Chapter 20: Analysis of nonlinear response

 The equation of motion for a single-degree-of-freedom system is given by

(1)
The displacement, velocity, and acceleration appear in Eq. 1 raised only to the power 1. In
addition, if the mass m, the damping constant c, and the stiffness k remain unchanged with
time the equation is linear. The methods of solution for single as well as multi-degree-of
freedom systems developed in the previous chapters are applicable for linear equations.
Some of these methods, particularly those that rely on superposition, are no longer valid
when the equations become nonlinear. The superposition methods include the Duhamel’s
integral method and the mode superposition method.
Nonlinearity in the system equations may arise from a number of different sources. One
such source is associated with the nature of damping. The linear viscous damping resistance
discussed so far arises from the slow motion of a body in a viscous fluid and is given by as
shown in Eq. 1 . At higher speeds, the resistance offered by the fluid becomes proportional
to the square of the velocity.
Analysis of nonlinear response

The resistance offered by wind is an

example of viscous damping. As
shown in Fig. 1, for low speed of the
moving body such resistance is
proportional to the speed of the
moving body. At a higher speed, the
resistance becomes proportional to
the square of the speed. Figure 1 Representation of a viscous damping force

 Damping produced by dry friction or Coulomb friction would also introduce nonlinearity in the
system. Such damping is given by
(2)
where is the coefficient of friction and N is the force perpendicular to the mating surfaces. The
force of dry friction is always opposed to the direction of motion. Special methods need to be
used to solve the equation that incorporate friction damping.
Analysis of nonlinear response
 Nonlinearity in damping is also produced when the damping arises from internal friction in the
structural material even at small deformations. This is different from the apparent damping
produced by yielding of the material at large deformation. The resistance produce by the
energy loss in internal friction is also known as hysteretic damping.
Because damping force may arise from a number of different sources and because there is
considerable uncertainty in the modelling of damping resistance, a simplified version of the
damping model, the linear viscous damping model, is most frequently employed in dynamic
analysis.
Nonlinearity in the system equations also arises when the structural displacements are large or
when the stress-strain relationship of the structural material become nonlinear. The former
type of nonlinearity is called geometric nonlinearity while the latter is called material
nonlinearity.
Geometric nonlinearity leads to the appearance of higher powers of variables or or both. One
example is the motion of a pendulum. The equation of motion of the pendulum is given by
(3)
Analysis of nonlinear response

 Using a power series expansion of sine function Eq. 3 can be expressed as

(4)
For small values of , the higher order terms of can be ignored and Eq. 4 takes its linear form
(5)
The treatment of nonlinearity introduced by nonlinear stress-strain relationship may be
important in dynamic analysis of structures. This is the case for structures subject earthquake
ground motion. Such structures are often designed to be strained into the inelastic range or to
yield during a design earthquake, so long as they do not collapse. This is because the alternative
design that aims to keep the structure elastic would be too expensive for common structures.
Since the material nonlinearity is the most important from the point of view of the design of civil
engineering structures, we limit our discussion to the nonlinear analysis of structures in which
the nonlinearity is introduced by nonlinear stress-strain relationship of the structural material.
Analysis of nonlinear response
 Figure 2 illustrates the three force terms in the
equation of motion. Figure 2a shows the
relationship between the inertia force and the
acceleration . The slope of the tangent to the
curve represents the mass at a specific instant of
time. Since the mass does not change with time
the slope is a constant equal to m.
The relationship between the damping force and
velocity is represented in Fig. 2b by a straight
line whose slope remains constant at c.
Figure 2c shows an idealized relationship
between spring force and displacement u. The
slope of the tangent to the curve, k, is constant
for small displacement but decreases with
displacement for larger displacement
(c) (d)
Figure 2
Analysis of nonlinear response
 Figures 2d, e, and f show details of three
different type of relationships between and
u as the body undergoes cyclic
displacements. Fig. 2d shows an elasto-
plastic model that is appropriate for certain
type of friction dampers. Figure 2e is referred
to as the Ramberg-Osgood model and
represents the response of ductile steel
(e)
structure. It is sometimes idealized by an
elasto-plastic model. Figure 2f, referred to as
Takeda model, represents the behaviour of
ductile concrete structures.
In each case illustrated in Fig. 2, the
relationship between and u is nonlinear. For
example in the elasto-plastic case the
stiffness reduces to zero as soon as the (f)
displacement exceeds corresponding to
point A on the curve. Figure 2
Analysis of nonlinear response

