0% found this document useful (0 votes)
206 viewsAnyl Chopra - Numerical Structural Analysis
Chopra, Dynamic, Structure
Uploaded by
adem ademovicCopyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
206 viewsAnyl Chopra - Numerical Structural Analysis
Chopra, Dynamic, Structure
Uploaded by
adem ademovicCopyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
/ 31
5S
Numerical Evaluation of
Dynamic Response
PREVIEW
Analytical solution of the equation of motion for a single-degree-of-freedom system is
usually not possible if the excitation—applied force p(t) or ground acceleration iig(t)—
varies arbitrarily with time or if the system is nonlinear. Such problems can be tackled by
numerical time-stepping methods for integration of differential equations. A vast body of
literature, including major chapters of several textbooks, exists about these methods for
solving various types of differential equations that arise in the broad subject area of ap-
plied mechanics. ‘The literature includes the mathematical development of these methods;
their accuracy, convergence, and stability properties; and computer implementation.
Only a brief presentation of a very few methods that are especially useful in
dynamic response analysis of SDF systems is included here, however. This presentation
is intended to provide only the basic concepts underlying these methods and to provide
a few computational algorithms. While these would suffice for many practical problems
and research applications, the reader should recognize that a wealth of knowledge exists
‘on this subject.
5.1 TIME-STEPPING METHODS
For an inelastic system the equation of motion to be solved numerically is
mii + cit + fo(u, i) = p(t) 6.11)
185 156 Numerical Evaluation of Dynamic Response Chap. 5
subject to the initial conditions
u=u0) «= 1(0)
‘The system is assumed to have linear viscous damping, but other forms of damping,
including nonlinear damping, could be considered, as it would become obvious later.
This is rarely done for lack of information on damping, especially at large amplitudes of
motion. The applied force p(t) is given by a set of discrete values p; = p(t,), 1 = 0 to
N (Fig. 5.1.1). The time interval
AG = hk (6.1.2)
is usually taken to be constant, although this is not necessary. The response is deter-
mined at the discrete time instants f;, denoted as time i; the displacement, velocity, and
acceleration of the SDF system are u;, ti, and ii, respectively. These values, assumed
to be known, satisfy Eg. (5.1.1) at time i
mii; + ct; + (fs) = Pi (5.1.3)
where (fs); is the resisting force at time i; (fs): = ku; for a linearly elastic system but
would depend on the prior history of displacement and velocity if the system were inelas-
tic. The numerical procedures to be presented will enable us to determine the response
quantities ;.1, 4:41, and i;,, at the time instant i +1 that satisfy Eq. (5.1.1) at time i+1:
imiigyy + clings + Cfsdint = Pist (5.14)
>
st
Py
Po pat
=!
b nb 4 tn
+] At,
Figure 5.1.1 Notation for time-stepping methods. Sec. 5.2 Methods Based on Interpolation of Excitation 187
‘When applied successively with = 0, 1,2,3,..., the time-stepping procedure gives
the desired response at all time instants ¢ = 1,2.3,.... The known initial conditions
provide the information necessary to start the procedure.
Stepping from time i to +1 is usually not an exact procedure. Many approximate
procedures are possible that are implemented numerically. The three important require-
ments for a numerical procedure are (1) convergence—as the time step decreases, the
numerical solution should approach the exact solution, (2) stability—the numerical solu-
tion should be stable in the presence of numerical round-off errors, and (3) accuracy—the
numerical procedure should provide results that are close enough to the exact solution.
Although these are important issues, they are mentioned only briefly in this book.
Three types of time-stepping procedures are presented in this chapter: (1) methods
based on interpolation of the excitation function, (2) methods based on finite difference
expressions of velocity and acceleration, and (3) methods based on assumed variation of
acceleration. Only one method is presented in each of the first two categories and two
from the third group.
