H3/4

Finite Element Program for Creep Analysis in Concrete Structures

Z.P. Bazant., E.C. Rossow, G. Horrigmoe
Dept. of CMI Engineering.
Nonhwestem University. Evanston. Illinois 60201. U.S.A.

Abstract

A new finite element program for the analysis of creep effects in concrete structures,
developed at Northwestern University, is briefly described. The program represents aD exten-
sion of the general purpose finite element code NONSAP developed at university of California
at Berkeley. The program uses a rate type creep law based on expanding the reiaxation func-
tion in Dirichlet series. The creep law is linear, the aging is accounted for, and the telll-
perature effects on creep rate as well as the rate of aging are considered. Xaterial proper-
ties can be input either as empirical creep test daca or as the double power law for creep.
The program caQ automatically set up the creep parameter from partial empirical information.
rhe program uses a step-by-step integration in time. Applications are envisaged chiefly for
nuclear concrete structures.
1. Int~oduci

"ith the increas1."lg public concern about the safety cf nuclear rea=tors. >: " ;-~c"':

imperative to evaluate the strengths and deformations of nuclear reactor veSSels and corototn-
ments very carefully 'and realistically, taking into account the actual inelastic behaVior of
the material. In concrete reactor contair.ments as -:e1l as vessels acti.."lg as energy-absor_
bing barriers, the major source of inelastic behaVior during the service life is the creep
of concrete. Ouring the past decade the lcnowledge of this phenomenon advanced tremendously
and methods that allow an accurate and realistic analysis on the service conditions have b~
developed.
A general purpose finite element code for the creep analysis of nuclear concrete struc_
tures has recently been developed by Anderson and coworkers ~1, 7 ,10J. This program utilizes
a rate type creep law proposed' in 1971 by Bazant [3], which is based on expanding the compli_
ance function for aging concrete (also called creep function) into a series of real exponen_
tials, called Dirichlet series. Subsequently, it was found, however, [8J that this formula-
tion has some theoretical disadvantages from the point of view of thermodynamic restrictions
resulting from the aging of concrete; although the general form of the creep constitutive
relation is correct from the thermodynaic·vwp~ the experimental creep data for
concrete are fit~ some parameters of the model can be negative although according to ther-
modynamic limitations they should be positive. This is not necessarily incorrect, since
these are internal variables while the thermodynamic restrictions must be satisfied only in
total by the ensemble of the internal variables. However, such a feature is a matter of con-
cern and it appears therefore safer to use a constitutive relation where the fulfillment of
the thermodynamic restrictions is assured. This appears to be the case for an analogous rate
type formulation which i~ based on a Dirichlet series expansion of the relaxation function
rather than the compliance function and is modeled by an aging Maxwell chain model [2,4,5,6J.
This formulation has been used in the extension of the f'inite element code XO::SAP (rniversity
of California, Berkeley) described in the present paper. •
2. Constitutive Relation for Creep
The creep law is assumed to be linear, i.e., following the principle of superposition
[2]. The linearity assumption is acceptable for concrete only for stresses less than about
one-half of the strength, which is typical of nuclear vessels and containments under service
conditions. Even vithin this range, significant deviations from linearity are observed upon
unloading and in a regime vi~h decreasing strains. In such cases, the program cannot yield
very good results. Another phenomenon which spoils the linearity even within the service
stress range is the simultaneous drying; however, this is not severe for nuclear concrete
structures because they are massive and also protected by the steel liner on the interior
face.
The aging of concrete, particularly the gradual change of creep properties as a function
of age, is taken into account. This phenomenon causes that the parameters of Dirichlet series
expansion of the relaxation function, representing the elastic moduli of the Maxwell chain
model, are dependent on the age of concrete.

'"Another large finite element program for creep in nuclear concrete structures has also been
recently developed; we do not discuss it, however, since the form of the constitutive re-
lation us~ in tbat program does not conform to the thermodynamic restrictions imposed by
tbe phenomenon of aging.

