You are on page 1of 64

Commission of the European Communities Joint Research Centre Ispra Establishment

EUR 11936/EN

An Explicit ALE Finite Element Formulation for 3D Transient Dynamic Fluid-Structure Interaction Problems

J. Donea S. Giuliani

PREPRINT November 1988

Abstract Finite element models, implemented in the computer code EURDYN3M, are presented for the prediction of the non-linear response of three-dimensional fluid-structure systems exposed to transient dynamic loading. An arbitrary Lagrangian-Eulerian kinematical description of the fluid domain is adopted in which the grid points can be displaced independently of the fluid motion. This formulation leads to an easy and accurate treatment of fluid-structure interfaces and permits significant fluid sloshing and swirling to occur without producing excessive distorsions of the computational mesh. Two numerical examples are presented to illustrate the potential of the proposed modelling procedures.


c _


1 ALE sliding surfaces 5.1.3 Semi-discrete momentum equations 3.2 Rezoning of a node with displacement constraints .2 Lagrangian sliding surfaces 5.3 Time partitioning 4 Structural analysis algorithm 5 Fluid-structure coupling 5.1 Semi-discrete mass equation 3.1.1 2.1 Computation of nodal accelerations 5.2.1 PHASE 1 : Lagrangian calculation 3.5 Arbitrary Lagrangian-Eulerian formulation Lagrangian and Eulerian descriptions of motion ALE description of motion Material.Contents 1 Introduction 2 The 3. 3.3 Numerical time integration 3. 3 .2 PHASE 2 : Connective transport 3.3 Momentum equations 5 7 7 9 11 13 15 15 16 16 19 19 20 21 21 23 24 28 29 29 30 31 32 35 36 36 36 39 40 40 3 Fluid analysis algorithm 3.3.1 Mass equation 2.1.1 Spatially discrete models 3.2 Internal energy equation 2. .2 Tangent plane orientation 5. spatial and mixed derivatives Conservation laws in the ALE description Weak variational form of the ALE conservation equations . . . . .2 2.1.2 Automatic mesh displacement prescription algorithm .1 Nodal masses and loads for fluid virtual surface .1 Rezoning of a generic node 3.3 Implementation of the algorithm 3. 2. .3.4 Semi-discrete internal energy equation 3.

5.2.2 Thin cylindrical vessel with hemispherical bottom .1 Spherical cap 6.2 Computation of nodal accelerations 41 44 44 45 6 Applications 6.

In these studies. a complicated slidesurface logic has to be introduced to treat the relative sliding at the interface between an inviscid fluid and a structure. although a clear delineation of fluid-structure interfaces is in principle afforded by the Lagrangian description. the alternative finite element method (FEM). For example. has been successfully extended to deal with problems in fluid mechanics. well proven in the structural mechanics field. These needs for safety evaluations have motivated the developement of computational methods capable to treat transient. we simply wish here to underline the leader role played by the Los Alamos. combined to the natural ability of the FEM to arbitrarily mix fluid and structural subdomains. On the other hand. This approach is limited by its inability to cope easily with the strong distorsions which often characterize flows of interest. Applications of FEM to hydro-structural problems have first been reported by Belytschko and Kennedy [2] and then by Donea. . strong distorsions could be handled with relative ease. the evaluation of structural integrity of reactor components under postulated energy excursions involves the analysis of structures of complex shape and their interaction with the fluid in which they are embedded. during these energy excursions both the fluid and the structure undergo non-linear response. but at the expense of precise interface definition and great complexity in handling fluid-structure coupling. a purely Lagrangian method was employed for the kinematical description of the fluid domain. Rather than attempting a necessarily superficial survey of the many computational techniques developed in this area (see for istance [l] for an overview). and Argonne groups in these developements. Furthermore. Until fairly recently. Such extension. if the fluid motion were described in Eulerian coordinates. In recent years. Giuliani and Fasoli-Stella [3]. Livermore. Furthermore.1 Introduction The analysis of problems in nuclear reactor safety and in several other engineering fields often requires a study of fluid-structure systems exposed to transient dynamic loading. numerical solutions to problems of fluid-structure interaction have been achieved by the finite difference method. has originated a new and very versatile computational technique for problems of fluid-structure interaction. non-linear fluid-structure interaction problems.

Because of the shortcomings of purely Lagrangian and purely Eulerian descriptions, efforts have recently been expended in the finite element area to develop integrated procedures with generalized kinematical descriptions of the fluid domain that possess both Eulerian and Lagrangian features. These generalized descriptions are usually referred to as mixed LagrangianEulerian, or Arbitrary Lagrangian-Eulerian (ALE) methods and were originally developed by Noh, Trulio and Hirt, Amsden and Cook [4,5,6] in finite difference formats. The theoretical framework for mixed LagrangianEulerian finite element descriptions has been established by Hughes et al. [7] in the context of viscous, incompressible flows. ALEfiniteelement methods for inviscid, compressibleflowshave been reported by Belytschko et al. and Donea et al. [8,9,10,11,15,17,20]. The purpose of the present report is to describe an ALE finite element method with automatic and continuous rezoning of the fluid mesh developed at the Joint Research Centre Ispra to predict the response of three-dimensional fluid-structure systems subjected to transient dynamic loading. Wefirstrecall the basic concepts underlying the Arbitrary LagrangianEulerian description and present the ALE form of the basic equations governing the flow of a compressible, inviscid fluid. This is followed by a description of the ALE fluid analysis algorithm which includes space discretization by finite elements, numerical time integration of the semi-discrete equations and a discussion of the prescription we propose for the automatic motion of the hydrodynamic grid. The structural analysis algorithm is then presented and fluid-structure coupling in the ALE formulation discussed. Finally, two numerical examples are presented to illustrate the potential of the proposed modelling procedures.


The Arbitrary Lagrangian­Eulerian formu­ lation

In this section, we first recall the b asic theoretical concepts underlying the Arbitrary Lagrangian­Eulerian (ALE) formulation. To start with, a b rief account is given of the classical concepts of La­ grangian and Eulerian descriptions of the motion of a continuum. Then the ALE description is introduced and the notion of material derivative in this generalized kinematical description is examined. On this basis, we derive ALE differential forms for the basic conservation equations of mass, momentum and energy and develop the associated weak variational forms which provide a basis for the formulation of finite element models.

2.1 L agrangian and Eulerian descriptions of motion
Consider a continuum B in the JV­dimensional vector space RN (N=l,2 or 3) with b asis e<, t = 1,...,JV and an open time interval D 0,Γ Π.€ R. We shall use the terms partielt or material point to denote a small volumetric element of the continuum B. The motion of Β is describ ed b y the trajectories along which the material points of the continuum move in space. Be X a material point of B. During time, Β always consists of the same material points though its configuration vary. We shall denote by Rx the configuration of Β at the initial time ío &nd by Rx its configuration at the present time t ; Rx is called the material domain and Rx the spatial domain. In the initial configuration Rx , a representative particle of the contin­ uum occupies a point Po in space (see Fig.l) and has the position vector X­(*),i«l,...(JV (1)

with respect to rectangular Cartesian axes. In the current configuration Rx, the particle originally at Po is located at the point Ρ and has the position vector x = (x,),» = l,...,JV (2)

