.

The Direct Stiness Method III

15

151

Lecture 15: THE DIRECT STIFFNESS METHOD III

152

TABLE OF CONTENTS
Page

15.1. The Remaining DSM Steps 15.2. Assembly: Merge 15.2.1. Governing Rules . . . . . . . . . 15.2.2. Hand Assembly by Augmentation and Merge 15.3. Solution 15.3.1. Applying Displacement BCs by Reduction . 15.3.2. Solving for Displacements . . . . . 15.4. PostProcessing 15.4.1. Recovery of Reaction Forces . . . . . 15.4.2. Recovery of Internal Forces and Stresses .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

153 153 153 154 156 156 157 157 157 158

The following material, as well as that presented in Lectures 14 and 15, is selectively extracted from Chapters 2 and 3 of the rst-year core graduate course ASEN 5007: Introduction to Finite Element Methods, which is taught every Fall. The ASEN 5007 course material is posted at http://www.colorado.edu/engineering/CAS/courses.d/IFEM.d/Home.html

152

153 15.1. The Remaining DSM Steps

15.2

ASSEMBLY: MERGE

Lectures 13 and 14 cover the initial stages of the DSM. The three breakdown steps: disconnection, localization and formation of member stiffness take us down all the way to the generic truss element: the highest level of fragmentation. This is followed by the assembly process. Assembly involves merging the stiffness equations of each member into the global stiffness equations. For this to make sense, the member equations must be referred to a common coordinate system, which for a plane truss is the global Cartesian system {x , y }. This is done through the globalization process covered in the previous Lecture. On the computer the formation, globalization and merge steps are done concurrently, member by member. After all members are processed we have the free-free master stiffness equations. Next comes the solution. This process embodies two substeps: application of boundary conditions and solution for the unknown joint displacements. To apply the BCs, the free-free master stiffness equations are modied by taking into account which components of the joint displacements and forces are given and which are unknown. The modied equations are submitted to a linear equation solver, which returns the unknown joint (node) displacements. The solution step completes the DSM proper. Postprocessing steps may follow, in which derived quantities such as internal forces and stresses are recovered from the displacement solution. F 15.2. Assembly: Merge 15.2.1. Governing Rules The key operation of the assembly process is the placement of the contribution of each member to the master stiffness equations. The process is technically called merge of individual members. The merge operation can be physically interpreted as reconnecting that member in the process of fabricating the complete structure. For a truss structure, reconnection means inserting the pins back into the joints. See Figure 15.1. Merge logic is mathematically governed by two rules of structural mechanics:
(3) y 1 x (1) 2 (2) 3

Figure 15.1. The disconnected example truss prior to merge. All member stiffness equations are in the global system. Reconnecting the truss means putting the pins back into the joints.

1. 2.

Compatibility of displacements: The displacement of all members meeting at a joint are the same. Force equilibrium: The sum of forces exerted by all members that meet at a joint balances the external force applied to that joint. (15.1)

153

Lecture 15: THE DIRECT STIFFNESS METHOD III

154

(a)
3) f( 3

2) f( 3

f3

(b)
3) f( 3

3
2) f( 3

(3)

(2)

Figure 15.2. The force equilibrium of joint 3 of the example truss, depicted as a free body diagram in 2) (3) (a). Here f3 is the known external joint force applied on the joint. Joint forces f( 3 and f3 are applied by the joint on the members, as illustrated in (b). Consequently the forces applied by the members on the 2) (3) joint are f( 3 and f3 . These forces would act in the directions shown in (a) if both members (2) and 2) (3) (2) (3) (3) were in tension. The free-body equilibrium statement is f3 f( 3 f3 = 0 or f3 = f3 + f3 . This (2) (3) (2) (3) translates into the two component equations: f x 3 = f x 3 + f x 3 and f y 3 = f y 3 + f y 3 , of (15.2).

The rst rule is physically obvious: reconnected joints must move as one entity. The second one can be visualized by considering a joint as a free body, although care is required in the interpretation of joint forces and their signs. Notational conventions to this effect are explained in Figure 15.2 for joint 3 of the example truss, at which members (2) and (3) meet. Application of the foregoing rules at this particular joint gives Rule 1: Rule 2:
2) (3) u( x3 = ux3 , 2) (3) u( y3 = u y3 .