H 3/4
-2-
 The effect of temperature variations (limited to temperatures under SOoC) upon the creep
behavior is taken into account. This includes boeh (a) the inerease of ehe creep race due to
heating, and (b) the aeeeleraeion of aging, as lIIOdeled by the equivalene hydration period
(maturity) [4,2]. Both of these effeces are lIIOdeled using ehe aetivation energy eoneept, wtth
different aetivaeion energies for the creep race and for the process of aging (hydration).
The ereep formulae ion used in the program is not eapable of handling the general effects
of time-variable pore bmlidity in eonereee. These effeces bring about a major eomplieation
oi t~e constitueive relation,[2J and require many additional variables to be used co eharac-
terize partial stresses in solid and water eomponents, and they also require a good model for
tensile craeking due to hygral stresses. Nevertheless, omission of moiseure effects appears
to be admissible for massive seructures such as nuclear vessels and coneainments. This is be-
eause of their thickness, as well as the use of the steel liner of the interior face, which
does not" permit any signifieant drying to occur expect in a narrow layer Rear the exterior
surfaee.
However, an apPlroximate consideration of the bJmidity effect which does noe cause any
major complication of ehe maehemaeical formulation, is possible with the present program. The
user is allowed to specify for eaeh finite element the mean ambient humidity (not the pore
humidity in concrete, and not a time variable quant:ity). The coeffieients of the creep law
are then adjusted to give a ereep funetion that: eorresponds to eonerete eylinders exposed to
t::1s environmental lImidity, kept constant for the duration of the ereep test. The approxi-
",ate praetical ereep predieeion model given in pares I and III of Ref. 7 is used for this Put'-
pose. As for the definition of shrinka:ge, the program ca .. use the formulas from Ref. 7
ur a set of experinentally observed shrinkage values can be specified. It must be remembered,
:r.a1: the foregoing considera::i"n "f ~idty effects is very erude and yields realistic values
only for the overall internal force resultants caused by humidity effeets within the eross
seetion of the structure. The method does not permit evaluaeing the distribution of the
stresses produeed by humidity effects throughout the eross section.

3. ~lumerica Algorithm
As has been elearly 4.ei:lonstrated before e.g. ,[2J, a rate type creep law allows a mueh
more effieient finite element analysis than the integral-type ereep law based" directly on the
principle of superposition. As already mentioned, the present program uses a rate-type creep
law based on expanding the relaxation function of eoncrete into a series of real exponentials,
called Dirichlet series. This ereep formulation may be visualized by the Maxwell ehain
model, the elastie 1I0duli~of whieh depend on the age of concrete, lIIore precisely the equiva-
lent hydration period. This creep formulation is completely specified by a set of relaxation
times, which are automatically selected in tbe program according to the rules given in Ref. 2
and ·5, and by the ag.e dependent elastic moduli of the Maxvell chain. ' These !Iloduli ar~ auto-
:nat ically generated fro:! t::'e relaxation function, ..hieb itself is automatieally determined"
fro:! t~e co:!pl iance f!.!r.ct::!..~n (creep function). The program also makes a cheek !>y reproduc ing
the creep curve on the ;,asis of t::'e )I.a1"ell chain model (as presented in Ref. 6).
T::'e rate t::"e cree;> law 1.5.2l consists of a set: of first order differential equations for
the ?areial stresses of :he ~xwel chain and for the total strain of the Maxwell chain. It
has ,een shown before that the !.!sual step-by~!, :aethods for ordinary ~iferntal equations
cannot be applied to this system of equations. This is beeause either· a nu:nerical instability

-3- H 3/1.
or a gross loss of ac.t.urac:.: result.s when the t.ime step of t.hese met. noc s gr eatl Y exc e~ 5 t :,~

shortest relaxation time, which is a necessity when the long t~e values, corespn~F :or
example to a 4(}-year service lifetime, are to be reached. The problem is coepletely avoided
by the so-called exponential algorithm [4,5,2J, which is based on an exact solution of the
differential equations for the Maxwell chaill under .the assumptior, that all parame!:ers are co","
stant within the time step. This, exponential algorithm ~ proven and demonstrated to be un-
conditionally stable, and highly accurate even for very,iong time steps. This allows increa-
sillg the time step roughly in a geometric progression, keeping it approximately constant in •
the log time scale.

4. Flow Chart of the Pr02ram

Instead of exhibiting the flow chart graphically, it is sufficient to describe it as
follows.
The program implements a step-by-step stable integration procedure combined with itera-
tions ill each time step. A typical time step consists of a two-cycle computation. In the
first cycle, estimates of the incremental quantities for the step are computed using the ma-
terial properties at the begillning of the time step. In the second cycle, the material pro-
perties are evaluated for the mid step based on the results of the previous cycle. The flow
of the program is outlined below (the subroutilles involved in each step of the program are
mentioned ill the parentheses).
1. Read and check the input data (INPUT, MATl'Alt).
2. Compute the material parameters for the Maxwell chain model (MATl'Alt, PRIt."Tl).
3. Initialize and set the step number to zero (MAINC).
4. Increment the step number and check whether the final step is completed. If it is, the
final output is printed and the program ends (MAINC, PRINT2)
5. For the current time step, the age of the concrete, applied external loads and tE!:!:pera-
. ture are determined and the it.eratiOn parameter is set. to zero (MAINe, CO:;STl).
6. The iteration parameter is illcrement.ed and the material properties are established based
on the current age of concrete (CONSTI, TEMBUH, DIFFUS, EMUF).
7. The instantaneous stiffness matru is established for each element (ASREEK).
8. The illcremental inelastic straills and the corresponding load vector are evaluate<! for
each element (ASREEO).
9. The structural linear equation system for the nodal displacement increments and the load
vector are assembled (MERGE).
10. Tbe prescribed boundary constraillts are imposed and the equation system is solved for the
incremental displacements (BOCOND, BANSOL).
11. If the iteration parameter equals one, the mid step stresses and strains are comput.ed and
tbe program returns to step 6 (ltETltCK, STRESS, ASREEP). If the iteration parameter e'l~as
two, program proceeds to step 12.
12. The incremental strains and stresses are computed, the total stresses, strains and dis~
placements are computed and selective out.put is printed. The program then returns to
step 4 (RETRCK, STRESS, PlUNT2).

