You are on page 1of 8

ASEN 6367 Advanced FEM for Solids, Plates and Shells. Fall 2011.

Midterm 1
Important: Work individually. No consultation or collaboration with others is permitted. Take Home, assigned Th Feb 24, 2011, due Th Mar 3, 2011 at class time (may be submitted earlier at AE 187 or instructor mailbox at AE 196 or OT 635) Begin each QUESTION on a separate page. Write your name on each page. Attach this exam as cover. Attach this exam as cover. Use engineering paper. QUESTION 1. Conceptual: 30 pts=5 pts per item. Be brief. References to answers in course Notes or in texbooks are OK. For items (ac), y(x) is an ordinary function of x whereas [y] is a functional of y(x). (a) Explain the differences between the ordinary differential dy, the variation y, and the variation . (b) Why is x = 0? (c) Why are variational forms (if they exist) desirable for numeric computation over weak forms? (d) How many canonical variational principles of linear elasticity exist? (Classical ones only, exclude hybrids) Which ones have been (so far) proven useful for FEM work? (e) Why do we bother to use the more complicated mixed or hybrid variational principles beyond our friend TPE to derive nite elements? (f) What two numerical problems must one must watch out for when doing nite element analysis of axisymmetric solids? QUESTION 2. Hybrid element development + application to design: 45 pts=10+15+10+10. Consider the plane, prismatic, hinged beam member shown in Figure Q1.1(a). Its mechanical behavior away from the hinged section is idealized by the Bernoulli-Euler (BE) model, also called engineering beam theory. This model only considers internal energy due to bending, ignoring contribution of the transverse shear. The Cartesian axes {x, y, z} are chosen as shown in that Figure. A prime will denote differentiation wrt x, so (.) = d(.)/d x. Except for the hinged section, the elastic modulus E and the second moment of inertia Izz = A y 2 d A are constant along x. The member is acted upon by a distributed lateral load q(x), expressed as force per unit length. The lateral displacement, positive along +y, is denoted by v(x). The cross section rotation about its neutral axis is, according to the BE model, (x) = dv(x)/d x = v (x), positive CCW. The internal forces include the bending moment M(x) and the transverse shear force V (x) = d M(x)/d x = M (x), with positive sign conventions illustrated in Figure Q1.1(b). Upon breaking up the member into nite elements, one encounters two element types: hinged and non-hinged. Non-hinged, 2-node elements are derived using the conventional Total Potential Energy (TPE) variational principle and well known cubic shape functions in Chapter 12 of the IFEM Notes: its stiffness matrix and consistent force vector are presented in 12.6 of that Chapter. The subject of this Question is the construction of a hinged, 2-node beam element according to the Equilibrium Stress Hybrid (ESH) variational principle introduced in Chapter 8 of the AFEM Notes. For this conguration, the ESH approach leads to the element stiffness equations faster than going through TPE, because it bypasses the error-prone use of complicated (piecewise cubic) shape functions that must allow akink (slope discontinuity) at the hinge section. Q11

Notation for the hinged element is introduced pictorially in Figures Q1.1(cf) and in Figure Q1.2. Items: (a) The ESH principle for the Bernoulli-Euler beam model is most readily developed starting from the classical Hellinger-Reissner (HR) principle. For an individual element such as shown in Figures Q1.1(c,d,e):
e HR

=
0

Le

M2 e e e e + M v qv d x m e 1 + m e 2 + f 1e v1 + f 2e v2 , 1 2 2E Izz

(Q1.1)

in which the master elds are M = M(x) and v = v(x). (The e superscripts in M and v are omitted to reduce clutter.) The end forces and moments f 1e , f 2e , m e and m e , dened in 1 2 Figure Q1.2(b), are prescribed.* Take (Q1.1) as given. For this item, show that if the bending moment M(x) satises strongly the equilibrium (balance) equation so that M = q, (Q1.1) can be converted, upon double integration by parts of the Mv term, to the ESH functional
e ESH e UE S H e e = UE S H WE S H , Le

e WE S H

M2 dx x=L e x=L e + M v x=0 M v x=0 2E Izz 0 Le M2 dx e e e e e e = M1 1 + M2 2 + V1e v1 + V2e v2 , 2E Izz 0 e e e e e = m 1 1 + m e 2 + f 1e v1 + f 2e v2 , 2 =

(Q1.2)

e in which the second form of U E S H results from the nodal value substitutions e e M1 = M(0), M2 = M(L e ), V1e = M (0), V2e = M (L e ), e e e e v1 = v(0), v2 = v(L e ), 1 = v (0), 2 = v (L e ).