2) 3) 1) 2) 3) f y 3 = f y(3 + f y(3 = f y(3 + f y(3 + f y(3 . (15.2) 1) 2) 3) 1) 2) 3) to f x(3 + f x(3 and of and f y(3 to f y(3 + f y(3 , respectively, changes nothing The addition of f x(3 because member (1) is not connected to joint 3. We are just adding zeros. But this augmentation enables us to write the key matrix relation:

2) 3) 1) 2) 3) f x 3 = f x(3 + f x(3 = f x(3 + f x(3 + f x(3 ,

f = f(1) + f(2) + f(3) . 15.2.2. Hand Assembly by Augmentation and Merge

(15.3)

To directly visualize how the two rules (15.1) translate to merging logic, we rst augment the member stiffness relations by adding zero rows and columns as appropriate to complete the force and displacement vectors. For member (1):
1) f x(1 10 (1) f y1 0 1) f x(2 10 1) = 0 f y(2 (1) 0 fx3 0 1) f y(3

0 0 0 0 0 0

10 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 0 154

1) u( x 1 0 1) u( 0 y1 (1) 0 ux2 1) . 0 u( y2 0 u (1) x3 0 1) u( y3

(15.4)

155 For member (2): For member (3):

15.2 (2) ux1 0 (2) 0 u y1 (2) 0 ux2 2) . 5 u ( y2 0 u (2) x3 5 (2) u y3

ASSEMBLY: MERGE

0 0 0 0 0 0

0 0 0 0 0 0

0 0 0 5 0 5

0 0 0 0 0 0

(15.5)

10 10 0 0 10 10

0 0 0 0 0 0 0 0 0 0 0 0

10 10 0 0 10 10

(3) ux1 10 (3) 10 u y 1 (3) 0 ux2 3) . 0 u( y2 10 u (3) x3 10 3) u( y3

(15.6)

According to the rst rule, we can drop the member identier in the displacement vectors that appear in the foregoing matrix equations. Hence the reconnected member equations are (1) fx1 10 0 10 0 0 0 ux1 ( 1 ) f y1 0 0 0 0 0 u y1 0 f (1) x 2 10 0 10 0 0 0 u x 2 (15.7) , (1) = 0 0 0 0 0 u y2 f y2 0 (1) ux3 0 0 0 0 0 0 f x3 u y3 0 0 0 0 0 0 1) f y(3 (2) fx1 0 0 0 0 0 0 ux1 ( 2 ) f y1 0 0 0 0 0 0 u y1 f (2) x2 0 0 0 0 0 0 ux2 (15.8) , (2) = f y 2 0 0 0 5 0 5 u y 2 (2) ux3 0 0 0 0 0 0 f x3 u y3 0 0 0 5 0 5 2) f y(3 (3) fx1 10 10 0 0 10 10 ux1 ( 3 ) f y1 10 0 0 10 10 u y 1 10 f (3) 0 0 0 0 0 ux2 x2 0 (15.9) . (3) = 0 0 0 0 0 u y2 f y2 0 (3) ux3 10 10 0 0 10 10 f x3 u y3 10 10 0 0 10 10 3) f y(3 These three equations can be represented in direct matrix notation as f(1) = K(1) u, f(2) = K(2) u, 155 f(3) = K(3) u. (15.10)

Lecture 15: THE DIRECT STIFFNESS METHOD III

156

According to the second rule, expressed in matrix form as (15.3), we have f = f(1) + f(2) + f(3) = K(1) + K(2) + K(3) u = K u, (15.11)

so all we have to do is add the three stiffness matrices that appear above, and we arrive at the master stiffness equations: fx1 20 f y 1 10 f x 2 10 = f y2 0 fx3 10 f y3 10 10 10 0 0 10 10 10 0 10 0 0 0 0 10 0 10 0 0 5 0 0 10 5 10 10 ux1 10 u y 1 0 ux2 . 5 u y 2 ux3 10 u y3 15

(15.12)

Using this technique member merging becomes simply matrix addition. This explanation of the assembly process is conceptually the easiest to follow and understand. It is virtually foolproof for hand computations. However, this is not the way the process is carried out on the computer because it would be enormously wasteful of storage for large systems. A computer-oriented procedure is discussed in 15.5. 15.3. Solution Having formed the master stiffness equations we can proceed to the solution phase. To prepare the equations for a linear solver we need to separate known and unknown components of f and u. In this Section a technique suitable for hand computation is described. 15.3.1. Applying Displacement BCs by Reduction If one attempts to solve the system (15.12) numerically for the displacements, surprise! The solution blows up because the coefcient matrix (the master stiffness matrix) is singular. The mathematical interpretation of this behavior is that rows and columns of K are linear combinations of each other (see Remark 15.1 below). The physical interpretation of singularity is that there are unsuppressed rigid body motions: the truss still oats in the {x , y } plane. To eliminate rigid body motions and render the system nonsingular we must apply the physical support conditions as displacement boundary conditions. From Figure 14.4(b) we observe that the support conditions for the example truss are u x 1 = u y 1 = u y 2 = 0, whereas the known applied forces are f x 2 = 0, f x 3 = 2, f y 3 = 1. (15.14) (15.13)

When solving the overall stiffness equations by hand, the simplest way to account for support conditions is to remove equations associated with known joint displacements from the master system. To apply (15.13) we have to remove equations 1, 2 and 4. This can be systematically 156

157

15.4

POSTPROCESSING

accomplished by deleting or striking out rows and columns number 1, 2 and 4 from K and the corresponding components from f and u. The reduced three-equation system is 10 0 0 0 10 10 0 10 15 ux2 ux3 u y3 = fx2 fx3 f y3 = 0 2 . 1 (15.15)

Equation (15.15) is called the reduced master stiffness system. The coefcient matrix of this system is no longer singular.
r = 3 and a rank deciency of d = N r = 6 3 = 3 (these concepts are summarized in Appendix C.) The dimension of the null space of K is d = 3. This space is spanned by three independent rigid body motions: the two rigid translations along x and y and the rigid rotation about z .
Remark 15.2. Conditions (15.13) represent the simplest type of support conditions, namely specied zero Remark 15.1. In mathematical terms, the free-free master stiffness matrix K in (15.12) has order N = 6, rank

displacements. More general constraint forms, such as prescribed nonzero displacements and multifreedom constraints, are handled as described in 15.6 and Chapters 89, respectively.