5.2 METHODS BASED ON INTERPOLATION OF EXCITATION
A highly efficient numerical procedure can be developed for linear systems by interpo-
lating the excitation over each time interval and developing an exact solution for a linear
system using the methods of Chapter 4. If the time intervals are short, linear interpol:
is satisfactory. Figure 5.2.1 shows that over the time interval f; <1 < fj1, the excitation
function is given by
Api
Pe) = pit at (5.2.1a)
where
ABi = Pit ~ Pi (5.2,1b)
and the time variable t varies from 0 to At,. For algebraic simplicity, we first consider
systems without damping; later, the procedure will be extended to include damping. The
P
Actual \, Pig
Figure 5.2.1 Notation for linearly
Het interpolated excitation, 158 Numerical Evaluation of Dynamic Response Chap. 5
equation to be solved is
Api
mit + ku = prt he
The response u(c) over the time interval 0 < t < Af; is the sum of three parts:
(1) free vibration due to initial displacement u; and Velocity i; at t = 0, (2) response
to step force p; with zero initial conditions, and (3) response to ramp force (Ap,/At;)t
with zero initial conditions, Adapting the available solutions for these three cases from
Sections 2.1, 4.3, and 4.4, respectively, gives
(6.2.2)
a ay Pig Ap (4 _ sinoyt
u(t) =u, cosine + Se sinaye +E —coscoye) + (z RE ) (5.2.3a)
and
HO) =u, sinayr +H cosagt + & sinaye + APL (1 =cosa,t) (5.2.36)
On oa k ko, Ah
Evaluating these equations at t= Af; gives the displacement u,4; and velocity #1 at
time i + 1:
tras =u; COSC, At) +“! sin(oy At)
oy
Sp 1
om Ati
on = 608(, At, — sin(@, At} (5.2.4a)
tit
= —u; sin(w, At) +“ cos(w, An)
@y
Pi Api _1
+ Sine At) + FO
‘These equations can be rewritten after substituting Bq. (5.2.1b) as recurrence formulas:
ini = Au; + Bit; + Cp; + Dpist (5.2.5a)
ign = Alu; + By + C’p; + D'pisn (5.2.5b)
{1 = cos(@, AG)) (5.2.46)
These formulas also apply to damped systems with the expressions for the coeffi-
cients A, B, ... , D! given in Table 5.2.1 for under-critically damped systems (i.e.,
£ <1). They depend on the system parameters w,, k, and ¢, and on the time interval
At= Ay.
Since the recurrence formulas are derived from exact solution of the equation of.
motion, the only restriction on the size of the time step Ar is that it permit a close
approximation to the excitation function and that it provide response results at closely
spaced time intervals so that the response peaks are not missed. This numerical procedure
is especially useful when the excitation is defined at closely spaced time intervals—as
for earthquake ground acceleration—so that the linear interpolation is essentially perfect.
If the time step A¢ is constant, the coefficients A, B, ..., D’ need to be computed only
once. Sec. 5.2 Methods Based on Interpolation of Excitation 159
TABLE 5.2.1 COEFFICIENTS IN RECURRENCE FORMULAS (¢ < 1)
2g
Jsomar-(1- Baal
25
2 At Ai
7 Sinep At + <— cosy At
sinwp Ar
3 oa)
Blac" (cose At — Sa sinwp AP
1-2
¢ 1
ae | sino ar + — cosy at
=n) at |)
: net ( $i sinap a1 +09 »)|
vi-3?
The exact solution of the equations of motion required in this numerical procedure
is feasible only for linear systems. It is conveniently developed for SDF systems, as
shown above, but would be cumbersome for MDF systems unless their response is
obtained as superposition of modal responses (Chapters 12 and 13).