(Q1.3)

This is the ESH principle to be used in the next 2 items for element derivation.** (b) The two master elds to be varied in (Q1.3) are: 1. The bending moment M e (x) over the element. 2. The end displacements and rotations. These can be identied with the nodal freedoms e e e e v1 = v(0), v2 = v(L e ), 1 = v (0) and 2 = v (L e ) pictured in Figure Q1.2(a). To simplify the moment assumption, take q(x) 0 over the element; thus lateral forces are only applied at end nodes. Consequently the moment variation can be linear in x. (Explain briey why.) Since M e (x) must vanish at the hinge, take M e (x) = M e ( ) = e Mr e f ( H )
def

(Q1.4)

Here = (2x1)/L e is the natural (isoparametric) coordinate dened in Figure Q1.1(f), H the natural coordinate of the hinge, e a stress parameter and Mr e f a reference moment
e * Do not confuse m e with M1 = M(0), say. The latter is an internal eld evaluated at node 1, not a 1 prescribed value.

** If you cant derive (Q1.2), dont stop. Assume it and continue with (b,c,d).

Q12

(a) 3D view of prismatic, hinged, plane beam member

q(x) z
z along neutral axis of cross section

(d) Element lateral deflection ve(x) sketch and nodal DOF (grossly exaggerated for visibility)

E & Izz constant along member

y,v
e

2 v1
e

hinge

1
x y
x along cross section centroid beam top surface (+y)

ve(x)

hinge

e v2

1
Scaled slope of M(x) diagram is stress parameter e

2
e M2 2

(b) Positive sense of beam internal forces: bending moment M(x) and transverse shear force V(x)

1 M1
e

M (x)

M(x) x V(x)
cross section at distance x

hinge

z
(e) Element bending moment diagram sketch if q(x) = 0 over element: M e(x) is positive if it compresses the top surface, as pictured in (b)

y 1

q(x)

hinge

x
L H1
e

LeH2 L
e

1
=1 = H
(f) Natural coordinate

2
=1

(c) 2-node, prismatic, hinged, plane beam Bernoulli-Euler element on x-y plane

Figure Q1.1. 2-noded, plane, prismatic, hinged, Bernoulli-Euler-modeled beam element to be formulated by a ESH variational principle.
(a) Element-level Degrees of Freedom (DOF) for ESH hinged beam element. Positive senses as shown. (b) Forcing actions conjugate to DOF pictured in (a) for ESH hinged beam element. Positive senses as shown. Action conjugate to stress parameter e is zero and thus omitted.
e v2 e f2 e 2 e m1

e 1

e v1

f1e

e m2

Stress parameter e can be interpreted as the scaled slope of the M(x) diagram over element -- see Figure Q1.1(e) -but cannot be readily visualized as a local physical quantity

Figure Q1.2. Degrees of freedoms and conjugate actions for hinged beam element of Figure Q1.1.

Q13

introduced only to make e dimensionless. (This Mr e f will disappear upon condensation.) The interpretation of e as a scaled slope of M e (x) is noted in Figure Q1.1(e). For the interface displacements we simply pick the nodal DOF as stated above. Expressing v e (x) within the element is messy because it consists of two cubic polynomial patches with a kink at the hinge, as pictured in Figure Q1.1(d). We dont need to worry about this mess, however, since the inner v e (x) has disappeared in (Q1.2). This feature makes the derivation much simpler. For convenience introduce the DOF vectors: ue = [ e +
e v1 e 1 e v2 e 2 ]T , e ue = [ v1 e 1 e v2 e 2 ]T ,

e = [ e ] ,

(Q1.5)

and their conjugates fe = [ 0 +


e v1

me 1

e v2

m e ]T , 2

fe = [ f 1e

me 1

f 2e

m e ]T . 2

(Q1.6)

Inserting (Q1.4) into the bending-energy term in (Q1.2), integrate to get


e Uc

=
0

Le

(M e (x))2 dx = 2E Izz

1 1

(M e ( ))2 J d, 2E Izz

with

J = d x/d = 1 L e , 2

(Q1.7)

as an explicit function of e . Then add the rest to express e S H as a quadratic form of the E degrees of freedom: e e e e e e e (Q1.8) ESH = E S H ( , v1 , 1 , v2 , 2 ). The ve element equations are obtained by making (Q1.8) stationary wrt to the ve DOF:
e ESH e

e ESH e v1

e ESH e 1

e ESH e v2

e ESH e 2

= 0.