15.3.2. Solving for Displacements Solving the reduced system by hand (for example, via Gauss elimination) yields ux2 ux3 u y3 = 0 0.4 . 0.2 (15.16)

This is called a partial displacement solution (also reduced displacement solution) because it excludes known displacement components. This solution vector is expanded to six components by including the three specied values (15.13) in the appropiate slots: ux1 0 u y1 0 u 0 u = x2 = . u y2 0 0.4 ux3 0.2 u y3 This is the complete displacement solution, or simply the displacement solution. 15.4. PostProcessing The last processing step of the DSM is the solution for joint displacements. But often the analyst needs information on other mechanical quantities; for example the reaction forces at the supports, or the internal member forces. Such quantities are said to be derived because they are recovered from the displacement solution. The recovery of derived quantities is part of the so-called postprocessing steps of the DSM. Two such steps are described below. 157

(15.17)

Lecture 15: THE DIRECT STIFFNESS METHOD III

158

15.4.1. Recovery of Reaction Forces Premultiplying the complete displacement solution (15.17) by K we get 20 10 10 0 10 10 0 2 10 0 0 10 10 0 2 10 0 10 0 0 0 0 0 10 f = Ku = = 0 0 5 0 5 0 1 0 10 10 0 0 10 10 0.4 2 10 10 0 5 10 15 0.2 1

(15.18)

This vector recovers the known applied forces (15.14) as can be expected. Furthermore we get three reaction forces: f x 1 = f y 1 = 2 and f y 2 = 1 that are associated with the support conditions (15.13). It is easy to check that the complete force system is in self equilibrium for the free-free structure; this is the topic of Exercise 15.1. For a deeper look at reaction recovery, study Section 4.3 of Chapter 3 of the IFEM Notes. 15.4.2. Recovery of Internal Forces and Stresses Often the structural engineer is not so much interested in displacements as in internal forces and stresses. These are in fact the most important quantities for preliminary structural design. In pinjointed trusses the only internal forces are the axial member forces. For the example truss these forces, denoted by F (1) , F (2) and F (3) , are depicted in Figure 15.3. The average axial stress e is obtained on dividing F e by the cross-sectional area of the member. 3 The axial force F e in member e can be obtained as follows. Extract the displacements of member e from the complete displacement solution u to form ue . Then e = Te ue . recover local joint displacements from u F (3) Compute the member elongation d e (relative axial displacement) and recover the axial force from the equivalent spring constitutive relation: d =
e

F (2)

u e xj u e yi

u e xi , u e yj

E A e F = d . Le
e

F (1)

(15.19)
e

Note that

and

are not needed in computing d .

Figure 15.3. Internal forces for the example truss are the member axial forces F (1) , F (2) and F (3) . Force arrow directions shown pertain to tension.

the global displacements of the member from (15.17): u(2) = [ u x 2 u y 2 u x 3 u y 3 ]T = [ 0 0 0.4 0.2 ]T . (2) = T(2) u(2) : Convert to local displacements using u

Example 15.1. Recover F (2) in example truss. Member (2) goes from node 2 to node 3 and (2) = 90 . Extract

u x2 cos 90 sin 90 0 0 ux2 0 y 2 sin 90 cos 90 0 0 u y 2 1 u = = u x3 0 0 cos 90 sin 90 u x 3 0 u y3 0 0 sin 90 cos 90 u y3 0 x3 u x2 The member elongation is d (2) = u compressive axial force.

0 0 0 0 0 0 = . 1 0.4 0.2 0 0.2 0.4 (15.20) (2) = 0.2 0 = 0.2, whence F = (50/10) (0.2) = 1, a

1 0 0 0

0 0 0 1

158

159

15.4

POSTPROCESSING

strain, e = E e ee as (average) axial stress, and F e = Ae e as the axial force. This is more in tune with the Theory of Elasticity viewpoint discussed in a previous Lecture.

Remark 15.3. An alternative interpretation of (15.19) is to regard ee = d e / L e as the (average) member axial

This concludes the exposition of the DSM for a plane truss structure. The same steps, however, apply to any structure. This universality has facilitate the development of general purpose FEM codes. One of the most widely used commercial codes in industry, called ANSYS (for ANalysis SYStem) will be used for the computer simulation portion of Lab #2, to analyze the 16-bay, three-dimensional truss structure tested at ITLL.

159