The coordinates X¿ are called material coordinates, while the coordinates x, are called spatial coordinates. The motion of Β during time is described by the application *:ÄxXDO,rc­+Bw, X,t ^ χ (3) (4)

such that, at any fixed time i, the image of Rx be Rx and the image of X € Rx be χ G Rz. This yields the system of equations Si«ft<(Xj,t)tt«l,...,JV (5)

which relates the components of the position vectors X and χ of the material point X in the material domain Rx and in the spatial domain Αχ. Equation (5) may be interpreted as a mapping of the initial configuration of the continuum into its current configuration. It b assumed that such a mapping is one­to­one and continuous, with continuous partial derivatives to whatever order is required. As a conse­ quence, the Jacobian of trasformation (5) should not vanish J =




In these conditions, there exists an inverse for transformation (5) which reads * ­ * , ( % , * ) , i«l,...,JV (7) Equation (7) may be interpreted as one which provides a tracing to its original position of the particle that now occupies the location x. The displacement vector u from Rx to R* is defined by u=x­X (8)

The components of u can be expressed either in terms of the material coordinates of X through the law of motion (5) : ν­*(Χ,,<)« (9)


In fact.t)ei (10) The representation (9) of the displacement is called Lagrangian. It is noted that the Lagrangian description of motion fixes the attention on the material points of the continuum..or in terms of the spatial coordinates of X through (7) : u = Ui{xk.jv (n) represent the position vector of a reference point Q0 in the initial configuration of the referential domain Rç (see fig.l). a third domain J2¿. let e=(fc). while the Eulerian description considers a fixed portion of the space occupied by the continuum and examines what goes on there during time. but on what we shall call reference points which may be moving with the continuum. while representation (10) is called Eulerian. In order to introduce the law of motion of the reference points. in addition to the material and spatial domains Rx and Rx.»'=i.rc-> RN 9 (12) .. while in the Eulerian description the computational mesh is kept fixed and the continuum moves through it. The coordinates & of the reference point Qo ure called mixed coordinates. One then defines the motion of the reference points by the application fcrP^x Zl0. 2. It follows that in the Lagrangian description the mesh of grid points moves with the continuum. called referential domain. This referential domain will play for the reference points the role played by Rx for the material points. With reference to the same rectangular Cartesian axes as before. the ALE description fixes the attention neither on material points nor on a fixed region of space.. it is convenient to consider.. but independently from the motion of its material points.2 A L E description of motion The Arbitrary Lagrangian-Eulerian (ALE) description represents a generalization of the purely Lagrangian and purely Eulerian descriptions of the motion of a continuum.

ί) (16) Clearly. One therefore writes *i = * . each of these points will move with its own law of motion.. at any fixed time f. For the material points.· and t.<) at at (17) Relation (17) gives a Lagrangian representation of the velocity..t) Therefore. t f „ < ) . for the mapping represented by Φ to be one­to­one. As far as the reference points are concerned. the velocity ν at time t is given by the time derivative of the position vector (5) taken with X held fixed v = ax(x i l t) 3u(X.t) at (19) .< ~ χ (is) such that. · · · » # (I 4 ) and it is assumed that the vector function Φ and its partial derivatives to whatever order is required are continuous in (. Later on.£. the displacement vector û from P { to Rz is given by û = x .ί) = Φ(&. at any time f. one material point Ρ with coordinates X.* = û(6k. one may associate with each point χ of the spatial domain P .*) at 10 aû(fr. In addition. the image of R¿ be Rx and any given point χ of the spatial domain Rx be the image of a reference point ζ € Rç. · = ! . the mixed Jacobian J of transformation (14) should not vanish J = dxi #0 (15) àîi At this point.· in the material domain Rx and one reference point Q with coordinates £. the above two points only coincide in χ at a given time t. it is important to note that. in the reference domain P £ χ = Φ(Χ. the velocity w of the reference points is given by w = (18) **(£*.

Details on the practical selection of the mesh velocity w are given in section 3. It is here denoted by the symbol ^ £ . 11 . we have seen that the ALE description has no basic depen­ dence on particles and treats the computational mesh as a reference frame which may be moving with an arbitrary velocity w . one can recover the classical kinematical descriptions. the following basic viewpoints may be individuated: 1. spatial and mixed derivatives Let us consider a physical property F of the continuum Β and denote by f(xj. 3. w = 0 : the computational mesh is fixed in space. selecting P j = Ρ χ and Φ = Φ yields the Lagrangian description for which relations (18.*) its spatial. choosing P¿ = P s and Φ as the identity transformation. f*(X j .By specializing the definition of the reference domain P¿ and of the law Φ of its motion.19) give û = χ ­ X = u w = ν A (20) (21) On the other hand. material and mixed representations. w = ν : the mesh moves in space at the same velocity as the particles.19) give û = w = 0 (22) Summarizing. In order to evaluate the time rate-of-change of property P . we recover the Eulerian description. this corresponds to the Lagrangian viewpoint. this corresponds to the Eulerian viewpoint. one has the choice between three types of derivative : • The material derivative is the variation of F per unit time felt by an observer moving with the material points. 2. Depending on the value of the velocity w . w φ ν φ 0 : the mesh moves in space at a velocity w which is in principle arbitrary.3 Material.2. In this case. For instance. respectively. this is the ALE viewpoint.t). £ = χ and relations (18. 2.t) and ƒ**(£/.

*. its material derivative includes two terms and is given by DF Dt dij df dx at x i df at X + *(*. To see this. since DF Dt ar(x. íi¿ as». (28) "at at at at 12 .<) = *. i).. i.<) (27) and take its partial derivative with respect to time with X held fixed.e. when F is given in mixed representation by ƒ**(£.Q at (23) When F is given in spatial representation.«) (24) X ¿ i (25) The first term in the right­hand side of (25) is the spatial derivative and the second term is the convective term of the material derivative.(Χ.(£. it is here represented by the notation ^ When Ρ is in material representation. it is here represented by the notation ^ . we consider relation (16).• The spatial derivative of Ρ is the variation per unit time at a fixed point in space. the second term is the convective term which may be expressed in terms of the relative velocity (v — w) of the material point with respect to the reference point. One obtains dxj 3Φ.·.· = Φ. its material derivative reads DF Dt df" dt df' at (26) The first term in the right-hand side of (26) is the mixed derivative of P. Finally. its material derivative is readily calculated. • The mixed derivative of Ρ is the variation per unit time felt by an observer moving with the reference points.

4 Conservation laws in the ALE description To simplify the subsequent developements. i. ­ Wj) c df' dti Ëk tdXj (31) DF Dt ar at + («i .or.v>j) Bf{xtt) dij (32) As we shall see in the next section. 2. in view of the definitions (17) of the particle velocity ν and (19) of the reference point velocity w: Vj = Wj + dij Hi at (29) Expression (29) indicates that a& at χ ι ^ dt< °x> (30) so that expression ( 26 ) of the material derivative in mixed description can be transformed into DF Dt' or. mo­ mentum and energy.e. α = Vi. represents the components of the relative velocity of the material point X with respect to the reference point ζ. expression (32) of the material deriva­ tive in mixed description is of fundamental importance for the formulation of ALE differential forms of the basic conservation equations for mass. while c. equivalently df" at + (υ.Wi (34) 13 . we shall write the material derivative in the ALE description given in (32) in the form Dt m r + CiFj (33) where F* indicates the partial derivative with respect to time for a given reference point (.