(Q1.9)

These linear equations can be arranged into the partitioned matrix form F GT G 0 e ue = 0 , fe (Q1.10)

where F is a 1 1 exibility matrix, and G is a 1 4 connection matrix. As nal result for this item, show the expressions of F and G in terms of the data. (c) By looking at the variational index of M(x) in (Q1.2), explain why the stress parameter e may be statically condensed from (Q1.10). (Static condensation is covered in Chapter 10 of the IFEM Notes.) Upon doing that obtain the element stiffness equations in terms of node displacements only: Ke ue = fe , in which Ke = GT F1 G. (Q1.11)

Entries of the 4 4 symmetric stiffness matrix Ke are functions of E, Izz , L e and H . Partial e 2 answer: K 22 = 3E Izz (1 + H )2 / L e (1 + 3 H ) . The rank of this matrix must be one. For H = 0, which means that the hinge is at the element midpoint, check that you get the stiffness equations (13.14) that appear in Chapter 13 of the IFEM Notes. (d) Apply the element stiffness equations of (c) to a simple design optimization problem that can be entirely done by hand. The problem is pictured in Figure Q1.3(a). A xed-xed Q14

prismatic beam AB of total span L is loaded by a point force P at its center C. Two hinges are symmetrically placed at distances 1 L from the end supports. Because of the symmetry about 2 C, it is possible to model the problem with just one hinged beam element, say the left half-span, as illustrated in Figure Q1.3(b). The displacement support conditions are: v1 = 1 = 2 = 0. This leaves only one unknown DOF: the midspan deection v2 = vC under P. Question: for which hinge conguration, dened by as design variable, is vC minimized in magnitude? Mathematically this is a constrained optimization problem: nd min |v ()| for [0, 1] Why is this conguration of interest to engineers? Minimizing vC for xed P means that the external energy V = P vC is minimized. Hence the effective rigidity under that loading is maximized. In structural optimization this is called a maximum stiffness design, which is of interest for certain classes of structures.
(a) Centrally-point-loaded, fixed-fixed, plane beam with 2 symmetrically placed hinges
C

y,v

Symmetry line

E & I zz constant along beam

hinge

P
C

hinge

= design variable

(b) One-element discretization by taking advantage of symmetry

BC: v 1 = 1 = 0

Does a minimum exist? For both = 0 (a simply supported beam) and = 1 (a doubly hinged cantilever), Mechanics of Material textbooks give vC = P L 3 /(48E Izz ), which provide a double check on your computations. For in between, |vC | gets smaller, so there is in fact an optimal value, which is readily computable by hand. For the beam element shown in Figure Q1.3(b), = 2 1 and = (1 + )/2. H H

;;
A

;;

x
L 2 L 2
1 L 2

1 L 2

L L/2 L 2 2

hinge

P/2
2

;;
1

;;

(1)

BC: 2 = 0

Figure Q1.3. Application of ESH-formulated hinged beam element to a simple design optimization problem.

Q15

(a) Parent element: Quad8 Ring z 4 8 1 d 7 r 5 2a 6 2


Quad8 generating cross section

(b) Child element: Shell2 Ring


Shell2 generating cross section

3 2h

morphing

i
2a

j 2h

(c) Visualization of the DOF of the Shell2 ring element uzj uri uzi zi urj

zj

Note: nodes i and j of Shell2 and nodes 8 and 6 of Quad8, respectively, are exactly at the same positions with respect to axes r and z. Their generating cross sections are drawn offset for visualization convenience

Figure Q1.4. Morphing ring elements for Question 3: 8-node quadrilateral to 2-node thin shell.

QUESTION 3. Analytical + Some thinking. 25 pts=15+10. The subject of this Question (never assigned previously) is a nite element fabrication method called morphing. The word generically means to undergo transformation (Merriam-Webster dictionary; From Gr. morphos: shape.) In the nite element area, the term identies the mapping of a parent element to a child element (also called source and target, respectively). through multifreedom constraints or MFC. The child element has a smaller number of degrees of freedom and often a smaller dimensionality. The process is described more carefully in Chapter 21 of the Notes, which is half done. Hopefully a good solution to this Question can be used to expand it. Candidate parent elements are usually selected from those that exhibit good performance above and beyond the call of duty, with the hope that those good genes are passed to the child. Here we chose the reduced-integration axisymmetric 8-node quadrilateral ring element introduced in Chapter 11 as parent. The child is a two-node ring element intended to model thin axisymmetric shells: think of a thin beam being rotated 360 around z. We will use the abbreviations Quad8 and Shell2 for them, respectively, being understood that both are ring elements. The process is illustrated in Figure Q1.4. For simplicity the element geometries are restricted as follows.
The term is nowadays used most often in image processing: gradually mapping in pseudo-time the image of an object onto that of another one by means of computer software, a process sometimes endowed with animation. Think of the second Terminator movie.