Example 5.1
‘An SDF system has the following properties: m = 0.2533 kip-sec/in., k = 10 kips/in.,
Tn = 1 sec (pn = 6.283 rad/sec), and ¢ = 0.05. Determine the response u(¢) of this
system to p(t) defined by the half-cycle sine pulse force shown in Fig. E5.1 by (a) using
piecewise-linear interpolation of p(t) with Ar = 0.1 sec, and (b) evaluating the theoretical
solution,
Sotution
1, Initial calculations
e FOr St 0.9691 wp = wn — 5? = 6.275
sinwp Ar = 0.5871 coswp At = 0.8095 160 Numerical Evaluation of Dynamic Response = Chap. 5
p, kips
, 10 sin (mt / 0.6)
8.66
10
10
Piecewise linear
interpolation
1, 8e¢
06 Figure E5.1
Substituting these in Table 5.2.1 gives
A=08129 B =0.09067 C =0.01236 D =0.006352
Al = —3,5795 BY = 0.7559 Cc’ = 0.1709 D! = 0.1871
2. Apply the recurrence equations (5.2.5). The resulting computations are summa-
rized in Tables B5.1a and E5.1b.
TABLE E5.1a NUMERICAL SOLUTION USING LINEAR INTERPOLATION OF EXCITATION
hi Pi Ce Deis Bley iy Aui u; Theoretical w;
0.0 0.0000 0.0000 0.0318 +—-.0000 0.0000 -0.0000 (0.0000 0.0000
0.1 5.0000 0.0618 0.0550 0.0848 0.9354 0.0258 (0.0318 0.0328
0.2 8.6602 0.1070 0.0635 0.2782 3.0679 0.1849 0.2274 0.2332
0.3 10.0000 0.1236 0.0550 0.4403 4.8558 0.5150 0.6336 0.6487
0.4 8.6603 0.1070 0.0318 0.4290 4.7318 0.9218 1.1339 1.1605
05 5.0000 0.0618 0.0000 0.1753 1.9336 1.2109 1.4896 1.5241
0.6 0.0000 0.0000 0.0000 -0.2735 -3.0159 1.1771 1.4480 14814
07 0.0000 0.0000 0.0000 0.6767 -7.4631 0.7346 0.9037 0.9245
0.8 — 0.0000 0.0000 0.0000 -0.8048 88765 0.0471 (0.0579 0.0593,
0.9 — 0.0000 0.0000 0.0000 -0.6272 -6.9177 -0.6160 0.7577 ~0.7751
1.0 0.0000 -25171 1.24320 1.
3. Compute the theoretical response. Equation (3.2.5)—valid for t < 0.6 sec,
Eq. (2.2.4) modified appropriately—valid for t > 0.6 sec, and the derivatives of these
‘two equations are evaluated for each fy; the results are given in Tables ES.la and E5.1b,
4. Check the accuracy of the numerical results. The numerical solution based on
piecewise linear interpolation of the excitation agrees reasonably well with the theoretical
solution. The discrepancy arises because the half-cycle sine curve has been replaced by
the series of straight lines shown in Fig. E5.1. With a smaller Ar the piecewise linear
approximation would be closer to the half-cycle sine curve, and the numerical solution
would be more accurate, Sec. 5.3 Central Difference Method 161
TABLE E5.1b NUMERICAL SOLUTION USING LINEAR INTERPOLATION OF EXCITATION
fi Pi Cpr Dip Alu uy Bis; iy Theoretical it;
0.0 0,000 0.0000 0.9354 + 0,0000 0.0000 0.0000 0.0000 0.0000
0.1 5.0000 0.8544 1.6201 0.0318 0.7071 (0.9354 0.9567
0.2 8.6602 1.4799 1.8707 0.2274 «= 2.3192 3.0679 3.1383
0.3 10.0000 1.7088 1.6201 0.6336 © 3.6708 4.8558 4.9674
0.4 8.6603 1.4799 0.9354 1.1339 35771 4.7318 4.8408
0.5 5.0000 0.8544 0.0000 14896 1.4617 1.9336 1.9783
0.6 0.0000 0,000 0.0000 14480-22799 -3.0159 —3.0848
0.7 0.0000 0,000 0.0000 0.9037 -5.6418 7.6346
0.8 0.0000 0.0000 0.0000 0.0579 -6.7103 9.0808
0.9 0.0000 0.0000 0.0000 -0.7577 -5.2295 -69177 —7.0771
1.0 0.0000 = 1.2432 -25171 2.5754
5.3 CENTRAL DIFFERENCE METHOD
This method is based on a finite difference approximation of the time derivatives of
displacement (e., velocity and acceleration). Taking constant time steps, Ar; = At, the
central difference expressions for velocity and acceleration at time i are
= 2uj + ui
(ary
(5.3.1)
‘Substituting these approximate expressions for velocity and acceleration into Eq. (5.1.3),
specialized for linearly elastic systems, gives
ui tit
(an?