The foregoing flow chart is, of course, routine, used ill many inelastic computer progra....

S. Determination of Creep Properties

The creep properties are determilled by the follovillg subroutille (~.ATPlt), calling for

-4- H 3/4
the subroutines aa specified.
1. Read the number of decades in log-time sCale to be considered, the number of steps per
decade, the number of elements in Maxwell chain, the time for the start of the first
time step and the first relaxation time.
2. Read the input option number (Sec. 6) and read the corresponding material data from
which the characterization of the compliance function is developed. In case the drying
is' specified, read also the characteristics for shrinIcage and drying creep.
3. Compute the discrete values of the relaxation function for various strain deviations
and various ages at the start of relaxation. (This consists in direct numerical soluCion
of a linear Volterra integral equation.) (RELAX).
4. Compute the discrete values of the Maxwell chain moduli at all discrete ages (MAXWLl).
5. As a check, compute discrete values of the compliance function from the discrete values
of Kaxvel.l chain moduli for various ages (CRCUIlV). Print the resulting values of the
compliance function and of their deviations from the initially specified values of the
compliance function. Also calculate and print the coefficient of variation of these
d....,iations.
Ouring the determination of the relaxation function, a subroutine for the compliance
function is repeatedly called. This subroutine has three options to be specified by theusee
1. The compliance function is evaluated by interpolation from a given array of discrete
values. The interpolation is linear in the logarithm of creep duration and in the loga-
rithm of age at loading. For durations and ages falling outside the range, extrapola-
tion is automatically used.
2. !~e compliance function is evaluated from a formula corresponding to the double power
1.0" [7].
3. If drying is specified, the compliance function is evaluated from a fOr<nula consisting
of the double power law enhanced by a drying term [7].
The ultimate result obtained in the subroutine MATPAR is the array of the discrete vari-
ables of the moduli of the Maxwell chain. For any specified age, the value of the Maxwell
chain moduli is computed from the discrete values (in subroutine function EMUF) by a linear
interpolation in the logarithm of the age of concrete. For argu~ents fal~ing outSide the
t:lllle range, a linear extrapolation from the extreme values 15 used.

6. InDut Options for Material Characterization

For convenience of the user, a number of input options are provided in the subroutine
KATPAII.:
1. The compliance function is specified as an array of discrete values for various load
durations and various ages at loading. ~o drying is considered.
2. Drying is considered, and the mean compliance function for th'e cross section is speci-
fied as in option 1. Also, the values of the mean shrinka~e strain of the cross section
are specified for various durations of drying and the g~ven a~e at the start of dryin~,

3. r~e cO::lplianc .. f"netion is give.'1 by t~e doubte po"er la'''' , for which all of its f i·:e
parameters are read. No drying is considered.
4. Same as option 3 but all double po"er la~ parameters except the 2B-dav elastic ~oduls

are generated from the given strengh'~d composition paraoeters of concrete (to ~ix
ratio of water, cement, sand, gravel, cement-type, and the unit ·"eight).

-5- Jot 3/t..

