Contents lists available at ScienceDirect Engineering Structures journal homepage: www.elsevier.com/locate/engstruct An efficient computational procedure for the dynamic analysis of liquid storage tanks RO. Ruiz*®, D. Lopez-Garcia*”, A.A. Taflanidis” *Deponment of Sacral Getechnic! Engineering Poti Universiad Case de hi, end "Deparment of il &envronmentlEnieeing& Earth cence, nies of Neve Da, 156 parc Hal. Note Dae, IV ASS, USA oa RerearchCeter for Integrate Natural iat Management ‘aera Accepted # Decenibet 2014 ‘valable enn 2 anay 2018 {X compatationally efficient numerical model is developed im this study for evaluating the dynamic behavior of liquid storage tanks. This model has higher complexity than the Housner model (which eot- respond tothe simplest and most popular approach for eppreximating the behavier of rectangular and CGrelar tanks) but stll enjoys high computational simplicity to facilitate implementation in practice, ‘while ts applicable to virtually any Kin of tank geometry, providing atthe same time a high degree fof accuracy, n the proposed model, the igi is assumed to be inviscid incompressible and rotation, and its motion is completely characterized by a velocity potential function. Thus, the Continuity and Equilibriam eqsations characterizing this motion take the form of Laplace and Bernoul equations, respectively The Laplace equation is solved though 420 finite element scheme and is then combined withthe Besnoull equation through the velocity potential funetion condensed atthe fre sutace of the liquid. Numerical details for the practical implementation af the proposed scheme ate discussed, Whereas the approximation is shown fo provide results with high accuracy for the dynamic behavior of diferent ype of tanks when compared tthe Housner model and ful rte element implementation. ‘As showa inthe examples considered the computational efficiency of the proposed mode s such that extensive parametric studies can be performed with smal numetial effort. which in turn makes the Proposed model very attractive not only for analysis purposes but alo forthe design of liquid storage Rowena Dynamic analysis of liquids fanke and other related cevices such ae tuned igu dampers. © 2014 Elsevier Li [A eights reserve, 1. Introduction The study ofthe dynamic behavior on liquid storage tanks has ‘gained significant attention in the last years as the seismic vullner- ability of these tanks represents a potential source of significant ‘economic loss due to structural failures, leakages ar environmental accidents (caused by the liquid spilled out) [1~2], whereas such tanks have been proposed to be used as mass dampers to mitigate the vibration induced by wind or seismic excitations [4). Several approaches have been proposed in this setting to model the ‘dynamical behavior of such liquid storage tanks. In earlier studies, the fuid was taken into account by adding a mass tothe structure, with characteristics computed by an analytical solution based on simplified geometries [5]. Later, Housner developed an analysis and design procedure, primarily for cylindrical and rectangular mal direst: roriz2ucc (RO. Ruz). dieing pice (D. Lepe-Garia storage tanks, based on a simple mechanical model (combination of mass-spring systems with different characteristics) that repre- sents the fluid. The computation of the physical constants in this procedure is based on the separation of the hydrodynamic ‘behavior into two components: (1) the impulsive component that is related to the mass that moves together with the structure: and (2) the convective component that takes into account the free Surface oscillations (6.7) This is a broadly adopted model in civil engineering since i€ provides closed form solutions for the transmitted force due to the liquid sloshing, and represents the basis of many design codes, e. API 650 [8], AWWA D100 [9] and the New Zealand recommendation guidelines NZSEE [10], that establish procedures for the seismic response analysis of liquid tanks based on this linear model proposed by Housner. At the same time, itis an approximation that is based on the asstimption that simplified flows can represent the actual fluid movement, restrict- ing its use to tanks with simple geometries (such as rectangular or circular tanks),RO Rue a Ensncering Stractres 65 (2015) 206-218 207 To obtain higher accuracy solutions, various high-fidelity proce- ddures [11-12] have been also established, applicable to arbitrary tank geometries, using the Finite Element Method (FEM) to model the fuid utilizing (depending on the numerical scheme) displace- iment, pressure or potential variables to characterize the fluid motion, For generalized thuid-structure interaction applications, a5 the equations of the structure are expressed in terms of dis placements it is convenient to also express the fluid equations through displacement variables. With respect tothe fluid motion ‘modeling, such a FEM approach, based om a displacement formula- tion, leads toa symmetric eigenvalue problem but it produces non zero spurious frequencies that are dificult to identify [12.14] and suppress [13.15] Additionally this approach requires discretizing a vector field (displacement) instead ofa scalar eld (pressure or Potential variables) increasing the number of degrees of freedom. In contrast, a FEM approach based on pressure or potential v ables involves fewer unknowns, increasing the computational elf- ciency and avoiding physical inconsistencies (16). In this case, the fuid-structure system leads to a non-symmetric eigenvalue prob- lem, though it is possible to keep the symmetry if the fluid is described in a redundant way using both pressure and potential variables (17-19). In