the internal energy equation may be rewritten in the computationally more attractive form (eO * + Ciiet) ·.In (33) and the following equations. g is the density. Thermal effects have been neglected in the form (37) of energy conservation. σ. N • Energy equation oe* + ode. + ¿να = a^ Dtj 14 (42) . .i (35) • Momentum equation QVi + QCjVij = o#j + obi. One may also write an internal energy equation in the form oi' + DCii¿ = OjDij where D» = \{vij + v») (40) is the rate of deformation (or stretching) tensor.· Cauchy stress tensor. a comma followed by an index indicates partial derivative with respect to the corresponding spatial coordinate.·. Since «·" « ( « · ) · ­ # (41) (39) we note that. the equations which govern the continuum in the ALE description are the following three conservation equations (see references [13. With the above notations.* = (ai. % = 1 .15.20] for details): • Mass equation Q* + Cißj = ~ßVi. b body forces per unit mass and e the total internal energy given by e = » + ^ViVi (38) where t is the specific internal energy. using the mass equation (35). .v¿)¿ + ob^ (37) (36) In relations (35­37). . .

i 2. we have Vg = 0 and the weighting function Φ may be chosen as unity.1 Mass equation The ALE differential form (35) of the mass equation may be rewritten as ■57 = (w ­ v) · Ve ­ ßV · ν (46) dt where.í oV-ydV = .­ the Kronecher delta.i ev-ndS K Jv{t) dt Jv[t) 75(0 15 (48) .5 (45) Weak variational form of the ALE conservation equations We shall now develop weak variational forms of the ALE conservation laws which will provide the basis for space discretization using finite element modelling.5.For an inviscid compressible fluid one has σ„ = -p6ij where ρ is the pressure and £. 2.< = ~(e» + P)vi. In this case OijDij = -pviti and the internal energy equation ( 42 ) reads (44) (43) (ßO* + «<(e0. Multiplying this equation by an arbitrary weighting function Ψ and integrating over a control volume V(t). the time derivative ρ* for an observer moving with the reference points is noted | j . for simplicity.­. It follows that (47) reduces to the simple form I ^dV = . / V^dV = I * v (w­v)­VedV­ / VoV>\dV Jv(t) dt Jv(t) ' Jv{t) * (47) v . In this case. we obtain the following weak form of the mass conservation equation. A useful particular form of (47) may be derived by assuming that the density is constant over the control volume V(t).

the following integral statement may be deduced from the weak form (51) of the internal energy equation ^r f oidV=l pV­vdV (52) e»(w­v)nd5­ / dtJvit) ISM Jvit) This integral form will be used in connection with ALEfiniteelement mod­ els in which an elementwise uniform density of internal energy is assumed. the ALE differential form (36) of momentum conser­ vation reduces to the simple form fi =fi(w fr "v) *Vv*+ebi " àk 16 (53) Equation (53) must be supplemented with appropriate boundary condi­ tions. We shall assume that on portion Sχ of the boundary the velocity .5. ' + p ) V ' V d F (51) If the density of internal energy DÌ is assumed uniform over the control volume V(t).which may be further transformed into d — I odV = I g(w ­ ν) · n dS (49) dtJvit) J sit) This integral form of the statement of conservation of mass will be em­ ployed in finite element models assuming an elementwise uniform density. In passing from (48) to (49) we have used the identity dV ^ ­ = V­wdT at 2. It reads /vm*a7 ( e 0 ^ = / v m * ( W " V ) ' V ( ß 0 ^ " / v w * ( e .2 Internal energy equation (50) The weak form of the internal energy equation (45) is derived in complete analogy with what has been done for the mass equation.3 Momentum equations For an inviscid fluid.5. 2.

After integration by parts of the pressure prescribed.ß(w .e. 17 . vanishing on Si) variation ¿v.· of the fluid velocity followed by integration over V(t).n&ip = Ti (54) The weak form of the momentum equation is obtained multiplying (53) by an arbitrary admissible (i. the following result is obtained m SI m I SviQ-^-dV Jv[t) at m [ öü.v) · Vvi. use of the divergence theorem and of the stress boundary relation (54). whereas on the remaining portion S2 the components T¿ of boundary loads are imposed .dV + Jv{t) Jv(t) I Sviobi dV + I pS-{6vi)dV + SviTi dS (55) i Equation (55) provides the basis for the formulation of spatially discrete models of the momentum equations using finite element approximations.

18 .Spatial Domain Material Domain Referential Domain Figure 1: Schematic diagram of domains and mappings for the ALE description.

3. to integrate the semi­discrete conservation equations forward in time to obtain the transient response. • t denotes a typical finite element. we will first define the basic variables in the form of local approximations over a typical finite element. to derive spatially discrete models of the ALE conservation equations using finite element modelling. We will use the following notation : • u. to prescribe the kinematics of the hydrodynamic mesh. i. lower case subscripts denote components of a vector.t) in an element can be described as product of shape functions N¡(x) that are independent of time and nodal values u. The usual assumption in discretizing partial differential equations in finite element schemes is that the functional dependence of the variables in space and time can be separated.. These are : 1. so that any local vector field u«(z.e. 3.3 Fluid analysis algorithm Three distinct tasks are involved in the construction of an ALE finite ele­ ment algorithm for transient fluid flow. • u¡ upper case subscripts denote node numbers./(t) which are independent of χ and incorporate the time dependence tt.1 Spatially discrete models For the purpose of providing a framework for our discussion of spatially discrete models of the ALE conservation equations. • · denotes a time derivative. to specify the nodal values of the grid velocity w. 2.t) = Nt{x)ua{t) (56) 19 .(x.

t) = Nt{x)va{t) w<(x. 5 and 6-node elements (tetrahedron.Here the fluid velocity ν and the arbitrary grid velocity w will be interpolated over a typical finite element by the same shape functions t../ w ­ v >­ n < i s < 59 > where Σ . 4. For thefluiddomain discretization.)f 20 + (l + aj)e·} (60) . the density gj is computed as a weighted average of the densities in the elements e and e on either side of face J H = \[{l-a. the discussion of 3Dfluidfiniteelement models will be restricted to the 8-node hexahedral element which has been implemented in the computer code EURDYN-3M.1 Semi-discrete mass equation In updating the mass of an element.(x. the integral form (49) of the conservation of mass is used.1. pyramid and prism) are also employed but they are treated by the code as hexahedral elements in which some nodes are coincident./(0 (57) Another set of shape functions is usually employed for the local description of the density Q and for the density of internal energy gi ρ(χ. In the remainder of this report. In general. the fluid properties in ( 58 ) are interpolated by a polynomial which is at least one order less than that for the velocities.t) .9j{x)oij{t) (58) While shape functions Ν must ensure inter-element continuity of the velocities ν and w.t) = JV>(x)u>. this is not necessarily the case for shape functions Φ. 3.ί) = *j{x)ej(t) ei(x. which gives !? £ = 3ï/v. To stabilize the mass equation. The density and specific internal energy are assumed elementwise constant and a trilinear local approximation is used to describe the fluid and mesh velocities. and the velocities are evaluated in terms of nodal values as indicated by (57). indicates a summation over the eight faces Sj of element e.«'n' = Ç i '/.