s. Same as option 4 except that the 28-<lay elastic modulus 1s also predicted from the
strength and composition parameters.
6. Drying is specified and the compliance function is defined by the double power law en-
hanced by the term for the creep increase due to drying [7J. A formula for the shrin_
kage strain development (accord ing to Ref. 7) is also spec if ied. Ane all parameters in
these formulas are read.
7. Same as option 6 but all material parameters excel't."the 28-day elastic modulus and the
final shrinkage strain are predicted from strengths and composition of concrete (and
also the given ambient humidity).
8. Same as 6 but all parameters except the "28-<lay elastic modulus are predicted frO!!! the
strengths and composition of concrete.
9. Same, as, option 6 but all parameters are predicted from the strength and composition of
concrete.
10. Two of the five double power la~ parameters (namely, Eo and ¢!' see Ref. 7) are deter-
mined so as to obtain the best fit (in the least sqUare sense) of the given array of
discrete values of the compliance function at various load durations and various ages.
The remaining three double power law parameters are given. This option is used when the
given arra: of discrete values for the compliance function is of limited range in time
and/or age. No drying is considered in this option. As a check, the coefficient of
variation for the deviations of the formula from the given array of values for the com-
'pliance function is automatically computed and printed.
11. Same as option 10 but the remaining three double power law parameters are not specified;
they are predicted from given strengths and composition of concrete.
12. Same as option 10 but drying is considered.
13. Same as opt ion 11 but drying is consideree.
The user selects his input option depending on the amount of information available to
him before the analysis. If sufficient test data have been obtained, as is frequently done
for nuclear concrete structures, tben the measured discrete values of the compliance function
should be'used (if the range of these variables is tOO limited, it is preferable to approxi-
mate these values by the double power law. If no experimental information is available, the
double power law parameters are predicted from composition. However, if there is some infor-
mation on the double law parameters for a similar concrete, an adjustment of some of its para-
meters is appropriate [11]. Moreover, if the short-time deformation (elastic modulus) 1&
measured, it should be also used to lIoprove the parameters of the double power law. The
c:hoice of the proper input option has a great effec:t on the accuracy in representing the
concrete properties.

7. Conclusion
It may be expec:ted that tbe ne~ computer program* will allow a rather effiCient, versa-
tile, and realistic analysis of the c:reep effects in nuclear concrete structures under 10n8-
time loading, and prescribes tecperature variations (however, without signific:ant drying
effeCts) •

*At the time of writ~g, the program described herein is still under developllent, although
rather near completion.

-6- H 3/4
Ac kno ..ledotment
Grateful appreciation is due to Oak Ridge National Laboratory, Oak Ridge, Tennessee,
operated by Union Carbide Corporation for the Department of Energy (subcontract 4OX-40434V).

ReferenceS"

1. ANDERSON, C. A., ''Numerical Creep Analysis of Structures," Los Alamos Scientific Labora-
tory Report LA-UR-8o-2585, Los Alamos, N.M., 1980.

B~;':"T, Z. P., "Theory of· Creep and Shrinkage in Concrete Structures: A Precis of Recent
Development," Mechanics Today, Pergamon Press, New York, Vol. 2, pp. 1-93, 1975,

3. SAZA.'."T, Z.P., WO, S.T., "Dirichlet Series Creep Function for Aging of Concrete," J. of
Eng. !-tech. Div., ASCE, Vol. 99, No. ~, pp. 367-387, April 1973; see also SMiRT -1-,---
Berlin, Vol. H, 1971.

4. !IAZA.'fI, Z.P., WO, S.T., "Thermoviscoelasticity of Aging Concrete," J. of Eng. Mech. Div.
ASCE. Vol. 100, No. EMJ, pp. 575-597, 1974; (also as ASCE Meeting Preprint No. 2110,
Oct. 1973).

5. BAZA::T, Z.P., WO, S. T., "Rate-Type Creep ·Law of Aging Concrete Based on !"Axwell Chain,"
Materials and Structures (RlLEM), Vol. 7, No. 37, pp. 45-60, 1974.

6. BAZA.'<"T, Z.P., ASCRAlU, A., "Computation of Age-Dependent Relaxation Spectra," ~

and Concrete Research, Vol. 4,·pp. 561-519, 1914.

7. SAZA:rr, Z.P., PANULA, L., "Practical Prediction of Time-Dependent Deformations of Con-

crete," Materials and Structures, (RILEM, Paris), Vol. 11, pp. 307-328, 415-434, 1918,
Vol. 12, pp. 169-183, 1979.

8. BAZA.'fI, Z.P., "Thermodynamics of Solidifying or Melting Viscoelastic Material," :L...2!

Eng. Mec!l. Oiv., ASCE, Vol. 105, No. EM6, Dec. 1979.

9. SHIra, P.O., COOK, W.A., ANDERSON, C;A., "Finite Element Analysis of Prestressed Con-
crete Reactor Vessels," Proc." 4th Conf. on Strucc. Mech. in Reactor Technology, San
Francisco, Calif., Aug. 1977.

10. SMITH, P.D., ANDERSON, C.A., ''NONSAP-G: A Nonlinear Stress Analysis Program for Concrete
Containments Onder StatiC, Dynamic and Long-Term Loadings, Los Alamos Scientific Labora-
tory, Report LA-7499-MS, Oct. 1978.

11. aAiANr, Z.P., PANULA, L., "Creep and Shrinkage Characterization for Analyzing Pre-
scressed Concrece Struccures," Prescressed Concrete Institute (PCl) Journal, Vol. 25,
pp. 86-122, 1980.

-7-