 In addition, when the velocity reverses direction at point B, the stiffness reverts to the initial
stiffness k. The force-displacement relationship is thus a function of both u and . In fact, the
relationship between and u is not single-valued but depends on the history of displacement
response. The Ramberg-Osgood and Takeda models are even more complicated.
It is apparent that an analytical solution is not possible when material nonlinearity exists, and
numerical methods must be used in the analysis. Such an analysis involves marching along
time in which response results are obtained at regular intervals of time
Single degree of freedom system
The average acceleration and linear acceleration methods can be adopted for the integration
of a nonlinear equation of motion. For this purpose, the equation of motion is expressed in an
incremental form. Th equation of motion at time nh is expressed as
(6)
Single degree of freedom system
 In a manner similar to Eq. 6, the equilibrium at time is represented by

(7)
On subtracting Eq. 6 from Eq. 7 we get the equation of motion in an increment form
(8)
If we assume that over the small interval of time h, the stiffness is constant and is equal to
the tangent stiffness at the beginning, we can write
  (9)
where is the tangent stiffness. A reference to Fig. 2c will
show that for a finite interval of time Eq. 9 will be in
error and the tangent stiffness should be replaced by
secant stiffness. The latter is, however, not known until
(n+1)h
the new displacement is determined. We will hope that h
for small h the error is will not be large.
nh
Fig. 2c
Linear acceleration method
 The time integration schemes discussed in Chapter 8 can be adopted to integrate the
nonlinear system of equations. As an example let us consider the linear acceleration
method. The following equations were derived in Chapter 8.
(10)
(11)
On setting and , Eq. 10 gives
(12)
Substituting in Eq. 11 we get
(13)
Finally, substitution of Eqs. 9, 12 and 13 in Eq. 8 gives
(14)
Linear acceleration method
 Once is obtaind from Eq. 14 , substitution in Eq. 13 will give . These values of and can be
added to and to obtain and , respectively. The acceleration at time can be obtained
similarly by using Eq. 12 to get . The three responses , , and obtained in this manner will
not quite satisfy the equilibrium equation at because of the error involved in using
tangent stiffness instead of secant stiffness. To improve the accuracy of the procedure,
the acceleration is obtained by enforcing total equilibrium at rather than by adding an
increment to the current value
(15)
A similar procedure can be used to derive the equations for the average acceleration
method.
It may be noted that the linear acceleration method is conditionally stable, and for
stability . The period T keeps changing with displacement; therefore the smallest value of
T should be used to ensure stability.
Average acceleration method
 The following equations were derived in Chapter 8

(16)
(17)
Equation 16 gives
(18)
From Eq. 17 we get
(19)
Substitution of Eqs 9, 18, and 19 in Eq. 8 gives
(20a)
or
(20b)
Example 1

 A tower supporting a tank of mass 0.1 kip.s2/in is

shown in the figure. The framing supporting the
tank is assumed to be massless, and the
damping in the system is represented by a
damper with viscous damping coefficient c = 0.2
kip.s/in. The tower is subjected to the blast
loading shown in Fig. b. The force-displacement
relationship is indicated in Fig. c, in which the
nonlinear part of the force-displacement relation
is given by

Obtain the response for first 0.8 s using the

average acceleration method

Figure: (a) Tower; (b) exciting force; (c) force-

displacement relationship
Example 1
 The smallest period of the system correspond to the initial stiffness ki = 8 k/in and is given by

s (a)
A time step of h = 0.1 s which is about 1/7 of the period is used in the integration. Since the
method is unconditionally stable, the selection of h is not governed by stability criterion. From
Eq. 19 we have
(b)
Equation 20 yields
(c)
Equations c and b give and and hence and . The equilibrium equation at , Eq. 15, is then
used to obtain
Example 1