the flux is entering the element e and density g* is favoured in the weighted average in equation (60) by choosing aj > 0.For the evaluation of the coefficient otj wefirstcompute the flux F through face J by F= f ( w . 3.1. if F is negative the flux leaves element e and density g' is favoured by choosing ctj < 0.n d S (61) JSj If F is positive.1.3 Semi-discrete momentum equations (64) The principle of virtual power (55) is applied to derive the semi-discrete equations expressing conservation of momentum in thefiniteelement mesh. To satisfy these requirements we take aj = a1iSign{F) (62) where 0 < a 0 < l . Conversely.v ) . = |((1 . is constant over the element.2 Semi-discrete internal energy equation Proceeding as described in the previous section for the mass equation and noting that the pressure. defined by an equation of state of the form ρ = f {β.aj){gi)' + (1 + aj)(gi)·) where ctj is function of 04 as in (62) and the flux F is given by (61) 3. a0 = 0 corresponds to a centered approximation. According to the Galerkin method. o j o = l gives a full donor approximation.i). the velocity variation is defined as Svi = Ni{x)6vu 21 (65) . from (52) we can derive the rate of change of internal energy in an element by !/ v .* W = E(*0//5/(w-v) ndS-p'^vndS (63) The energy density gi is computed by UK).

as shown by ( 66 ).^ f Ì p ' ™ 22 W . of element contributions of the type (67) M}j = J gNjNjdV • v denotes the global vector of nodal accelerations. as shown in (66). the discrete analog to the momentum equations is obtained. the repetition of subscript J implies a summation over all nodes in element e. (68) • Ft indicates global nodal loads induced by transport of momentum. a typical component of vector F t b obtained by assembly of element contributions of the form P«')' « fy. Independent variations of all velocity degrees of freedom in the mesh are introduced in turn into the principle of virtual power (55) and by invoking the arbitrariness of such variations. and subscript » denotes components of a vector. The semi-discrete equation governing a typical node I b found in the form ^l. The governing equations for the discrete mesh form a set of ordinary differential equations in time which may be written in condensed form as [M]v = F t + F è + F p -rF where • [M] b the global mass matrix consisting. ""t™ .NtgNjVijdV m YtJ^NIg{wj-vi)j^ViJdV + W NjTidS (66) where £)e indicates a summation over all the elements to which the node I belongs.where the iV/'s are the velocity shape functions defined in (57).

The viscous pressure used in EURDYN­3M b composed.8 for explicit Lagrangian calculations. a b the dilatational wave speed and CA. [23]. b added to the fluid pressure ρ when calculating shock propagation problems. of a quadratic and a linear term in the divergence of the fluid velocity: g _ Í e\C\l*{V · v) 2 ­ CLl aV · v] when V · v < 0 \ 0 when V v > 0 . According to Wilkins et al. . 3.• F» denotes global nodal loads due to body forces g6< &<. 23 . CA « 2. the practical implementation of the ALE description requires that an au­ tomatic mesh déplacement prescription algorithm be supplied.NiebidV • Fp represents the nodal loads induced by the fluid pressure ρ (70) v-r-Lnz« • F accounts for the externally applied loads Ti (F. q. The formulation suggested by Flanagan and Be­ lytschko [24] has been implemented. However. where Vt b the element volume. CL are numerical coefficients.2 Automatic mesh displacement prescription algo­ rithm The freedom in moving the fluid mesh offered by the ALE formulation b very attractive. / = \/Ve. C¿ « 0.7)e = / NjTidS (71) (72) An artificial viscous pressure. as suggested by Wilkins [23]. As a consequence. Anti­hourglassing forces may abo be applied in EURDYN­3M to control zero energy modes which arbe due to one­point integration of first­order isoparametric elements. it can be overshadowed by the burden of specify­ ing grid velocities w well suited to a particular problem.)' = fY. * 73 ' In thb expression.

J.R. pyramid and tetrahedron). 5 and 4-node elements (hexahedron. 3. K.C. Let Δι. These elements with linear.In the following paragraph we describe an automatic mesh displacement prescription algorithm for three-dimensional finite element meshes consisting of 8. Consider a generic tetrahedron (I. An early version of the 3D rezone algorithm was presented in [16].2. In other words.2.r.ζ components). Consider. Ispra ALE computer code EURD YN-3M. L) in the influence domain of node I as shown in fig. after rezone the tetrahedron basb area a (opposite to node I) and height h are given respectively by o = yjpp + qq-\-rr (74) h = (ρΔχχ + qAx2 + rAx s + 3u ) / o where. The basic task of the rezone algorithm b to displace node I so as to render the tetrahedra defining its influence domain as regular as possible. a portion of a three-dimensional finite element mesh consisting of hexahedral. pyramid and/or tetrahedral elements.u are the following determinants zzj 1 *u XJJT * S K 2 X*L XSL (75) (76) 1 1 1 Ρ = 24 . To a generic nodal point I we associate an influence domain (dashed in fig. we wbh to minimize both the squeeze and the dbtorsion of each tetrahedron in the influence domain of a given node.q.3 be the displacements of node I required to achieve its rezone (t indicates x.y. prbm.. as shown in fig. defined by the set of tetrahedra obtained by joining node I to all nodes connected to it via the element sides. v' indicates the volume of the tetrahedron before rezoning υ = {pxu + qxu + rx s / + u)/3 while p.1 Rezoning of a generic node. or trilinear local velocity field are the only 3D hydrodynamic elements currently implemented in the J. prbm. t = 1.

The coordinates of the centroid of the triangle defined by nodes J.3) are ii = {zu + xiK + x. K.3) and the dbtances between these points and the centroid are s D'j = Σ(χ</ + Δχ< ­ ï.2.) * {xij ­ *i)/D j i=l 25 .3 (79) and the distances between J. L (fig. K.XlJ xur XlL Xi/ XSJ 1 1 1 1 1 *SK XSL xw XJJC ' = 2 XLIC XlL XlL Xij 1 XSJ xu " = "S XJJC XlL *2K *SK XJL XSL (77) The total volume V of the influence domain and its mean height h are v = Σ>' h = ZV Ση a (78) where η represents the number of tetrahedra in the influence domain. L and the centroid are Dj = DK DL = = (80) The projection of I on the lines joining the centroid and J. K% L gives the points J\K\L' (fig. r )/3. » = 1.

ΚΫ + Wt Σ.*i)* {xiL .* i ) ' (xiir .*») (ΧΙΧ-*0(Χ«Τ-*0 i>î x?i ^îr .D 'K = Σ(χα s + Δ*ί-*ί)*{Χίκ-£ί)/Οκ D'L = Yí{XiI + ^Xi-Xi)*{XiL-^i)lDL (81) «=ι These distances are clearly a measure of the distorsion of the tetrahedron. while the difference h — h b a measure of its squeeze. + (XlX. yields a system of linear equations [S] {Δχ. Considering the influence domain of node ƒ as a whole we define E = Wr Σ {h .} = {R} (84) where the elements of the symmetric matrix [S] and of the right-hand vector {R} are WaE ^ (XIJ .Xi){xu .Ι&ιΫ η η + (D'K)* + (DLY\ = minimum (82) as a criterion for minimizing globally the squeeze and distorsion of the tetrahedra in the influence domain.*i)* + Dj DK + D\ s» = ^ E S + w 'Vi [(XIJ . where w.­gl)(x»L­g>)1 J s» = ι*Σδ+ 26 . = w* = 1 η 4En(Dj + DK + Dl) (83) The minimization of (82) with respect to Δ ι .