Q16

Quad8. Generating cross section is a 2a 2h rectangle, with sides parallel to the {r, z} axes, as pictured in Figure Q1.4(a). The r axis is made to pass through midside nodes 6 and 8 for convenience. Shell2. Generating cross section is a 2a 2h rectangle, with sides parallel to the r, z axes, as pictured in Figure Q1.4(b). Here 2h is the shell thickness dimension. Typically h << a for a thin shell element, although that is not apparent in the Figure because the thickness dimension is exaggerated for clarity. The generating cross sections overlap in the sense that end nodes i and j of Shell2 occupy the same location of midside nodes 8 and 6, respectively, of Quad8. The cross sections are drawn offset in Figure Q1.4(a,b) for visualization convenience. Both elements are considered completely free, that is, not connected to other elements or given boundary condtions. Material properties are not relevant to morphing. The Quad8 element has 8 nodes (labeled 1, 2, . . . 8) and 16 displacement degrees of freedom arranged as (Q1.12) ue = [ u r 1 u z1 u r 2 u z2 u r 3 u z3 . . . u r 8 u z8 ]T . Q where u r 1 , for example, is the radial displacement of node 1. The Shell2 element has 2 nodes (labeled i and j) and 6 displacement degrees of freedom arranged as ue = [ u ri S u zi i ur j uzj j ]T . (Q1.13)

where denotes the end node rotations depicted in Figure Q1.4(c), positive CCW. (The more common rotation symbol: , is not used because of clash with the circumferential coordinate.) To carry out the morphing process, the kinematics of Quad8 and Shell are related through a transformation matrix that connects the freedoms: u e = T Q S ue , Q S (Q1.14)

in which T Q S is an 16 6 matrix. Let Ke and fe denote the stiffness matrix and consistent force Q Q vector, respectively, of the parent element Quad8. The corresponsing entities for the child element Shell2 are obtained by a congruential transformation: Ke = TT S Ke T Q S , S Q Q fe = TT S fe . S Q Q (Q1.15)

So the main idea is: once T Q S is built, we get two elements for the development price of one. The main use of morphing is the development of reduced order model (ROM) elements. Those are special elements that conceal multiscale parent details, such as holes, inclusions, cracks or composite fabrication. But any of those would take far too long for a midterm exam (might be OK for a term project), so Question 3 is about an parent-child pair that can be done quickly. Fast development of custom ROMs is important for commercial software houses, since formulating and testing a new nite element from scratch may incur high labor expenses (good programmers cost now about $300K per year). Savings can be substantial if (and that is a big if) the child behaves. To build T Q S one needs to know the displacement eld of the child element. For Shell2, it is given by the interpolation of the bar and (Bernoulli-Euler) beam elements given in Chapter 11 and 12, respectively, of the IFEM Notes. Those are reproduced below for convenience and relabeled for {r, z} physical coordinates. Let {, } denote the usual quadrilateral natural coordinates dened Q17

over generating cross sections. For the rectangular generating cross sections of Figure Q1.4, they are linked to r and z by r = d + a , z = b , = r d , a = z , b 1 = , r a 1 = , z b = = 0. z r (Q1.16)

Introduce the shape functions N1 = 1 (1 ), 2 N2 = 1 (1 + ), 2 N3 = 1 (1 )2 (2 + ), 4 N6 = 1 (2a)(1 + )2 (1 ). 8 (Q1.17) u z b u z = , r a

N4 = 1 (2a)(1 )2 (1 + ), 8 Then for Shell2:


bar bend u r (, ) = u r + u r ,

N5 = 1 (1 + )2 (2 ), 4

bar u r = u ri N1 + u r j N2 ,

bend ur = z

(Q1.18)

u z (, ) = u zi N3 + i N4 + u z j N5 + j N6 . in which as usual and vary within the range [1, +1]. Items in Question: (a) Construct the T Q S matrix. There are actually several ways to do this. By far the simplest is collocation: the displacements of Shell2 are evaluated at the nodes of Quad8 and used to ll out T Q S . Any other ideas welcome. Whatever you come up with, please indicate your method. (b) Having build a T Q S you now have a possible shell ring element Shell2. How would you test whether this element behaves? (Dont actually do it, just write down some ideas for that; reading the (incomplete) material in Chapter 21 may suggest ways and means.)

Q18

You might also like