 The response values for the first 0.8 s are shown in the accompanying table. Sample
calculations involved in proceeding from 0.3 s to 0.4 s follow
 Table: Response of a nonlinear system.
t p
(s) (in.) (in./s) (in./s2) (kips) (kips/in.) (in.) (in./s) (in.) (in./s) (kips) (in./s 2)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
0.0 0.0000 0.0000 0.0000 0 8.0000 0.0192 0.3846 0.0192 0.3846 0.1538 7.6924
0.1 0.0192 0.3846 7.6924 1 7.9987 0.1198 1.6273 0.1391 2.0119 1.1093 24.8835
0.2 0.1391 2.0119 24.8835 4 7.9312 0.3626 3.2277 0.5016 5.2396 3.8635 40.8861
0.3 0.5016 5.2396 40.8861 9 7.1053 0.6111 1.7432 1.1128 6.9828 7.2691 3.3438
0.4 1.1128 6.9828 3.3438 9 3.5974 0.5965 −2.0350 1.7093 4.9477 8.0000 −29.8955
0.5 1.7093 4.9477 −29.8955 6 0.0000 0.2225 −5.4450 1.9318 −0.4973 8.0000 −79.0054
0.6 1.9318 −0.4973 −79.0054 0 8.0000 −0.3459 −5.9234 1.5859 −6.4207 5.2328 −39.4866
0.7 1.5859 −6.4207 −39.4866 0 8.0000 −0.6952 −1.0026 0.8907 −7.4833 −0.3288 18.2546
0.8 0.8907 −7.4833 18.2546 0
Example 1
 The response values for the first 0.8 s are shown in the Table. The force-displacement
relationship shows that for u > 1.5 in, fs is constant at 8 kips. The displacement increases
from 1.1128 in. to 1.7093 in. during the interval 0.4 s to 0.5 s. Evidently, the governing
equation for fs changes from to during the period. However, because of the use of discrete
time interval, the change is not detected until
Also, as soon as the velocity reverses, for example at A in the figure, the unloading path is
followed and the stiffness changes from 0 to 8 kips/in. Theoretically, this change should
take place when the velocity is zero. Because of the discrete time interval used in the
calculation, the change in the velocity is detected at 0.6 s, where the velocity is calculated
to be - 0.4973 in/s. It is apparent that the velocity changed between 0.5 and 0.6 s.
Simultaneously with a change in the direction of velocity, the stiffness changed from 0 to 8
kips/in. However, in the calculation kT = 0 was assumed was assumed for the entire duration
of time from 0.5 s to 0.6 s. Evidently, the use of finite step in the integration gives rise to
errors of this type whenever a transition occurs in the force-displacement relationship.
Similar transition also occurs at point C in Fig. c.
 Example 1
It should be noted that in the iterations following 0.6 s, kT has been changed to 8 kips/in., and
the spring force follows the unloading path so that at 0.7 s,
fS = 8 + 8×u = 8 − 8×0.3459 = 5.2328 kips. (d)

Errors involved in numerical integration of nonlinear systems

The errors introduced into the numerical integration because the procedure, in fact,
provides a solution of a finite difference approximation of the equation of motion
rather than of the original difference equation are common to both linear and nonlinear
systems. Additional sources of error are, however, present in the latter case. These
additional errors can be classified as those arising due to (1) use of tangent stiffness
in place of secant stiffness, and (2) delay in detecting the transitions in the force–
displacement relationship.
Errors involved in numerical integration of nonlinear systems
 The errors due to the use of tangent stiffness can be minimized by using a process of
iteration. Consider, for example, the graphical representation of Equation 20b shown in Figure
3a. The relationship shows a nonlinearity because the slope is not constant, being dependent
on displacement and hence on time. In a static load case will be equal to and the
nonlinearity will be even more severe. The presence of mass and damping terms in the
expression for smooth out the nonlinearity. In particular, when h is small the constant terms
and are much larger when compared to .