)PL] ~w* ΣΚ1" η ­ * S ) P J + (xs* ­ X»)PJC + (xsx ­ XS)PL] If To denotes the determinant of the matrix [S] the solution of the linear 27 .Is) Dj + {X2K .Xs) Dì S22 = ^ΙΣ4+ wÆ {Xjj-Zi)* Dj + (Χικ-**)* DK + (XìL-2ì) Di l-ì s» = * Σ 5 + ^ιΣ (XU . fol ­ £ 3 ) ~w* Σ Κ * " ­ *i)*V + (xijr ­ XI)PJC + (XIL ­ *ι)Ρι] ~w* Σ ( ( * " ­ **)pJ + ( * « ­ *»)Pjr + (X 2 L ­ X.ZÌ){xsK ­ îs) DK (X2L .It){XsL -li) Dì S* = ^ΣΤ2 W + Î1 V [ ( ^ ­ g » ) a .*2){xsJ .*s) (XlL ­ Xi){xsL .NÆ (Xu ­ xi){xsj ­ Is) Dj + (χικ - *I){XSK DK . (*»* ­ XsY .

t = 1.algebraic system (84) b 1 Ri Su Δχι = — Ri Sn Γ0 Äs sn 1 Sn Ri Δχ* = — S21 Ri To Ssi Rs 1 Sis Sa Sss Sis Sa Sss (85) Sn Su Ri Axs = — Sn Su Ri To Ssi Ssi Rs 3.=i s is y=i is λ s is \j=i Is Σ ^w Ι Σ cit Sij J Σ ^i» Ι Σ CtjSij I i=i \ s (87) If the nodal déplacement b constrained to occur along a line with direction cosines C. iel ÖLiCrtÖ-iC^.2. the minimization of E in (82) with thb constraint gives Δχ ί = Σ σ ί < ' ί ' » = 1 ' 2 ' 3 j=i (86) where Cu and C Jt . * = 1.2.3 (88) where t* b given by *' = Σ ^2< Ι Σ CijSij J Σ Cx Ι Σ c*jS%j fil- T.2 Rezoning of a node with displacement constraints If the nodal displacement b constrained to lie in a plane. the minimization of JE in (82) with thb constraint gives Axi = Ctt\ i = are easily obtained by solving the following system of equations s .3 are the direction cosines of two orthogonal directions lieing in the plane and ii.2.3.) 28 (89) .CiiRi is1 1 σ Σ «Λ » ..

However.3. To enhance the numerical efficiency of explicit time integration algorithms. in which different time steps are used in different parts of the mesh. Let ƒ be for a given node the ratio of the grid velocity and the orthogonal projection on it of the fluid velocity. momentum and internal energy b divided into two phases : Phase 1 b an explicit Lagrangian phase in which convective contributions are ignored and Phase 2..3 Numerical time integration The algorithm for numerical time integration of the semi-discrete conservation equations for mass.2. The magnitude of the final grid velocity b then evaluated by w\ = -Xvi W¡ = Wi w¡ = (1 -r Atø for : ƒ < . partitioning techniques are being used.λ for : . In expression (91) A b a numerical coefficient (0 < λ < 1).3 Implementation of the algorithm The rezone algorithm described above has been implemented into the ALE finite element code EURDYN-3M. The code uses an explicit time-integration method and at each time station a first approximation to the grid velocity b computed for each ALE node by * = «T" + ^ (90) where At b the time-step size and Δχ.are given by relations (85). b the convective phase. The components of the grid velocity computed by relation (90) are then limited by an empirical inequality of the type (-λ) < ƒ < (1 + λ) (91) which b dictated by numerical stability and accuracy requirements. Explicit time integration methods are only conditionally stable and the time-step size Δί must be limited so that sound waves do not travel more than one element during the time increment. to simplify the exposition of the two phases in the 29 .λ < ƒ < (1 + λ) for : ƒ > (1 + λ) (92) 3. so as to fully exploit local stability conditions. which b abo explicit.

. The two phases in the explicit time integration are as follows : 3.105 below). + j. so that tn = J^At and tn+1 = tn + At.3. Then. time integration based on a single time step will be considered first.« i(pf + iç)(v?­-ν?) 2 Aíf (97) (98) 30 .3.1 PHASE 1 : Lagrangian calculation Let superscript n denote the time level. FJ is the nodal load (69) induced by momentum transport arising from the previous time­step Phase 2 calculation (see eq.present time integration procedure.3. the following Lagrangian quantities (denoted by the superscript L) are evaluated : • Nodal coordinates : XL = = χ η + Δίν η + 1 ' 2 (94) (95) • Element volume Element density : vtL = ƒ (xL) • : 0L _ M? (96) • Element internal energy andÌ pressure (by iteration on97 and 98) : <f pf if = = = C ç fit f.» + jm + | m | ^ Here. D etaib on the practical implementation of partitioning techniques will be given in section 3. Thefluidvelocity at time t n+1 /' b calculated first from the momentum equations (67) and using a diagonal mass matrix Vn+1/J = v n­l/2 + Δ* [j.

2 and 3. contributions from the convective terms in the conservation equations must be added to the time-advanced L-values. Thb b the objective of Phase 2. In the Eulerian or ALE case.t/. dS (101) •r ' = $ i < f + $ » ƒ. 3.The above L-values represent the end-of-step values in case of a purely Lagrangian calculation. «f-f (-» .)n+1/2n.3.63) : (102) + Δί ƒ m gL{wi .3.2 PHASE 2 : Convective transport We assume that the mesh velocities w n+1 / 2 have been evaluated using the automatic rezone atgorithm described in sections 3."i)"*1'*^.»>>"+1/.1.. & Thb completes the two steps contained in a cycle of the explicit time integration procedure. 31 .59) : AC"1 = K • Element density : * " = ^ΪΓ • Element internal energy (see eq.2.n+1 = /(x n + 1 ) (100) • Element mass (see eq.^*) uw) (105) • Nodal loads due to momentum transport (see eq.3.2. The following end-of-step quantities are then evaluated in sequence : • Nodal coordinates : xn+l=xn + Atwn+l/2 ^ • Element volume : V.2.»> « s • Element pressure : (irø) »τ1 -/ur 1 .69) : Ftu = ƒ Nrfiwj .

• the time step Atj used in the integration of the equation of motion at each node b given by: 2*Δί^ η < At.·η b the minimum critical time step found in the mesh and k b a positive integer.η (106) where Δί^. first introduced by Belytschko [21] (see abo [22]). Thb improves the efficiency of the numerical time integration process and saves computer time.3. The element computations required to evaluate the nodal loads are consequently reduced with respect to a procedure based on a single time step throughout the mesh. < 2*+1Δί^. • when the node with the largest critical time step in the mesh has been integrated forward in time. 32 . Linear displacement interpolations are used for the element nodes with larger critical time steps. With thb technique the nodes with the smallest critical time step are integrated at every time step.3. the critical time step Δί} b the minimum critical time step found in the elements to which the node belongs. It uses the following rules: • fort each nodal point I.3 Time partitioning An explicit-explicit time partitioning. • the element variables (density. has been implemented in the code EURDYN-3M. all the element variables are advanced and a new time-partitioning cycle b entered. internal energy and pressure) are updated whenever a new acceleration has to be computed at a node of the element. while the other nodes are only periodically processed.

Figure 2: Influence domain of a node I in a three-dimensional mesh. 33 .

Figure 3: Generic tetrahedron in the influence domain of node I. 34 .

The material b assumed to satisfy the von Mbes yield criterion and an botropic hardening rule b considered.4 Structural analysis algorithm A 4-node thin-shell element b used in EURDYN-3M to model the containment vessel influid-structuresystems. The formulation of the element follows the lines indicated by Hughes [25].[26].[27]. The elastic-plastic response of the structural material b idealized through a trilinear stress-strain diagram. Numerical integration of the constitutive law b based on the classical elastic predictor-radial return concept.[28] and b presented in detail by Casadei in reference [29]. The element results from the degeneration of an 8-node boparametric continuum element obtained upon introduction of classical thin-shell hypotheses. An explicit central-difference method b used to update displacements and velocities in the structural mesh: u n + 1 = η η + Δίν η + 1 /' v ^ ' + Aia" vn+i/a = (107) (108) 35 .

so that all nodes on the sliding interfaces remain permanently aligned. The second technique b used where fluid and solid nodes cannot be kept aligned.1 Computation of nodal accelerations In order to illustrate the coupling procedure alongfluid-solidinterfaces.1 ALE sliding surfaces The purpose of thb paragraph b to show thatfluid-structurecoupling may be achieved in a very simple and elegant manner if the fluid b treated in the ALE formulation. As shown in fig. The following conditions are prescribed at each point of the interface between an inviscid fluid and a deforming structure: 36 . we may constrain the fluid nodes to remain contiguous to the structural nodes. in thb case Lagrangian sliding surfaces are employed. so that all nodes on the sliding interface remain permanently aligned. 5. The first applies to the permanently submerged parts of the structural domain and makes use of the freedom allowed by the ALE technique to force the fluid nodes to remain contiguous to the structural nodes. It b clear that such a permanent alignment of nodes along the interface greatly facilitates the flow of information between fluid and structural domains and permitsfluid-structurecoupling to be effected in the simplest manner. the movement of the fluid mesh may be chosen completely independent from the movement of the fluid itself.4. two nodes are placed at each point of the interface: one fluid node and one structural node. 5.1. let us consider a structural member embedded in a fluid which b allowed to slide along the faces of the structure. In particular.5 Fluid-structure coupling Suitable conditions must be prescribed alongfluid-solidinterfaces to allow relative sliding of thefluidand solid. Two distinct techniques are employed in EURDYN-3M to deal with fluid-solid interfaces. Since the fluid b treated in the ALE formulation.

In order to specify the above conditions.Vsn (HO) Since the two nodes must have a common normal velocity at all times.t2 to indicate respectively the normal and tangential directions.Vsn = 37 TJ (115) . 2. n b set up at each node of the interface (fig. ­ v s ) · -£ = 0 Now dn dt η'+Δ.4. 3. a local coordinate system í i . The condition of common normal velocity implies that VFn .­η« At (ill) (112) (113) (114) and since (v*­vj)n = 0 relation (112) reduces to VSn — Vf* . referring to fig. ti.n] = 0 1 dt or (v* ­ v s ) · n + ( v .4). See next paragraph for the selection of the tangent plane orientation. we prescribe that ¿[(v. the normal velocity of the fluid coincides with the normal velocity of the solid. Λ Vfn . the grid velocity of the fluid coincides with the material velocity of the solid.­v s ). so that tx and t2 lie in a plane tangent to the sliding interface and n b normal to it. í j . the condition under point (1) implies that W/r = v 5 (109) The conditions under points (2) and (3) are enforced through the nodal loads. S to indicate the fluid and the solid and n.1. the tangential velocities of the fluid and the solid are unconstrained. Indicating by w the grid velocity and by v the material velocity and using the subscripts F.

The tangential velocities at the fluid and solid nodes are unconstrained and result from the equations of motion Msvsti — Fati 38 .3 contributed by the relevant fluid and solid elements. the complementary accelerations are linked by *Fn — &Sn = ^7 = A„ (118) and a second relationship between these accelerations may be obtained from the requirement of equilibrium which reads MFaFn + Msasn = 0 It follows that * Fn (119) Ms * MF + A MS*" "» = W^Ms^ (120) Thb completes the computation of normal accelerations which ensure a common normal velocity at the fluid and solid nodes on the interface. Mp and Ms are respectively the mass contributed by the elements adjacent to the fluid and solid nodes.4. and aj?n. i = 1. the condition of common normal velocity implies different normal accelerations at the two nodes if the normal to the interface changes direction with time.í2. referring to fig. In view of (115 and 116).2.n)-system of the assembled internal nodal loads F{. To evaluate these normal accelerations.aSn are individual complementary accelerations. The common normal acceleration o„ b obtained from the equation of motion {MF+Ms)an = Fn (117) where. we pose VFn = VSn On + ttpn On + asn (H6) where an is a common normal acceleration.Thus. The normal load Fn b obtained by transformation into the (í1.

MsVst.xj. the fluid tangential velocity at the inteface node may produce a spurious transport of fluid across the intersecting sides. x s — Χχ and Χχ — Xj. Posing Μ Σ a? and dividing equation (122) by 5 we obtain t"-f-=0 i=l ώ ( 123 ) (124) Thb means that the spurious transport of fluid vanishes if one selects the tangent plane as being perpendicular to the direction defined by the following direction cosines c o s a g l i .3.3 are the components of the tangential velocity in the global reference system Xi.­. « = 1. if the tangent plane orientation b not properly defined. If S. It follows that.3 39 (125) . = Fsta MFVFÍI = FFtx MFVFI* = FFti (121) where the tangential loads are again obtained by transformation into the {ti. The amount of fluid lost on some faces may be compensated by the amount gained on other faces by a suitable choice of the tangent plane orientation to which the tangential velocity b assumed to be parallel.2 Tangent plane orientation In general the element faces around a node belonging to a sliding surface are not parallel.2.. i = 1.t2. » = 1. 5.2. an exact compensation of the spurious fluxes can be obtained by imposing the following condition s Σ>£ =0 i=l (122) where v.2. i = 1.1.n)~ system of the assembled internal forces JF¡.3 are the projections of the surfaces of the faces sur­ rounding a given node on the planes x2 — x3.xj.2.­.

1 Nodal masses and loads for fluid virtual surface In the actual fluid sliding surface we define the domain Dj of a given node I as the sum of the surfaces of the element faces to which the node belongs.2. It then follows that the nodes of the fluid virtual surface and of the solid surface are aligned and their accelerations can be computed as described in section 5. The nodal loads and masses Fu.5. In EURDYN-3M the problem of fluid-structure coupling along Lagrangian sliding surfaces b faced by preparing a new fluid virtual surface. equivalent to the real one.5). 5. near a fluid free surface).1. We assume that the loads F¡e and mass Mfe per unit surface in an element face are given by the sum over the nodes defining the face itself of the nodal loads and masses divided by their domain surface F/e Fi \r* Fu \ D J M* = Ç ^ (126) where t indicates the vector components. The interface involves two Lagrangian surfaces: the first consists of fluid nodes while the second consists of structural nodes. Every face of the actual fluid surface covers one or several parts of the virtual surface faces (fig. interface nodes cannot be kept aligned and Lagrangian sliding surfaces must be introduced to treat fluid-structure interaction.Mj of thb last surface may be obtained from the contributions of every covered part Sp by the following integrals Fij = / NjF/'dS is· Mj = ƒ fNjM/edS (127) 40 .2 Lagrangian sliding surfaces Under some circumstances (e. but with nodes which coincide with those of the solid surface.g. and I the face nodes.

5. while for the solid nodes they are multipUed by the shape functions and added to those obtained by (117). 41 . These integrals are numerically evaluated in the code using a 2 χ 2 Gaussian rule.where F/e. and Nj are the corresponding shape functions. The tangential fluid accelerations.2 Computation of nodal accelerations Considering thefluidvirtual surface and the solid surface and assuming the tangent plane orientation as defined in section 5. that are unconstrained. the nodal normal accelerations can easily be computed by relation (117) while the tangential accelerations for the solid result from the equa­ tions of motion Msvsti Msvst3 — Fstx = Fsti (128) The normal acceleration for the actualfluidnodes b set equal to the normal acceleration at the adjacent point on the solid surface and computed using relation VFn = T.1.NJVSnJ (129) J where Nj are the shape functions of the adjacent solid node and J the relevant nodes of the solid element face.2 to obtain the (ti.2. n)system.t2.Míe are the loads and mass per unit surface of the relevant face. may be ob­ tained from the equations of motion MpilFtx MFVFU = Fpti = f>ia (130) The complementary accelerations are computed using (120) where masses and velocities for the fluid are the nodal values while for the solid they are evaluated at the adjacent point in the solid face by the relations VSn = J^NjOSnJ J M8 = ΣΝ'Μ" 3 (131) For the fluid nodes the complementary accelerations are then added to those abtained by (129).

n) at each node of thefluid-solidinteface.t*. Local coordinate system (ti.Fluid Figure 4: ALE sliding surfaces. 42 .

Actual and virtual fluid surface in the fluid-solid interface.Figure 5: Lagrangian sliding surfaces. 43 .

8.7) and suitable displacement and rotation constraints have been imposed to the nodes belonging to the symmetry plane. first hardening slope: 2. recently developed at J. externally loaded with a constant pressure P(t) = 600 starting at time t = 0.6). 44 .3.6 Applications As an illustration of the ALE finite element modelling procedures discussed in the present paper. we shall present results for two test problems obtained using the new three-dimensional finite element computer code EURDYN3M. release 1. • bilinear stress-strain law (botropic hardening): first yield stress: 2.1 Spherical cap An academic problem consisting of a rigidly clamped spherical cap (fig.R.9. Only a symmetric sector has been dbcretized using 20 shell elements (fig. • Young's modulus: 1.050 x 107. Computed deformed meshes at various times are shown in fig. Ispra. have been chosen for their ability to validate the new code against exbting reference solutions. • Pobson's ratio: 0. These two problems. The properties of the employed elastic-plastic material are (nondimensional units): • initial density: 2.C.4 χ IO4. has been run to check the structural module of the EURDYN-3M code. 6. Our computations are in good agreement with results obtained using different elements and codes [18].1 χ 10s. the first of which b purely structural while the second involvesfluid-structureinteraction. Hbtories of the external pressure work and of the apex deflection are shown infig.45 χ IO"4.

except for the 9 edges near the free surface where Lagrangian sliding surfaces have been introduced to deal with interface nodes which do not remain aligned. • constitutive law: Ρ = A[l-Ei/{CV)]exp-cv +B[l-Ei/(DV)]exp-DV +{eoi-Ei)E2/V {MPa) where: V = go/e. Sliding of water along theflexiblevessel has been treated in ALE form.5.0. The top of the vessel b clamped on a rigid cover. The air gap above the water free surface b modelled as a void. as well as the water free surface have been taken as purely consbts of 23 shell elements for the vessel and 258 hydrodynamic elements to model the charge and the water. The properties of the explosive charge are [19]: • initial density g0 = 270 (Kg/ms).2 Thin cylindrical vessel with hemispherical bottom As example offluid-structureinteraction. The vessel b nearly completely filled with water and loaded by the detonation of an explosive charge located on the axis. Β = 1159. Λ = 17039. 11 shows the vessel thickness as a function of the distance from the rigid cover. 10). Fig. The employed finite element mesh for a symmetric sector of the configuration b shown infig. the dynamic response of a thin cylindrical vessel with hembpherical bottom has been computed (fig. The interface between the explosive charge and the water. 45 .6. The rest of the hydrodynamic domain has been treated in the ALE formulation and the corresponding nodes have been moved automatically by the computer program.12.

0 x lO'PiP.0 ebe Ps = [Ζ + ^ . • Poisson'e ratio: 0.796 Pu = 0.1483.0.C = 9.τ Α . and: g0i0 = 1629. • constitutive law: Ρ = 1.0 {MPa m*/m*). Εχ = 1029. B2 = 1.V)] (132) Z= Ax(l-V)-l. Es = 0.0.ν ) 2 ] /[A s (l . = 0.BiP¡) Pi = 1.0Bsgi/V {ΜΡα). Αχ = 2. D = 2. ft = 0.V)/V H{V = 1) then Ρ.1. where: V = eo/fi. + 2. The properties of the water are: • initial density: ß0 = 1000 {Kg /m 8 ). 46 .14 and: ßot0 = °·0 (AfPo m*/ms).3333. As = 2.086. ί Ι .398.4. Λ = Ps(Bx + AxPs . The properties of the ebtoplastic vessel material are: • initial density: 7900 {Kg/ms).8293. As = 0.0 ­ Bs{l .

The important increase of the charge volume. One abo notes that the automatic mesh displacement prescription al­ gorithm has performed quite well.14 gives a comparison between the EURD YN­3M and EURD YN­ 1M (2D­axbymmetric) results for the meridional profiles of the longitudinal and hoop strains. second hardening slope: 1867.93 χ IO5 (AíPo). 47 . • trilinear stress­strain law (botropic hardening): first yield stress: 275. Fig.• Young's modulus: 1. the progressive impact of water on the rigid vessel­cover and the successive deformation of the vessel itself are clearly seen on the figures.0 {MPa). A fairly good agreement between the two computations may be noted in spite of the coarse 3­D discretization employed in the circumferential direction. Fig. 3 and 4 ms.0 {MPa). second yield stress: 354. 13 shows the deformed configurations of the computational mesh at times 1. first hardening slope: 5895.0 {MPa).0 {MPa). 2.

Figure 6: Spherical cap problem. 48 .

Figure 7: Finite element mesh for spherical cap problem. 49 .

(d) t«7«10"4i (e) t*8«10"< . 50 .Pigure 8: Spherical cap computation. Deformed mesh at various times (displacements magnification factor «10): (a) t«1«1o"< . (c) t*5«10"'6 . (b) t*3«10"* .

51 .000 o» (¡L χ Γ Ώ /Β Ώ S ρ BB Ρ y& fe.-6.5. L _ ι co 00 TíMe <χ10 _4 > Figure 9: Spherical cap computation./ I Ο) I 1 Γ .000 3.000 2. Work done by external pressure (a) and apex deflection (b) versus time.000 I -. w V. EEURD Y1Ï­1: 20 conical s h e l l elements. 3>EURD YN­2: 20 9­node elements. jj.000 - .000 ι 1 1 O ^ 4.000 JÉ - L Ο 1.000 ^^«> ^sffi^^QK Α Π V> -8.000 - (b) Q o c Λ é £-1.000 ¿ ι \ -2. " V. —BURD YN­3M: 20 shell elements. Ψ α> τ> φ Β S Ώ % «Ρ.

52 .ι V///////////////M\ Rigid cover Thin vessel Figure 10: Thin cylindrical vessel problem.

000 •Ν r ^ -*^^ >^ \ -—" ^ υ t.500 ΙΛ Χ -··* I « c 1. 53 .000 .500 vt nrm g o OisTance ( C H > <x10 '> Figure 11: Vessel thickness versus dbtance from rigid cover.UUU •N 1 1 -r- 1 *i - o 2.J. u .500 ν I "κ 2.

Figure 12: F i n i t e element mesh f o r thin cylindrical vessel problem 54 .

(c) t*3 me.♦Ä ^^ * 55 . (b) t=2 ms. (d) t=4 ms.Figure 13: Thin cylindrical vessel problem. Deformed mesh at various times: (a) t=l ms. „.

56 .D Q D . h — □/ . Η Π Ο Β Β Ε Π π η « .soo - ° D - . Meridional p r o f i l e s of longitudinal (a) and hoop strain (b): — BÜRD YN­3M.200 ■ ƒ B\ °\ —U . ^ " . τ ') . 1 -ι (b) .800 . f.δ ο δ ο S o Cover disTcxnce ( C M > 1. p.000 ' ■ δ ο CM UI Cover disTunce <CM) <X10 ' ) Figure 14: Thin cylindrical v e s s e l problem.000 7 o *S T CΧ1 0 „ . SD RMN­1ÍÍ (axisymmetri β ) .400 ö i.

Lagrangian-Eulerian finite element formulation for inconpressible viscous flows. Theory and structure of the AFTON codes. Meeting on Fast Reactor Safety and Related Physics. Press. New York. 17-38. pp. Advanced Structural Dynamics.J. Ithaca. p. J. 1964. Donea. B. Kennedy.A. coupled Eulerian-Lagrangian code. 227-253. Zimmermann. 1980. Cook. Des. Fasoli-Stella and S. Liu and T. 117. J. [6] C. Cornell University. Fasoli-Stella and Α. US-Japan Seminar on Interdisciplinary Finite Elemet Analysb. 49. Belytschko and J.References [l] P. Survey of computational methods for transient fluid-structure problems in reactor safety. Proc. Chicago. NY. Finite element solution of transientfluid-structureproblems in Lagrangian coordinates. 1975.Applied Science Publishers. 14. CEL : A time-dependent. [3] J. Belytschko. 172-177. Noh.G.W.1966. Vol. An arbitrary LagrangianEulerian computing method for all flow speeds. J. Computer models for subassembly simulation.F.M. Engrg. Kennedy. Nucl. W. ΠΙ. 191-253. Hirt. Comput. eds. P. Donea ed. two-space dimensional. 1976. Finite element approach to pressure wave attenuation by reactor fuel subassemblies. Methods in Computational Physics. Kirtland Air Force Base Rept. Air Force Weapons Laboratory. J. [4] W.K.F. [7] T. [8] T. pp. 1427-1435.M. pp. AFWL-TR66-19.1978. A. 1974. Schoeberle. Adler et al. Academic Press. [5] J. Belytschko and J. Quasi-Eulerian finite element formulation for fluid-structure interaction. No. Hughes.M.. [2] T. Jones. Trulio. [9] T.. 1979. Phys. Amsden and J..V. Internat. Kennedy and D.L.K. Giuliani. pp. pp. Technology. paper 7857 .R.

[17] J. Fluid -Structure Interaction in Problems with Interfaces Involving Sharp Corners.PVP-60. 1979. Donea. [11] J.Halleux and A. Vol. Donea. Donea. Giuliani. 1978. Fasoli-Stella. An Algorithm for Continuous Rezoning of the Hydrodynamic Grid in Arbitrary Lagrangian-Eulerian Computer Codes. 255-290. J. The computer code EURDYN-1M for transient dynamic fluid-structure interaction. Montreal. paper B l / 3 . Giuliani. An arbitrary Lagrangian-Eulerian finite element procedure for transient dynamic fluid-structure interaction problems. pp. Berlin.V. paper B l / 2 . Joint ASME/CSME Pressure Vesseb and Piping Conference. The classical field theories. Vol. EUR 6751. Halleux. 4th Internat. San Francbco. P. S. [14] J. Giuliani. P. HI 1. [12] C. on Structural Mechanics in Reactor Technology. 72. 5th Internat. 1977. Canada. An Arbitrary LagrangianEulerian Finite Element Method for Transient Dynamic FluidStructure Interactions. [13] J. Giuliani. Commission of the European Communities. Jones. S. Fasoli-Stella and S. Donea and S. Applied Science Publishers.1982. Trans. Computer Methods in Applied Mechanics and Engineering. 205-212. 1980. Conf. 689-723. Nuclear Engineering and Design. [16] S. 33. Trans. 1980. Vol. Fasoli-Stella.P. Springer. J. pp. Advanced Structural Dynamics. paper B l / 3 . [10] J. [15] J. on Structural Mechanics in Reactor Technology. Giuliani and J. 6th International Conference on SMIRT.. Jones. S. Encyclopedia of Physics.V.P. P. Berlin. Donea. 58 . 1981. Donea ed. Donea. Lagrangian and Eulerian finite element techniques for transient fluid-structure interaction problems. pp.1982.Halleux and A. Toupin. Paris. Conf. 1960. Giuliani. Finite element analysis of transient dynamic fluid-structure interaction. 17-21 August. J. Truesdell and R.P.

1981. Comp. August 12-16. Comp. The Norwegian Institute of Technology. [22] S. paper E2/1. [23] M. Norway. Nonlinear Finite Element Analysis of Shells: Part I . Explicit Time Integration of Structure-Mechanical Systems. Donea ed.Three-dimensional Shells. Applied Science Publishers. 59 . [26] T. [21] T. Recent Improvements of the Non-linear Transient Dynamic Structural Computer Programs EURDYN. Cronshagen and P. Donea. A Uniform Strain Hexahedron and Quadrilateral with Orthogonal Hourglass Control.R. Liu. [25] T. Meth. North-Holland.E. Eng. 4th International Conference on SMIRT. Appi.1981..W.P. EUR 6694. San Francbco. Appi. Halleux. pp. Liu. UCRL51574 Rev. 1985. [19] I. Trondheim. pp. Flanagan and T. R. 331-362. 1980. Chapter 10 in Computational Methods for Transient Analysis. Commission of the European Communities. Giuliani and J. S. 26. The University of Trondheim. Nonlinear Finite Element Analysis of Shells: Part II .1981.[18] J. 1980.P. Volume 2. Key.E. Hoskin and M.. Hughes editors.J.G.J. Charge Development and Analysis of Recent UK Experiments in the COVA Model Test Programme.K. Grantham. Hughes and W. May 30. 1. Mech. International Journal for Numerical Methods in Engineering. 167-181. Advanced Structural Dynamics. Meth. Lancefield. Blum.. T. USA. E. N.R. London. Belytschko. J. Lawrence Livermore Laboratory.Wilkins. 27.J. Europe-Us Symposium.Two-dimensional Shells. Belytschko. Belytschko and T. pp.1975. 17. Donea. Improvement in Transient Dynamic Time Integration with Application to Spent Nuclear Fuel Shipping Cask Impact Analyses. Arbitrary Lagrangian-Eulerian Finite Element Methods. Mech. 1977.L.J. 15-19 August. 1983. [20] J.R. Hughes and W. Cameron. [24] D. Eng. 679-706.K. Vol.

Eng. pp. 1981.R.J.K. Wunderlich et al. 60 . 39. Nonlinear Finite Element Shell formulation Accounting for Large Membrane Strains.1983. ed. Nonlinear Dynamic Finite Element Analysis of Shells.. Meth. Levit. Hughes and E. Hughes. Springer.R. Mech. Comp. 69-82. Carnoy. [29] F. Casadei.. EUR report (in preparation). Liu and I. Commission of the European Communities. Berlin. A Nonlinear S-D Shell Element Implementation for Transient Problems. Appi.J. W. [28] T. In "Nonlinear Finite Element Analysis in Structural Mechanics". W.[27] T.