Hist + kui = pi 63.2)
In this equation 1, and u;_1 are assumed known (from implementation of the procedure
for the preceding time steps). Transferring these known quantities to the right side leads
to
me mc 2m
[ae 7 aa ft eS lar 7 af 7 [« 7 are a
or
huis = Bi (6.3.4)
where
(5.3.5) 162 ‘Numerical Evaluation of Dynamic Response Chap. 5
and
me
oe 3
sit [4 2at 636)
The unknown u;;1 is then given by
Mis1 (5.3.7)
The solution w:.1 at time i +1 is determined from the equation of motion at time i
without using Eq. (5.1.4), the equation of motion at time i + 1. In particular, the elastic
and damping forces can be computed explicitly using known displacements u; and w;_1
and velocities 4, and 2;_,. Such methods are called explicit methods.
Thus wo and u_y are required to determine u; the specified initial displacement ug
is known, To determine u_1, we specialize Eq, (5.3.1) for i = 0 to obtain
mya = Quo ta
"0" ~ Tat (aye eo)
Solving for w; from the first equation and substituting in the second gives
2
ty = uo — Attiig) + iin (639)
‘The initial displacement uo and initial velocity wo are given, and the equation of motion
at time 0 (% = 0),
milo + Cito + kuo = po
provides the acceleration at time 0:
ki
fig = BO Ho ~ ko (63:10)
i
Table 5.3.1 summarizes the above-described procedure as it might be implemented on
the computer.
The central difference method will “blow up,” giving meaningless results, in the
presence of numerical round-off if the time step chosen is not short enough. The specific
requirement for stability is
at 1
Th O01 i
Central difference 0 Linear
(/-Lineat acceleration z acceleration
Jf Average acceleration
00
02
0 ol 02 03 04
Figure $5.2 (a) Amplitude decay versus A/T; (b) definition of AD and PE; (c) petiod elongation
versus At/T,
short enough to keep the distortion of the excitation function to a minimum. A very fine
time step is necessary to describe numerically the highly irregular earthquake ground
acceleration recorded during earthquakes; typically, Ar = 0.02 sec is chosen and this
dictates a maximum time step for computing the response of a structure to earthquake
excitation.
One useful, although unsophisticated technique for selecting the time step is to
solve the problem with a time step that seems reasonable, then repeat the solution with
a slightly smaller time step and compare the results, continuing the process until two
successive solutions are close enough.
The preceding discussion of stability and accuracy applies strictly to linear systems.
The reader should consult other references for how these issues affect nonlinear response
analysis. 174 Numerical Evaluation of Dynamic Response Chap. 5
5.6 ANALYSIS OF NONLINEAR RESPONSE: CENTRAL
DIFFERENCE METHOD
The dynamic response of a system beyond its linearly elastic range is generally not
amenable to analytical solution even if the time variation of the excitation is described by
a simple function. Numerical methods are therefore essential in the analysis of nonlinear
systems, The central difference method can easily be adapted for solving the nonlinear
equation of motion, Eq. (5.1.3), at time i. Substituting Eqs. (5.3.1), the central difference
approximation for velocity and acceleration, gives Eq. (5.3.2) with ku; replaced by (fs)i-
which can be rewritten to obtain the following expression for response at time i + 1:
huiys = Bi (6.6.1)
where
cot in 6.6.2)
and
me 2m
pi [as - sl wan (Bit ante 6.63)
Comparing these equations with those for linear systems, itis seen that the only difference
is in the definition for f,. With this modification Table 5.3.1 also applies to nonlinear
systems.
The resisting force (fs): appears explicitly as, it depends only on the response at
time i, not on the unknown response at time i+1. Thus it is easily calculated, making the
central difference method perhaps the simplest procedure for nonlinear systems. Although
attractive in this regard, the method is not popular for practical or research applications
because more effective methods are available.
5.7 ANALYSIS OF NONLINEAR RESPONSE: NEWMARK’S METHOD
In this section Newmark’s method, described in Section 5.4 for linear systems, is extended
to nonlinear systems. Although not as simple as the central difference method, it is
perhaps the most popular method because of its superior accuracy.
The difference between Eqs. (5.1.3) and (5.1.4) gives an incremental equilibrium
equation:
m Aiiy + ¢ Att; + (Afs)i = Api (1.
The incremental resisting force
(Afs)i = (Kidsee Mui (5.7.2)
where the secant stiffness (k;)gec, shown in Fig. 5.7.1, cannot be determined because 4.44
is not known. If we make the assumption that over a small time step Ar, the secant Method 175
Sec. 5.7 Analysis of Nonlinear Response: Newmi:
Foie
Gd:
uy
Py Pan Figure 5.7.1
stiffness (K;)sec could be replaced by the tangent stiffness (k;)r shown in Fig. 5.7.1, then
Eq. (5.7.2) could be approximated by
(fs): ~ Kir uy (6.7.3)
Dropping the subscript T from (Kj)r in Eq. (5.7.3) and substituting it in Eq. (5.7.1)
gives
«mM Bil; +c At; + kj Quy = Ap; (5.7.4)
The similarity between this equation and the corresponding equation for linear systems,
Eq. (5.4.12), suggests that the noniterative formulation of Newmark’s method presented
carlier for linear systems may also be used in the analysis of nonlinear response. All that,
needs to be done is to replace k in Eq. (5.4.14) by the tangent stiffness &; to be evaluated
at the beginning of each time step. This change implies that step 1.3 of Table 5.4.2
should follow step 2.1. For nonlinear systems step 2.5 and Eq. (5.4.17) would give
different values of ii; and the latter value is preferable because it satisfies equilibrium
at time i + 1
This procedure with a constant time step Ar can lead to unacceptably inaccurate
results. Significant errors arise for two reasons: (1) the tangent stiffness was used
instead of the secant stiffness, and (2) use of a constant time step delays detection of the
transitions in the force-deformation relationship.
First, we consider the second source of error, illustrated by the force-deformation
relation of Fig. 5.7.2a. Suppose that the displacement at time i, the beginning of a time
step, is u; and the velocity 1; is positive (ie., the displacement is increasing); this is
shown by point a. Application of the previously described numerical procedure for the
time step results in displacement 4, and velocity i;,, at time i + 1; this is shown by
point b. If 1i;41 is negative, then at some point 2/ during the time step, the velocity
became zero, changed sign, and the displacement started decreasing, In the numerical
procedure, if we do not bother to locate b’, continue with the computations by starting 176 Numerical Evaluation of Dynamic Response Chap. 5
Js
2 Numerical
Exact
Numerical
uM
fa) (b)
Figure 5.2.2
the next time step at point b, and use the tangent stiffness associated with the unloading
branch of the force~deformation diagram, this procedure locates the point c at the end of
the next time step with displacement u;42 and negative velocity. On the other hand, if
the time instant associated with b/—when the velocity actually became zero—could be
determined, computations for the next time step would start with the state of the system
at b’ and determine the displacement and velocity at the end of the time step, identified
as c'. Not locating b’ has the effect of overshooting to b and not following the exact
path on the force-deformation diagram. These departures from the exact path would
‘occur at each reversal of velocity, leading to errors in the numerical results. A similar
problem arises at sharp corners in the force-deformation relationship, as in elastoplastic
systems.
These errors could be avoided by locating 6’ accurately. This could be achieved
by retracing the integration over the time interval 1; to t;1 with a smaller time step,
say, At/4, Alternatively, an iterative process may be used in which integration is re-
sumed from time i with a step smaller than the full time step, whose size is progres-
sively adjusted so that at the end of such an adjusted time step, the velocity is close
to zer0.
Now, we return to the first source of error that is associated with the use of
tangent stiffness instead of the unknown sccant stiffness, and is illustrated by the force
deformation relation of Fig. 5.7.2b. The displacement at time i, the beginning of a
time step, is shown as point a. Using the tangent stiffness at a, numerical integration
from time i to time i + 1 leads to the displacement 1,,1, identified as point 6. If we
were able to follow the curve exactly, the result may have been the displacement at
4’. This discrepancy accumulating over a series of time steps may introduce significant
errors.
These errors can be minimized by using an iterative procedure. The key equation
that is solved at each time step in Newmark’s method is Eq. (5.4.13), which, modified
for nonlinear systems, becomes
& Au; = Api (5.7.5) Sec. 5.7 Analysis of Nonlinear Response: Newmark's Method 177
where A; is given by Eq. (5.4.15) and
i
kj + 0+ 5.76)
tame ae” (5.7.6)
For convenience in notation we drop the subscript / in k; and replace it by 7 to emphasize
that this is the tangent stiffness; also, the subscript i is dropped from Au; and Ajy.
Equations (5.7.5) and (5.7.6) then become
ky Au = Ap (6.7.7)
ky
and
=kp+—oc+—t (5.7.8)
Bar” Ba”
Figure 5.7.3a shows a schematic plot of Eq. (5.7.7). The relationship is nonlinear because
the tangent stiffness kr depends on the displacement u and hence the slope kr is not
constant. In static analysis of a nonlinear system, ky = kr and the nonlinearity in kr is
the same as in ky. In dynamic analysis the presence of mass and damping terms in ky
decreases the nonlinearity because the constant term m/B (Ar)? for typical values of At
is usually much larger than kr.
Baal
—t {- 2
2 =
ey ia
: af i
al) ln Ye
0
a | a |e
® ©)
Figure 5.7.3. teration within a time step for nonlinear systems: (a) modified Newton-Raphson
iteration; (6) Newton-Raphson iteration.
The iterative procedure is described next with reference to Fig. 5.7.3a. The first
iterative step is the application of Eq. (5.7.7) in the procedure described previously:
ky Ou = Ap (3.7.9)
to determine Au‘) (corresponding to point » in Fig. 5.7.2b), the first approximation to
the final Aw (corresponding to point &/ in Fig. 5.7.2b). Associated with Au‘ is the tue
force Af‘, which is less than Af, and a residual force is defined: AR® = Ap-Af™. 178 Numerical Evaluation of Dynamic Response Chap. 5
‘The additional displacement Au” due to this residual force is determined from
ky Au® = AR® = ap—af” (5.7.10)
This additional displacement is used to find a new value of the residual force, and the
process is continued until convergence is achieved. This iterative process for the time step
i toi+1, summarized in Table 5.7.1, is known as the modified Newton-Raphson method.
TABLE 5.7.1 MODIFIED NEWTON-RAPHSON ITERATION
1.0. Initialize data.
Wau (=U) ARM HAG hr =
2.0 Calculations for each iteration, j = 1, 2, 3,.
21 Solve: ky Au) = ARY = Au?
a _ 0-0 D
2.2 uP = Wiz? + au?
23° af = 2) — f+ (hr — kr) du.
24 ARGH = aR®— af)
3.0. Repetition for next iteration. Replace j by j + 1 and repeat calculation steps 2.1 to 2.4,
The iterative process is terminated after £ iterations when the incremental displace-
ment Au becomes smali enough compared to the current estimate of Au = S7i_, Au;
that is,
‘Au®
Au
Then the displacement increment over the time step i to i +1 is given by