particular, Olson and Bathe presented such a linear formulation based on velocity potentials and pressures [17], expanded later to take also into account gravity loads (20) ‘An important aspect of this formulation is its suitability for both time-history and frequency analysis of fluids with free surface. “More recently, an increased number of even more complex proce- dures have been proposed, for example taking into account non- linear sloshing due to large fee surface motions [21-24 and including identification of damping effects introduced at the tank walls due to viscosity effects in the thin interface layer [25]. How- ever, implementation of these procedures is almost exclusively rel- gated to scientific and research professional environments due to the complexity ofthe formulations and the high level of expertise required for their implementation. Despite such high-fidelity modeling developments and advances in computer and computational science, the philosophy ofthe analysis methods of design codes is still based on analytical expressions and equivalent mechanical models. Though undoubt- edly some practitioners are utilizing commercial software to solve ‘multi-physic problems under seismic loads, avoiding the use of simplified models proposed by the design codes or even the com- plex implementation of the procedures described above, such approaches are still not widely used (presumably because asignif- icant background is required not only in the software know-how but also in the theoretical knowledge about the involved physics) Furthermore, many traditional software packages used for seismic and structural analysis lack luid-structure interaction modules, enforcing engineers to work with alternative packages that were not designed to perform seismic analysis. There i a gap for a meth- dology that is more simple and attractive than the commercial packages but still maintains the accuracy ofthe advanced methods presented in the literature Motivated by this realization, the main novel contribution of this work is to develop a simplified, computationally efficient framework, utilizing a FEM modeling based on potential variables and a static condensation approach while assuming the tank walls, as rigid, for describing the dynamic behavior of arbitrary geometry liquid scorage tanks under seismic excitation. This approach can facilitate a computationally efficient description of the dynamic behavior of tanks (supporting frequency and time domain analysis, as well a eigenvalue analysis) including its interaction with a sup- porting structure (as needed for TLD design applications), though it ‘annat provide detailed predictions for localized failure phenom- ena seated to the tank walls (which are considered rigid). The pro- posed numerical procedure expresses the linear sloshing problem as a second order linear system of equations, where the indepen ddent variables are the vertical elevation of the free surface and the excitation is directly related to the ground acceleration. The fluid is assumed ideal while the tank walls and bottom are assumed rigid. As the fluid is considered ideal, itis possible to adopt a FEM formulation based on potential variables, reducing the number of unknowns and avoiding problems with spurious fre- quencies. The rigid tank assumption simplifies the fuid-structure coupling since itis not necessary to generate a mesh forthe tank ‘walls and bottom and to match it withthe fluid mesh, In this sense, the procedure is easier to implement than the ones cited previ- ously [12-20], Furthermore, the proposed procedure allows study- ing the sloshing effect over the tank support rather than the sloshing effect over the tank itself 12-20}. Although the method- ology is standard, the numerical procedure offers significant advances and physical insight as (1) the equations are expressed in terms of physical variables (Iree surface elevation and ground acceleration), (2) the system of equations is similar to that of ‘mass-spring systems, (3) the approach is valid for any tank geom- etry, (4) the formulation is suitable for both time-history and fre quency analysis, (5) it allows for a straightforward. coupling between rig tanks and elastic structures, (6) the procedure is rel= ative easy to implement or understand, and (7) it offers significant advances over the models suggested by the design codes. Ulti- imately, the proposed numerical procedure. ftom now on denoted as Simplified Sloshing Model (SSM), enjoys such computational simplicity and effcieney that it can be used for various tasks such as parametric studies, preliminary dimensioning of tanks, seismic performance identification, or even design and dimensioning of tuned liquid dampers. 2. Description of the Simplified Sloshing Model (SSM) A2D schematic diagram of aliquid storage tank is presented in Fig, I. It is important to mention that, although the formulation of the proposed Simplified Sloshing Model could be either 20 or 3D, this paper presents in detail only a 2D formulation because it is the case most extensively studied in the literature, and the valida- tion will be made considering several 2D existing examples ana- lyzed independently by other authors. An inertial system of feference x-2 is located at the middle of the non-pertutbed free surface and an auxiliary coordinate 7 is defined to measure the rel- ative displacement between the free surface and the coordinate system. Let @ represent the volume of liquid, 1, the non-perturbed free surface (at 2=0), I, the free surface at any time t, and T' the walls and bottom surfaces (al these variables are also shown in Fig. 1). The liquid motion is modeled using principles of Mass and Momentum Conservation, while