18.

FAST NONLINEAR ANALYSIS

The Dynamic Analysis of a Structure with a Small Number of
Nonlinear Elements is Almost as Fast as a Linear Analysis

18.1 INTRODUCTION

{ XE "Fast Nonlinear Analysis Method" }{ XE "FNA Method" }The response of

real structures when subjected to a large dynamic input often involves significant
nonlinear behavior. In general, nonlinear behavior includes the effects of large
displacements and/or nonlinear material properties.

{ XE "P-Delta Effects" }The use of geometric stiffness and P-Delta analyses, as

summarized in Chapter 11, includes the effects of first order large displacements.
If the axial forces in the members remain relatively constant during the
application of lateral dynamic displacements, many structures can be solved
directly without iteration.

{ XE "Large Strains" }The more complicated problem associated with large

displacements, which cause large strains in all members of the structure, requires
a tremendous amount of computational effort and computer time to obtain a
solution. Fortunately, large strains very seldom occur in typical civil engineering
structures made from steel and concrete materials. Therefore, the solution
methods associated with the large strain problem will not be discussed in detail in
this chapter. However, certain types of large strains, such as those in rubber base
isolators and gap elements, can be treated as a lumped nonlinear element using
the Fast Nonlinear Analysis (FNA) method presented in this chapter.
18-2 STATIC AND DYNAMIC ANALYSIS

{ XE "Nonlinear Stress-Strain" }The more common type of nonlinear behavior is

when the material stress-strain, or force-deformation, relationship is nonlinear.
This is because of the modern design philosophy that “a well-designed structure
should have a limited number of members which require ductility and that the
failure mechanism be clearly defined.” Such an approach minimizes the cost of
repair after a major earthquake.

18.2 STRUCTURES WITH A LIMITED NUMBER OF NONLINEAR

ELEMENTS

{ XE "Buckling Analysis" }A large number of very practical structures have a

limited number of points or members in which nonlinear behavior takes place
when subjected to static or dynamic loading. Local buckling of diagonals,
uplifting at the foundation, contact between different parts of the structures and
yielding of a few elements are examples of structures with local nonlinear
behavior. For dynamic loads, it is becoming common practice to add
concentrated damping, base isolation and other energy dissipation elements.
Figure 18.1 illustrates typical nonlinear problems. In many cases, those nonlinear
elements are easily identified. For other structures, an initial elastic analysis is
required to identify the nonlinear areas.

In this chapter the FNA method is applied to both the static and dynamic analysis
of linear or nonlinear structural systems. A limited number of predefined
nonlinear elements are assumed to exist. Stiffness and mass orthogonal Load
Dependent Ritz Vectors of the elastic structural system are used to reduce the
size of the nonlinear system to be solved. The forces in the nonlinear elements
are calculated by iteration at the end of each time or load step. The uncoupled
modal equations are solved exactly for each time increment.

Several examples are presented that illustrate the efficiency and accuracy of the
method. The computational speed of the new FNA method is compared with the
traditional “brute force” method of nonlinear analysis in which the complete
equilibrium equations are formed and solved at each increment of load. For many
problems, the new method is several magnitudes faster.
FAST NONLINEAR ANALYSIS 18-3

Nonlinear
(Hysteretic) Elements Friction Device

Damping Element

Nonlinear Element

Gap Element
Tension Only Element

Typical Bridge Deck Joint

Friction Pendulum Type Base Isolators Gap Elements between Adjacent Frames

{ XE "Tension Only Element" }{ XE "Gap Element" }Figure 18.1 Examples of

Nonlinear Elements

18.3 FUNDAMENTAL EQUILIBRIUM EQUATIONS

{ XE "Damping:Nonlinear" }{ XE "Nonlinear Equilibrium Equations" }The FNA

method is a simple approach in which the fundamental equations of mechanic
(equilibrium, force-deformation and compatibility) are satisfied. The exact force
equilibrium of the computer model of a structure at time t is expressed by the
following matrix equation:
18-4 STATIC AND DYNAMIC ANALYSIS

Mu
&&(t) + Cu& (t) + Ku(t) + R(t) NL = R(t) (18.1)

where M, C and K are the mass, proportional damping and stiffness matrices,
respectively. The size of these three square matrices is equal to the total number
of unknown node point displacements Nd. The elastic stiffness matrix K neglects
the stiffness of the nonlinear elements. The time-dependent vectors
u
&&(t), u& (t), u (t) and R(t) are the node point acceleration, velocity, displacement
and external applied load, respectively. And R(t) NL is the global node force
vector from the sum of the forces in the nonlinear elements and is computed by
iteration at each point in time.

{ XE "Effective Stiffness" }If the computer model is unstable without the

nonlinear elements, one can add “effective elastic elements” (at the location of
the nonlinear elements) of arbitrary stiffness. If these effective forces, K e u (t ) ,
are added to both sides of Equation (1), the exact equilibrium equations can be
written as:

&&(t) + Cu& (t) + ( K + K e )u (t) = R(t) −R(t) NL + K e u (t)

Mu (18.2)

where K e is the effective stiffness of arbitrary value. Therefore, the exact

dynamic equilibrium equations for the nonlinear computer model can be written
as:

&&(t) + Cu& (t) + Ku (t) = R(t)

Mu (18.3)

The elastic stiffness matrix K is equal to K + K e and is known. The effective

external load R (t ) is equal to R ( t ) − R (t) NL + K e u (t) , which must be evaluated
by iteration. If a good estimate of the effective elastic stiffness can be made, the
rate of convergence may be accelerated because the unknown load term
− R (t) NL + K e u (t) will be small.

18.4 CALCULATION OF NONLINEAR FORCES

At any time the L nonlinear deformations d (t ) within the nonlinear elements are
calculated from the following displacement transformation equation:

d(t ) = bu (t) (18.4)

FAST NONLINEAR ANALYSIS 18-5

Also, the rate of change with respect to time in the nonlinear deformations, d& (t ) ,
are given by:

d& (t ) = b u& (t) (18.5)

Note that for small displacements, the displacement transformation matrix b is

not a function of time and is exact. The displacement transformation matrix b for
a truss element is given by Equation (2.11).

If the time-history deformations and velocities in all nonlinear elements are

known, the nonlinear forces f (t ) in the nonlinear elements can be calculated
exactly at any time from the nonlinear material properties of each nonlinear
element. It is apparent that this can only be accomplished by iteration at each
point in time.

18.5 TRANSFORMATION TO MODAL COORDINATES

The first step in the solution of Equation (18.3) is to calculate a set of N

orthogonal Load Dependent Ritz vectors, Φ , which satisfy the following
equations:
2
Φ T MΦ = ΙI and Φ T KΦ = Ω (18.6a) and (18.6b)

where I is a unit matrix and Ω2 is a diagonal matrix in which the diagonal terms
are defined as ω2n .

The response of the system can now be expressed in terms of those vectors by
introducing the following matrix transformations:

u (t) = ΦY(t) & (t)

u& (t) = ΦY && (t)
&&(t) = ΦY
u (18.7)

The substitution of those equations into Equation (18.1) and the multiplication of
both sides of the equation by Φ T yield a set of N uncoupled equations expressed
by the following matrix equation:

&& ( t ) + ΛY
IY & (t ) + Ω 2 Y(t ) = F(t ) (18.8)
18-6 STATIC AND DYNAMIC ANALYSIS

in which the linear and nonlinear modal forces are given by:

F(t) = Φ T R(t) = Φ T R(t ) − Φ T R(t) NL + Φ T K e u (t) (18.9)

The assumption that the damping matrix can be diagonalized is consistent with
the classical normal mode superposition method in which damping values are
assigned, in terms of percent of critical damping, at the modal level. The diagonal
terms of the Λ matrix are 2 ξn ωn in which ξn is the damping ratio for mode n. It
should be noted that the forces associated with concentrated dampers at any
location in the structure can be included as part of the nonlinear force vector.

Also, if the number of LDR vectors used is equal to the total number of degrees
of freedom Nd,, Equation 18.8 is exact at time t. Therefore, if very small time
steps are used and iteration is used within each time step, the method converges
to the exact solution. The use of LDR vectors significantly reduces the number of
modes required.

Because u (t) = ΦY(t) , the deformations in the nonlinear elements can be

expressed directly in terms of the modal coordinate as:

d (t) = BY(t) (18.10)

where the element deformation - modal coordinate transformation matrix is

defined by:

B = bΦ (18.11)

It is very important to note that the L by N B matrix is not a function of time and
is relatively small in size; also, it needs to be calculated only once before
integration of the modal equations.

At any time, given the deformations and history of behavior in the nonlinear
elements, the forces in the nonlinear elements f(t) can be evaluated from the basic
nonlinear properties and deformation history of the element. From the basic
principle of virtual work, the nonlinear modal forces are then calculated from:

F(t) NL = BT f (t) (18.12)

FAST NONLINEAR ANALYSIS 18-7

The effective elastic forces can also be rewritten as:

F(t) e = Φ T K e u (t) = Φ T b T k e b u(t ) = BT k e d(t ) (18.13)

where k e is the effective linear stiffness matrix in the local nonlinear element
reference system.

18.6 SOLUTION OF NONLINEAR MODAL EQUATIONS

The calculation of the Load Dependent Vectors, without the nonlinear elements,
is the first step before solving the modal equations. Also, the B deformation-
modeshape transformation matrix needs to be calculated only once before start of
the step-by-step solution phase. A typical modal equation is of the form:

&y&(t)n + 2 ξn ωn y& (t) n + ωn 2 y(t) n = f (t) n (18.14)

where f(t)n is the modal load and for nonlinear elements is a function of all other
modal responses at the same point in time. Therefore, the modal equations must
be integrated simultaneously and iteration is necessary to obtain the solution of
all modal equations at time t. The exact solution of the modal equations for a
linear or cubic variation of load within a time step is summarized by Equation
(13.13) and is in terms of exponential, square root, sine and cosine functions.
However, those computational intensive functions, given in Table 13.2, are pre-
calculated for all modes and used as constants for the integration within each
time step. In addition, the use of the exact piece-wise integration method allows
the use of larger time steps.

The complete nonlinear solution algorithm, written in iterative form, is

summarized in Table 18.1.
18-8 STATIC AND DYNAMIC ANALYSIS

{ XE "Algorithms for:Fast Nonlinear Analysis" }Table 18.1 Summary of

Nonlinear Solution Algorithm

I INITIAL CALCULATION - BEFORE STEP-BY-STEP SOLUTION

1. Calculate N Load Dependent Ritz vectors Φ for the structure without the
nonlinear elements. These vectors have Nd displacement DOF.
2. Calculate the L by N B matrix. Where L is the total number of DOF within
all nonlinear elements.
3. Calculate integration constants A 1 − − − for the piece-wise exact
integration of the modal equations for each mode.

II NONLINEAR SOLUTION at times ∆t, 2∆t, 3∆t - - - - - - - -

1. Use Taylor series to estimate solution at time t .

Y(t ) = Y(t - ∆t ) + ∆t Y& (t - ∆t ) + ∆t Y

&&(t - ∆t )
2
& (t ) = Y
Y & (t - ∆t) + ∆t Y
&& (t - ∆t )
2. For iteration i, calculate L nonlinear deformations and velocities.
d(t )i = BY(t )i and d& (t )i = BY
& (t )i
3. Based on the deformation and velocity histories in nonlinear elements,
calculate L nonlinear forces f (t )i .
4. Calculate new modal force vector F(t )i = F(t) - BT [f (t )i − k e d(t )]
5. Use piece-wise exact method to solve modal equations for next iteration.
& (t )i , Y
Y(t )i , Y && (t )i
N N

∑|
n =1
f (t )in | - ∑|
n =1
f (t )in-1|
6. Calculate error norm: Err = N

∑|
n =1
f (t )in |

7. Check Convergence – where the tolerance, Tol , is specified.

If Err > Tol go to step 2 with i = i + 1

If Err < Tol go to step 1 with t = t + ∆t

FAST NONLINEAR ANALYSIS 18-9

18.7 STATIC NONLINEAR ANALYSIS OF FRAME STRUCTURE

The structure shown in Figure 18.2 is used to illustrate the use of the FNA
algorithm for the solution of a structure subjected to both static and dynamic
loads. It is assumed that the external columns of the seven-story frame structure
cannot take axial tension or moment at the foundation level and the column can
uplift. The axial foundation stiffness is 1,000 kips per inch at the external
columns and 2,000 kips per inch at the center column. The dead load is 80 kips
per story and is applied as concentrated vertical loads of 20 kips at the external
columns and 40 kips at the center column. The static lateral load is specified as
50 percent of the dead load.

20 ft 20 ft

MEMBER PROPERTIES
BEAMS I=80,000 IN4 A=300 IN2
CENTER COLUMN I=100,000 IN4 A=300 IN2
7 at 13 ft = 91 ft

OUTER COLUMNS I= 50,000 IN4 A=200 IN2

MODULUS OF ELASTICITY E=4,000 KSI
FOUNDATION STIFFNESS k=1000 K/IN
WEIGHT PER STORY w= 80 Kips

k 2k k

Figure 18.2 Properties of Frame Structure

{ XE "Uplifting Frame" }For the purpose of calculating the dynamic response,

the mass of the structure is calculated directly from the dead load. The
fundamental period of the structure with the external columns not allowed to
uplift is 0.708 seconds. The fundamental period of the structure allowing uplift is
1.691 seconds.
18-10 STATIC AND DYNAMIC ANALYSIS

The static load patterns used to generate the series of LDR vectors are shown in
Figure 18.3. The first load pattern represents the mass-proportional lateral
earthquake load. The second pattern represents the vertical dead load. The last
two load patterns represent the possible contact forces that exist at the foundation
of the external columns. It is very important that equal and opposite load patterns
be applied at each point where a nonlinear element exists. These vectors allow
for the accurate evaluation of member forces at the contact points. For this
example, the vectors will not be activated in the solution when there is uplift at
the base of the columns because the axial force must be zero. Also, the total
number of Ritz vectors used should be a multiple of the number of static load
patterns so that the solution is complete for all possible loadings. In addition, care
should be taken to make sure that all vectors are linearly independent.

Figure 18.3 Four Static Load Vectors Used in Analysis

For this example, the dead load is applied at time zero and reaches its maximum
value at one second, as shown in Figure 18.4. The time increment used is 0.10
second. The modal damping ratios are set to 0.999 for all modes; therefore, the
dynamic solution converges to the static solution in less than one second. The
lateral load is applied at two seconds and reaches a maximum value at three
seconds. At four seconds after 40 load increments, a static equilibrium position is
obtained.

It should be noted that the converged solution is the exact static solution for this
problem because all possible combinations of the static vectors have been
FAST NONLINEAR ANALYSIS 18-11

included in the analysis. The magnitude of the mass, damping and the size of the
time step used will not affect the value of the converged static solution.

Load Dead Load

Lateral Load

1.0 2.0 3.0 4.0 5.0

Time - Seconds

Figure 18.4 Application of Static Loads vs. Time

It is of interest to note that it is impossible for a real structure to fail under static
loads only, because at the point of collapse, inertia forces must be present.
Therefore, the application of static load increments with respect to time is a
physically realistic approach. The approximate static load response of the frame
is shown in Figure 18.5.
600

400 LEFT
RIGHT
200

-200

-400

-600
0 1 2 3 4 5
TIME - Seconds

Figure 18.5 Column Axial Forces from “Static” Loads

18-12 STATIC AND DYNAMIC ANALYSIS

18.8 DYNAMIC NONLINEAR ANALYSIS OF FRAME STRUCTURE

The same frame structure that is defined in Figure 18.2 is subjected to Loma
Prieta Earthquake ground motions recorded on the east side of the San Francisco
Bay at a maximum acceleration of 20.1 percent of gravity and a maximum
ground displacement of 5.81 inches. The acceleration record used was corrected
to zero acceleration, velocity and displacement at the end of the record and is
shown in Figure 18.6.

25

20

15

10

-5

-10

-15

-20

-25
0 1 2 3 4 5 6 7 8 9 10
TIME - seconds

Figure 18.6 Segment of Loma Prieta Earthquake - Percent of Gravity

The dead load was applied as a ramp function in the time interval 0 to 1 second.
The lateral earthquake load is applied starting at 2 seconds. Sixteen Ritz vectors
and a modal damping value of 5 percent were used in the analysis. The column
axial forces as a function of time are shown in Figure 18.7.
FAST NONLINEAR ANALYSIS 18-13

600

400
LEFT
RIGHT
200

-200

-400

-600
0 1 2 3 4 5 6 7 8 9 10
TIME - seconds

Figure 18.7 Column Axial Forces from Earthquake Loading

It is of considerable interest to compare the behavior of the building that is not

allowed to uplift with the behavior of the same building that is allowed to uplift.
These results are summarized in Table 18.2.

Table 18.2. Summary of Results for Building Uplifting Problem from the Loma
Prieta Earthquake ξ = 0.05
Max. Max. Max. Max. Max. Compu-
Displace- Axial Base Base Strain tational
ment Force Shear Moment Energy Time
Uplift (inches) (kips) (kips) (k-in) (k-in) (seconds)
Without 3.88 542 247 212,000 447 14.6
With 3.90 505 199 153,000 428 15
Percent
+0.5 % -6.8% -19.4% -27.8% -4.2% +3%
Difference

{ XE "Uplifting Frame" }The lateral displacement at the top of the structure has
not changed significantly by allowing the external columns to uplift. However,
allowing column uplifting reduces significantly the base shear, overturning
18-14 STATIC AND DYNAMIC ANALYSIS

moment and strain energy stored in the structure. It is apparent for this structure,
that uplifting is a “natural” base isolation system. This reduction of forces in a
structure from uplifting has also been observed in shaking table tests. However, it
has not been used extensively for real structures because of the lack of precedent
and the inability of the design engineer to easily compute the dynamic behavior
of an uplifting structure.

For this small nonlinear example, there is a very small increase in computational
time compared to a linear dynamic analysis. However, for a structural system
with a large number of nonlinear elements, a large number of Ritz vectors may be
required and the additional time to integrate the nonlinear modal equation can be
significant.

Table 18.3 presents a summary of the results if the same structure is subjected to twice
the ground accelerations of the Loma Prieta earthquake. One notes that all significant
response parameters are reduced significantly.
Table 18.3 Summary of Results for Building Uplifting Problem from Two Times
the Loma Prieta Earthquake - ξ = 0.05
Max. Max. Max. Max. Max.
Displace- Column Base Base Strain
ment Force Shear Moment Energy Max. Uplift
Uplift (inches) (kips) (kips) (k-in) (k-in) (inches)
Without 7.76 924 494 424,000 1,547
With 5.88 620 255 197,000 489 1.16
Percent
-24% -33% -40% -53% -68%
Difference

The maximum uplift at the base of the external columns is more than one inch;
therefore, these may be ideal locations for the placement of additional energy
dissipation devices such as viscous dampers.

18.9 SEISMIC ANALYSIS OF ELEVATED WATER TANK

A nonlinear earthquake response analysis of an elevated water tank was

conducted using a well-known commercial computer program in which the
FAST NONLINEAR ANALYSIS 18-15

stiffness matrix for the complete structure was recalculated for each time step and
equilibrium was obtained using iteration. The structural system and analysis had
the following properties:
92 nodes with 236 unknown displacements
103 elastic frame elements
56 nonlinear diagonal brace elements - tension only
600 time steps at 0.02 seconds
The solution times on two different computers are listed below:
Intel 486 3 days 4,320 minutes
Cray XMP-1 3 hours 180 minutes
The same structure was solved using the FNA method presented in this chapter
on an Intel 486 in less than 3 minutes. Thus, a structural engineer has the ability
to investigate a large number of retrofit strategies within a few hours.

18.10 SUMMARY

It is common practice in engineering design to restrict the nonlinear behavior to a

small number of predefined locations within a structure. In this chapter an
efficient computational method has been presented to perform the static and
dynamic analysis of these types of structural systems. The FNA method, using
LDR vectors, is a completely different approach to structural dynamics. The
nonlinear forces are treated as external loads and a set of LDR vectors is
generated to accurately capture the effects of those forces. By iteration within
each time step, equilibrium, compatibility and all element force-deformation
equations within each nonlinear element are identically satisfied. The reduced set
of modal equations is solved exactly for a linear variation of forces during a
small time step. Numerical damping and integration errors from the use of large
time steps are not introduced.

{ XE "Effective Stiffness" }The computer model must be structurally stable

without the nonlinear elements. All structures can be made stable if an element
with an effective stiffness is placed parallel with the nonlinear element and its
stiffness added to the basic computer model. The forces in this effective stiffness
element are moved to the right side of the equilibrium equations and removed
18-16 STATIC AND DYNAMIC ANALYSIS

during the nonlinear iterative solution phase. These dummy or effective stiffness
elements will eliminate the introduction of long periods into the basic model and
improve accuracy and rate of convergence for many nonlinear structures.

It has been demonstrated that structures subjected to static loads can also be
solved using the FNA method. It is only necessary to apply the loads slowly to a
constant value and add large modal damping values. Therefore, the final
converged solution will be in static equilibrium and will not contain inertia
forces. It should be noted that it is necessary to use Load Dependent Vectors
associated with the nonlinear degrees of freedom, and not the exact eigenvectors,
if static problems are to be solved using this approach.

The FNA method has been added to the commercial program ETABS for the
analysis of building systems and the general purpose structural analysis program
SAP2000. The ETABS program has special base isolation elements that are
commonly used by the structural engineering profession. Those computer
programs calculate and plot the total input energy, strain energy, kinetic energy
and the dissipation of energy by modal damping and nonlinear elements as a
function of time. In addition, an energy error is calculated that allows the user to
evaluate the appropriate time step size. Therefore, the energy calculation option
allows different structural designs to be compared. In many cases a good design
for a specified dynamic loading is one that has a minimum amount of strain
energy absorbed within the structural system.

As in the case of normal linear mode superposition analysis, it is the

responsibility of the user to check, using multiple analyses, that a sufficiently
small time step and the appropriate number of modes have been used. This
approach will ensure that the method will converge to the exact solution.

Using the numerical methods presented in this chapter, the computational time
required for a nonlinear dynamic analysis of a large structure, with a small number of
nonlinear elements, can be only a small percentage more than the computational time
required for a linear dynamic analysis of the same structure. This allows large
nonlinear problems to be solved relatively quickly.