Figure 3: Nonlinear system:

iterations within time step:
(a) use of initial tangent
stiffness; (b) use of current
tangent stiffness.
Errors involved in numerical integration of nonlinear systems
 The iteration is begun with the equation

(21)
For the softening type of nonlinearity shown in Figure 3a, where the spring stiffness decreases with
an increase in displacement, the equivalent spring force due to a displacement of is less than that
required to balance the force . As a result, a residual force remains as shown in Figure 3a. Let this
residual force be denoted by . Additional displacement is caused by the residual force so that
(22)
This additional displacement is used to find a new value of the residual force, and the process is
continued until convergence is achieved. Iteration within a time step can be expressed as

(23)
Errors involved in numerical integration of nonlinear systems
 In Eq. 23 represents the spring force. Also, iteration commences with . When the iterative
process has converged, that is, when or is sufficiently small, the total incremental
displacement is given by
(24)
The incremental acceleration and velocity can now be obtained from relationships similar to
Eq. 18 and 19.
The procedure outlined above, in which initial stiffness kT is used in each iteration is referred
to as the modified Newton Raphson method. The iteration process will converge to the
correct displacement, but the rate of convergence may be slow. A better rate of convergence
is achieved by using the current tangent stiffness in place of the initial stiffness. This,
however, requires that tangent stiffness be evaluated at each iteration, involving additional
computation. The procedure is illustrated in Fig. 3b and is called the full Newton Raphson
method.
Errors involved in numerical integration of nonlinear systems

The errors involved in numerical integration during transitions in the force–displacement

relationship can be minimized by using a sub-increment of time to carry out the integration
during an interval in which a transition is detected. For example, if it is found that the
velocity changed sign during the interval from tn to tn+1, integration can be retraced in the
interval tn to tn+1 with a smaller time step, say, h/5.
Alternatively, an iterative process may be used in which integration is resumed from tn with a
smaller step whose size is progressively adjusted so that at the end of such an adjusted time
step, the velocity is close to zero.
The criteria for the selection of time step used in integration are similar to those for a linear
system. The natural period of the system, however, varies as the stiffness changes. The limits
imposed on ωh or h/T to minimize truncation errors and to ensure stability of the system
should therefore be based on the largest value of ω or the smallest value of the period.
Example 2
In Example 1 use an iterative procedure to take care of the nonlinearity in the force-displacement
relationship in the progression from 0.3 s to 0.4s using average acceleration method

 The iterations procedure is given by

The following values are known

Figure 3a
Example 2

 The first iteration is

51.1053

Figure 3a
Multiple degree-of-freedom system

 The equations of motion for a MDOF system with nonlinear material properties can be
expressed as
(25)
In Eq. 25 mass and damping matrices are constant. However, because of material
nonlinearity the spring forces are not directly proportional to the displacement. This makes
the calculation of spring forces much more involved. Methods of determining the vector of
spring forces for nonlinear systems are described in standard texts. As well, standard
computer programs have algorithms to determine them.
The analysis procedures for nonlinear MDOF systems are similar to those for SDOF systems.
They can be divided into explicit and implicit methods. The explicit methods involve less
computations than the implicit methods. However, they have strict conditions on the time
step used in the numerical integration to guarantee stability. On the other hand some of the
implicit methods are unconditionally stable. One such method is the average acceleration
method, which we describe in the following.
Average acceleration method for the integration of nonlinear MDOF system

 As in the case of SDOF system the equations of motion in an incremental from are

(26)
where
, (27)
The remaining steps are
(28)
(29)
(30)
where
(31)
(32)
Average acceleration method for the integration of nonlinear MDOF system

 The incremental displacement vector is obtained by solving Eq. 30. Substitution of in

Equation 29 provides the value of . The incremental values and are added to and ,
respectively to obtain and . The acceleration at time is then obtained by enforcing
equilibrium between inertia, damping, and spring forces and applied load at time .
Since the average acceleration method is unconditionally stable, no restriction need be
placed on the time step to ensure stability. If another method which is only conditionally
stable is used the time step should not be larger than a fraction of the smallest period of
the structure, taking into account the fact that such period may change with time.