the tank walls and bottom are considered to be rigid. The liquid is assumed to be invisci, incompressible, and irrotational, allowing its mation to be com- pletely defined by a velocity potential function ¢. Additionally, body forces are assumed conservative and nonlinear terms areneglected. Thus, Mass and Momentum Conservation equations take the form of Laplace and Bernoulli equations, which are given, respectively, by (25) Ag=0 a here pis pressute pis the fluid density Iisthe potential ofthecan- servative forces and tis the horizontal acceleration of the tank. The term dx must be introduced because the motion of the liquid is ‘expressed with respect toa coordinate system (x2) that moves along ‘with the tank, whereas the horizontal acceleration tis expressed ith respect to a fixed coordinate system (full details can be found in [4). tis important to notice that Ag defines the velocity of any particle ofthe liquid. Therefore, Vip m corresponds to the velocity ‘ofthe liquid projected aver the normal vector n, Two normal vectors are defined: na, perpendicular tothe free surface [sand np perpen- dicular to the tank walls and bottom, Smal displacements at the free surface are assumed, in order to simply the problem, Thus, the lin- cearized boundary condition atthe free surface is given by [27|: Poona agarh In turn, the normal velocity of the liquid at Fis zero due to the Tigid condition ofthe tank walls and bottom, and the boundary con- dition at Fis given by [27] Vom =0at Ty @) AA specific manipulation of the Laplace and Bernoulli equations is requited to combine both equations such thatthe sloshing dynamics could be expressed only by the elevation of the free surface Section 2.1 shows the procedure based on the FEM to handle the Laplace ‘equation, Its important to note that Laplace equation defines only the spatial solution, in other words, the solution for the velocity of any particle of the Muid (2s a function ofthe walls and free surface velocities) but dees not give information about how the velocities change over time, Section 2.2 shows how to maniptlate the Ber nouli equation to make st compatible with the Laplace equation. ‘The combination ofthese equations addresses the temporal varia- tion, allowing forthe modeling ofthe time-history evolution of the liquid motion, Section 2.3 then discusses how the shear forces trans- mitted by the rank tothe supporting groune can be estimated, Such «a connection, ultimately between the free surface elevation and the ‘transmitted forces to the ground, allows the analysis of liquid stor age tanks supported by elastic structures (which for example facili- tate the analysis of sloshing dampers). The numerical details of all ‘these components are summarized in Section 24 2.1. Laplace equation modification Galerkin's Weighted Residual Method is applied to solve the Laplace equation, In particular, (1) is expressed in a weak formula- 3g funetion, leading to [26 3) ‘Then, Green's Fitst Mentity is applied to include boundary condi- tions in an explicit way [vivo ners] veo mar, [ov-veda-0 ant by ncang bound condone hve [ wir. [9's oi0-0 a ‘A typical FEM discretization (more details on this later is then applied to the velocity potential g. the auxiliary coordinate m and ‘the weighting function y such that: 08.2.0) = N28) 0%) = Nal ¥K.2,0) = Nexziie) (10) Itis important to mention that gis a vector built with the values of ‘the nodal velocity potential inthe volume #, and that isa vector ‘that contains the values of the nodal displacements ofthe free sur- face (here, the bar under the Greek letters is use to distinguish vec tors and functions. indicating vector variables when the bat is used and scalar functions when itis not used). Also, N is the vector of, interpolation functions for the velocity potential 9, while Ny is the vector of interpolation functions for the auxiliary coordinate » ‘within the FEM scheme. In this study linear functions are adopted for both N and Nq, BY pastiioning @ into components associated land non-assaciated with the free sutTace (@, and gy, respectively) (7) is transformed into a linear system of ordinary dilferential equations given by: G OV[%,] [Du De) [ee] an) (5 ol[a]-lor sc][2]-° “ with c= [Nigar (12) Dy De r ; p= [2x Pe] J owrone a The vector mis included only to complete the second block of equa- tons, allowing to express (11) in a matrix form. A static condensa- tion s finally performed in order to express 11) only in terms ofthe surface variables, such that the condensation is defined by: J= Dy —DuD,'Dy (14) and then becomes Gy Ja —0 15) Which is the final expression of the Laplace Equation, where the free slevation velocity and velocity potential of the free surface are related such thatthe coupling with the Bernoulli Equation is posst- ble. [tis important to notice that the static condensation introduced here [shown in (14)} is key in this procedure because: (a) it will make possible in the next section to express the equations of ‘motion in terms of variables whose physical meaning can be readily appreciated (vertical displacements ofthe liquid surface); and (b) it makes the method computationally efficient as it considerably reduces the number of variables involved. 22 Bernoulli equation modification Inorder to solve the Bernoulli equation, itis necessary to define the potential I. Under this setup, the potential of volume forces is defined as 11= gz, where g is the gravity acceleration. The Bernoull, equation is evaluated atthe free surface, imposing 2= m, Pliny =O and leg = Pleo: Then (2) is simplified as: @ + en tile Oat ly 18) Furthermore, (16) is paticularized for each node located at the free surface using the previous FEM discretization leading to: