^3)
component = 0
[]
[ interface2 ]
type = S t r u c t u r e A c o u s t i c I n t e r f a c e
variable = ‘p ’
neighbor_var = ‘ disp_y ’
boundary = ‘ Interface ’
D = 1e -6 # ( fluid density in Giga kg per m ^3)
component = 1
[]
[ interface3 ]
type = S t r u c t u r e A c o u s t i c I n t e r f a c e
variable = ‘p ’
neighbor_var = ‘ disp_z ’
boundary = ‘ Interface ’
D = 1e -6 # ( fluid density in Giga kg per m ^3)
component = 2
[]
[]

An Executioner block is specified, wherein, options such as the solve_type, PETSc settings, time step, TimeIntegrator,
and automatic_scaling are specified. Listing 6 presents the Executioner block of the input file.

Listing 6: Executioner block for the acoustic FSI input file.


[ Executioner ]
type = Transient
solve_type = ’ NEWTON ’
p e t s c _ o p t io n s _ i n a m e = ’- pc_type - pc_f actor_ mat_so lver_p ackage ’
p e t s c _ o p t io n s _ v a l u e = ’ lu superlu_dist ’
start_time = 0.0
end_time = 4.0
dt = 0.002
dtmin = 0.00001
nl_abs_tol = 1e -14
nl_rel_tol = 1e -14
l_tol = 1e -14
l_max_its = 25
t i m e st e p_ t ol e ra n c e = 1e -8
a u tom ati c_ sca lin g = true
[ TimeIntegrator ]
type = NewmarkBeta
[]
[]

Finally, the required outputs such as pressures and wave heights at specific points in the model are captured using
Postprocessors and the data is written into csv files. Listing 7 presents the Postprocessors block of the input file.

Listing 7: Postprocessors block for the acoustic FSI input file.


[ Po stprocessors ]
[ Wave_1 ]
type = PointValue
point = ‘ -0.79 0.0 1.805 ’
30 Dhulipala ET AL

variable = Wave
[]
[ Pressure_1 ]
type = PointValue
point = ‘ -0.79 0.0 1.805 ’
variable = p
[]
[]

References

1. Sandberg G, Ohayon R. Computational aspects of structural acoustics and vibration. Springer Science and Business Media
. 2009.

2. Zhao C, Chen J, Wang J, Yu N, Xu Q. Seismic mitigation performance and optimization design of NPP water
tank with internal ring baffles under earthquake loads. Nuclear Engineering and Design 2017; 318: 182–201. doi:
https://doi.org/10.1016/j.nucengdes.2017.04.023

3. Zhao C, Yu N. Dynamic response of generation III+ integral nuclear island structure considering fluid structure interaction
effects. Annals of Nuclear Energy 2018; 112: 189–207. doi: https://doi.org/10.1016/j.anucene.2017.10.011

4. American Society of Civil Engineers . Fluid-structure interaction during seismic excitation. Reston, VA, USA: Committee
on Seismic Analysis of the Committee on Nuclear Structures and Materials of the Structural Division of ASCE . 1984.

5. Mir FUH, Yu CC, Whittaker AS. Experimental and numerical studies of seismic fluid-structure interaction in a
base-supported cylindrical vessel. Earthquake Engineering and Structural Dynamics 2021; 50(5): 1395–1413. doi:
https://doi.org/10.1002/eqe.3402

6. Oberkampf WL, Trucano TG, Hirsch C. Verification, validation, and predictive capability in computational engineering
and physics (No. SAND2003-3769). Albuquerque, NM, USA: Sandia National Laboratories . 2003.

7. Oberkampf WL, Trucano TG. Verification and validation in computational fluid dynamics. Progress in Aerospace Sciences
2002; 38(3): 209–272. doi: https://doi.org/10.1016/S0376-0421(02)00005-2

8. Yu CC, Whittaker AS. Analytical solutions for seismic fluid-structure interaction of head-supported cylindrical tanks.
Journal of Engineering Mechanics 2020; 146(10): 04020112. doi: https://doi.org/10.1061/(ASCE)EM.1943-7889.0001831

9. Jacobsen LS. Impulsive hydrodynamics of fluid inside a cylindrical tank and of fluid surrounding a cylindrical pier. Bulletin
of the Seismological Society of America 1949; 39(3): 189–204. doi: https://doi.org/10.1785/BSSA0390030189

10. Housner GW. Dynamic pressures on accelerated fluid containers. Bulletin of the Seismological Society of America 1957;
47(1): 15–35. doi: https://doi.org/10.1785/BSSA0470010015

11. Veletsos A, Yang J. Earthquake response of liquid storage tanks. Raleigh, North Carolina, USA: Proceedings of Second
Annual Engineering Mechanics Division Specialty Conference . 1977.

12. Chalhoub MS, Kelly JM. Theoretical and experimental studies of cylindrical water tanks in base isolated structures.
Berkeley, CA, USA: UCB/EERC-88/07. Earthquake Engineering Research Center, University of California at Berkeley .
1988.

13. Yu CC, Whittaker AS. Review of analytical studies on seismic fluid-structure interaction of base-supported cylindrical
tanks. Engineering Structures 2021; 233: 111589. doi: https://doi.org/10.1016/j.engstruct.2020.111589

14. LSTC . LS-DYNA keyword user’s manual-R11. Livermore, CA, USA: Livermore Software Technology Corporation . 2018.

15. Systèmes D. ABAQUS 2018-unified FEA products . 2018.

16. Pal NC, Bhattacharyya SK, Sinha PK. Experimental investigation of slosh dynamics of liquid-filled containers. Experimen-
tal Mechanics 2001; 41(1): 63–69. doi: https://doi.org/10.1007/BF02323106
Dhulipala ET AL 31

17. Goudarzi MA, Sabbagh-Yazdi SR. Investigation of nonlinear sloshing effects in seismically excited tanks. Soil Dynamics
and Earthquake Engineering 2012; 43: 355–365. doi: https://doi.org/10.1016/j.soildyn.2012.08.001

18. Park JH, Bae D, Oh CK. Experimental study on the dynamic behavior of a cylindrical liquid storage tank subjected to seismic
excitation. International Journal of Steel Structures 2016; 16(3): 935–945. doi: https://doi.org/10.1007/s13296-016-0172-y

19. Compagnoni ME, Curadelli O. Experimental and numerical study of the response of cylindrical steel tanks under seismic
excitation. International Journal of Civil Engineering 2018; 16(7): 793–805. doi: https://doi.org/10.1007/s40999-017-
0218-3

20. Everstine GC. Finite element formulations of structural acoustics problems. Computers and Structures 1997; 65(3): 307–
321. doi: https://doi.org/10.1016/S0045-7949(96)00252-0

21. Virella JC, Prato CA, Godoy LA. Linear and nonlinear 2D finite element analysis of sloshing modes and pressures in
rectangular tanks subject to horizontal harmonic motions. Journal of Sound and Vibration 2008; 312(3): 442–460. doi:
https://doi.org/10.1016/j.jsv.2007.07.088

22. Rodriguez CG, Flores P, Pierart FG, Contzen LR, Egusquiza E. Capability of structural–acoustical FSI numerical model
to predict natural frequencies of submerged structures with nearby rigid surfaces. Computers & Fluids 2012; 64: 117–126.
doi: https://doi.org/10.1016/j.compfluid.2012.05.011

23. Rawat A, Matsagar VA, Nagpal AK. Numerical study of base-isolated cylindrical liquid storage tanks using
coupled acoustic-structural approach. Soil Dynamics and Earthquake Engineering 2019; 119: 196–219. doi:
https://doi.org/10.1016/j.soildyn.2019.01.005

24. Bunting G, Crane NK, Day DM, Dohrmann CR, Ferri BA, Flicek RC, Hardesty S, Lindsay P, Miller ST, Munday LB,
Stevens BL, Walsh T. Sierra/SD-Theory Manual-4.54. Albuquerque, NM, USA: Sandia National Laboratories . 2019

25. Bunting G, Miller ST. Partitioned coupling for structural acoustics. Journal of Vibration and Acoustics 2020; 142(1):
011012. doi: https://doi.org/10.1115/1.4045215

26. Rawat A, Matsagar V, Nagpal AK. Seismic analysis of steel cylindrical liquid storage tank using coupled acoustic-structural
finite element method For fluid-structure interaction. International Journal of Acoustics & Vibration 2020; 25(1). doi:
https://doi.org/10.20855/ijav.2020.25.11499

27. Rawat A, Mittal V, Chakraborty T, Matsagar V. Earthquake induced sloshing and hydrodynamic pressures in rigid liquid
storage tanks analyzed by coupled acoustic-structural and Euler-Lagrange methods. Thin-Walled Structures 2019; 134:
333–346. doi: https://doi.org/10.1016/j.tws.2018.10.016

28. Chen YH, Hwang WS, Ko CH. Sloshing behaviours of rectangular and cylindrical liquid tanks subjected to har-
monic and seismic excitations. Earthquake Engineering & Structural Dynamics 2007; 36(12): 1701–1717. doi:
https://doi.org/0.1002/eqe.713

29. Phan HN, Paolacci F. Fluid-structure interaction problems: an application to anchored and unanchored steel storage tanks
subjected to seismic loadings. Thessaloniki, Greece: 16th European Conference on Earthquake Engineering . 2018.

30. Permann CJ, Gaston DR, Andrš D, Carlsen RW, Kong F, Lindsay AD, Miller JM, Peterson JW, Slaughter AE, Stogner
RH, Martineau RC. MOOSE: Enabling massively parallel multiphysics simulation. SoftwareX 2020; 11: 100430. doi:
https://doi.org/10.1016/j.softx.2020.100430

31. Veeraraghavan S, Bolisetti C, Slaughter AE, Coleman JL, Dhulipala SLN, Hoffman WM, Kim K, Kurt E, Spears R, Munday
LB. MASTODON: An Open-Source Software for Seismic Analysis and Risk Assessment of Critical Infrastructure. Nuclear
Technology 2020: 1–23. doi: https://doi.org/10.1080/00295450.2020.1807282

32. Yu CC, Mir FUH, Whittaker AS. Validation of numerical models for seismic fluid-structure-interaction
analysis of nuclear, safety-related equipment. Nuclear Engineering and Design 2021; 379: 111179. doi:
https://doi.org/10.1016/j.nucengdes.2021.111179
32 Dhulipala ET AL

33. Rienstra SW, Hirschberg A. An Introduction to Acoustics. Eindhoven University of Technology . 2004.

34. Kohnke PC. ANSYS Theory Reference: Release 5.6. ANSYS, Incorporated . 1999.

35. Koutromanos I. Fundamentals of finite element analysis: linear finite element analysis. Blacksburg, VA USA: John Wiley
and Sons . 2018.

36. Wang X, Bathe KJ. Displacement/pressure based mixed finite element formulations for acoustic fluid–structure inter-
action problems. International Journal for Numerical Methods in Engineering 1997; 40(11): 2001–2017. doi:
https://doi.org/10.1002/(SICI)1097-0207(19970615)40:11%3C2001::AID-NME152%3E3.0.CO;2-W

37. Bathe KJ, Nitikitpaiboon C, Wang X. A mixed displacement-based finite element formulation for acoustic fluid-structure
interaction. Computers and Structures 1995; 56(2): 225–237. doi: https://doi.org/10.1016/0045-7949(95)00017-B

38. Kirk BS, Peterson JW, Stogner RH, Carey GF. libMesh: a C++ library for parallel adaptive mesh refinement/coarsening
simulations. Engineering with Computers 2006; 23(3–4): 237–254. doi: https://doi.org/10.1007/s00366-006-0049-3

39. Balay S, Abhyankar S, Adams MF, Benson S, Brown J, Brune P, Buschelman K, Constantinescu EM, Dalcin L, Dener V,
Gropp WD, Hapla V, Isaac T, Jolivet P, Karpeev D, Kaushik D, Knepley MG, Kong F, Kruger S, May DA, Curfman McInnes
L, Tran Mills R, Mitchell L, Munson T, Roman JE, Rupp K, Sanan P, Sarich J, Smith BF, Zampini S, Zhang H, Zhang H,
Zhang J. PETSc/TAO Users Manual. Tech. Rep. ANL-21/39 - Revision 3.16, Argonne National Laboratory 2021.

40. Jiang W, Hales JD, Spencer BW, Collin BP, Slaughter AE, Novascone SR, Toptan A, Gamble KA, Gardner R. TRISO
particle fuel performance and failure analysis with BISON. Journal of Nuclear Materials 2021; 548: 152795. doi:
https://doi.org/10.1016/j.jnucmat.2021.152795

41. Spencer BW, Hoffman WM, Biswas S, Jiang W, Giorla A, Backman MA. Grizzly and BlackBear: Structural component
aging simulation codes. Nuclear Technology 2021: 1–23. doi: https://doi.org/10.1080/00295450.2020.1868278

42. Xia Y, Plummer M, Mattson E, Podgorney R, Ghassemi A. Design, modeling, and evaluation of a dou-
blet heat extraction model in enhanced geothermal systems. Renewable energy 2017; 105: 232–247. doi:
https://doi.org/10.1016/j.renene.2016.12.064

43. Doulgerakis N, Tehrani PK, Talebinejad I, Kosbab B, Cohen M, Whittaker AS. Software Commercial Grade Dedication
Guidance for Nonlinear Seismic Analysis. Report prepared by SC Solutions for the Department of Energy under grant
number DE-NE0008857 . 2021.

44. Mei CC. Wave propagation. Boston, MA, USA: Massachusetts Institute of Technology: MIT OpenCouseWare,
https://ocw.mit.edu/, License: Creative Commons BY-NC-SA . 2000.

45. Bolisetti C, Whittaker AS. Site Response, Soil-Structure Interaction and Structure-Soil-Structure Interaction for Per-
formance Assessment of Buildings and Nuclear Structures. Buffalo, NY, USA: Technical Report MCEER-15-0002,
Multidisciplinary Center for Earthquake Engineering Research . 2015

46. Mir FUH, Yu CC, Whittaker AS. Rocking response of liquid-filled cylindrical tanks. Earthquake Spectra 2021; 37(3):
1698–1709. doi: https://doi.org/10.1177/8755293020981973

47. Blacker TD, Bohnhoff WJ, Edwards TL. CUBIT mesh generation environment. Volume 1: Users manual (No. SAND-94-
1100). Albuquerque, NM, USA: Sandia National Laboratories . 1994.

48. Bolisetti C, Coleman JL, Hoffman WM, Whittaker AS, Parsi SS, Redd J, Cohen M, Kramer K, Kirchman P, Bowers H, Lal
K. Seismic isolation of major advanced reactor systems for economic improvement and safety assurance. Idaho Falls, ID,
USA: Idaho National Laboratory (INL/EXT-20-59608-Rev000) . 2020

49. Parsi SS, Lal KM, Kosbab BD, Ingersoll ED, Shirvan K, Whittaker AS. Seismic isolation: A pathway
to standardized advanced nuclear reactors. Nuclear Engineering and Design 2022; 387: 111445. doi:
https://doi.org/10.1016/j.nucengdes.2021.111445

You might also like