Rate-type creep law of aging concrete based on Maxwell chain



Similar documents
Long term performance of polymers

Numerical Analysis of Texas Cone Penetration Test

MATH 132: CALCULUS II SYLLABUS

Stress Strain Relationships

Introduction to time series analysis

Electrochemical Kinetics ( Ref. :Bard and Faulkner, Oldham and Myland, Liebhafsky and Cairns) R f = k f * C A (2) R b = k b * C B (3)

Vilnius University. Faculty of Mathematics and Informatics. Gintautas Bareikis

DETERMINATION OF TIME-TEMPERATURE SHIFT FACTOR FOR LONG-TERM LIFE PREDICTION OF POLYMER COMPOSITES

Lecture Notes to Accompany. Scientific Computing An Introductory Survey. by Michael T. Heath. Chapter 10

Incorporating Internal Gradient and Restricted Diffusion Effects in Nuclear Magnetic Resonance Log Interpretation

STRAIN-LIFE (e -N) APPROACH

1.- L a m e j o r o p c ió n e s c l o na r e l d i s co ( s e e x p li c a r á d es p u é s ).

CALCULATING THE RESPONSE OF AN OSCILLATOR TO ARBITRARY GROUND MOTION* By G~ORGE W. I-IouSl~ER

Forecasting Methods. What is forecasting? Why is forecasting important? How can we evaluate a future demand? How do we make mistakes?

Overview of Topics. Stress-Strain Behavior in Concrete. Elastic Behavior. Non-Linear Inelastic Behavior. Stress Distribution.

Algebra Unpacked Content For the new Common Core standards that will be effective in all North Carolina schools in the school year.

Figure 2.1: Center of mass of four points.

Stress Relaxation Study of Paper and Plastic Film based Packaging Material

Mesh Moving Techniques for Fluid-Structure Interactions With Large Displacements

In order to solve this problem it is first necessary to use Equation 5.5: x 2 Dt. = 1 erf. = 1.30, and x = 2 mm = m. Thus,

Engineering Problem Solving and Excel. EGN 1006 Introduction to Engineering

APPLIED MATHEMATICS ADVANCED LEVEL

Prentice Hall: Middle School Math, Course Correlated to: New York Mathematics Learning Standards (Intermediate)

A Strategy for Teaching Finite Element Analysis to Undergraduate Students

Lesson 3: Isothermal Hydrostatic Spheres. B68: a self-gravitating stable cloud. Hydrostatic self-gravitating spheres. P = "kt 2.

The Basics of FEA Procedure

Models for Product Demand Forecasting with the Use of Judgmental Adjustments to Statistical Forecasts

The Heat Equation. Lectures INF2320 p. 1/88

Integration of a fin experiment into the undergraduate heat transfer laboratory

In mathematics, there are four attainment targets: using and applying mathematics; number and algebra; shape, space and measures, and handling data.

Building a Smooth Yield Curve. University of Chicago. Jeff Greco

Demand Forecasting When a product is produced for a market, the demand occurs in the future. The production planning cannot be accomplished unless

TI-B 103 (94) Test Method Activation Energy for the Maturity Method

Mathematical Modeling and Engineering Problem Solving

4 Thermomechanical Analysis (TMA)

Using Excel (Microsoft Office 2007 Version) for Graphical Analysis of Data

Introduction to Error Analysis

ANALYSIS OF STRUCTURAL MEMBER SYSTEMS JEROME J. CONNOR NEW YORK : ':,:':,;:::::,,:

SIMULATION SOFTWARE AND NUMERICAL ISSUES. Alexander L. Pugh Massachusetts Institute of Technology, Cambridge, USA

Developing an ANSYS Creep Model for Polypropylene from Experimental Data

Structural Dynamics of Linear Elastic Single-Degree-of-Freedom (SDOF) Systems

Understanding Poles and Zeros

Technology of EHIS (stamping) applied to the automotive parts production

1 Error in Euler s Method

Systems with Persistent Memory: the Observation Inequality Problems and Solutions

Numerical Analysis of Independent Wire Strand Core (IWSC) Wire Rope

Review of Fundamental Mathematics

An important theme in this book is to give constructive definitions of mathematical objects. Thus, for instance, if you needed to evaluate.

NUMERICAL ANALYSIS OF THE EFFECTS OF WIND ON BUILDING STRUCTURES

Modelling of Short Term Interest Rate Based on Fractional Relaxation Equation

Hydraulics Prof. A. K. Sarma Department of Civil Engineering Indian Institute of Technology, Guwahati. Module No. # 02 Uniform Flow Lecture No.

Ultrasonic Detection Algorithm Research on the Damage Depth of Concrete after Fire Jiangtao Yu 1,a, Yuan Liu 1,b, Zhoudao Lu 1,c, Peng Zhao 2,d

Computer Graphics. Geometric Modeling. Page 1. Copyright Gotsman, Elber, Barequet, Karni, Sheffer Computer Science - Technion. An Example.

LABORATORY EXPERIMENTS TESTING OF MATERIALS

A Decision-Support System for New Product Sales Forecasting

Objectives. Experimentally determine the yield strength, tensile strength, and modules of elasticity and ductility of given materials.

2 Applications to Business and Economics

PART TWO GEOSYNTHETIC SOIL REINFORCEMENT. Martin Street Improvements, Fredonia, Wisconsin; Keystone Compac Hewnstone

Lymon C. Reese & Associates LCR&A Consulting Services Tests of Piles Under Axial Load

Stability of Evaporating Polymer Films. For: Dr. Roger Bonnecaze Surface Phenomena (ChE 385M)

INVESTIGATION OF VISCOELASTICITY AND CURE SHRINKAGE IN AN EPOXY RESIN DURING PROCESSING

General Theory of Differential Equations Sections 2.8, , 4.1

Three dimensional thermoset composite curing simulations involving heat conduction, cure kinetics, and viscoelastic stress strain response

Applications to Data Smoothing and Image Processing I

Numerical Simulation of CPT Tip Resistance in Layered Soil

Second Order Linear Nonhomogeneous Differential Equations; Method of Undetermined Coefficients. y + p(t) y + q(t) y = g(t), g(t) 0.

Measurement of Soil Parameters by Using Penetrometer Needle Apparatus

Deflection Calculation of RC Beams: Finite Element Software Versus Design Code Methods

Adequate Theory of Oscillator: A Prelude to Verification of Classical Mechanics Part 2

Continued Fractions and the Euclidean Algorithm

Technical Trading Rules as a Prior Knowledge to a Neural Networks Prediction System for the S&P 500 Index

SOFTWARE FOR GENERATION OF SPECTRUM COMPATIBLE TIME HISTORY

Fire-induced damage in prestressing steels

International Journal of Information Technology, Modeling and Computing (IJITMC) Vol.1, No.3,August 2013

Parametric Curves. (Com S 477/577 Notes) Yan-Bin Jia. Oct 8, 2015

In-situ Load Testing to Evaluate New Repair Techniques

Heat Transfer Prof. Dr. Aloke Kumar Ghosal Department of Chemical Engineering Indian Institute of Technology, Guwahati

CH 6: Fatigue Failure Resulting from Variable Loading

Industry Environment and Concepts for Forecasting 1

A matrix method for degree-raising of B-spline curves *

Politecnico di Torino. Porto Institutional Repository

Diablo Valley College Catalog

3-D WAVEGUIDE MODELING AND SIMULATION USING SBFEM

Should we Really Care about Building Business. Cycle Coincident Indexes!

DRAFT. Algebra 1 EOC Item Specifications

Tensile Testing Laboratory

Credit Number Lecture Lab / Shop Clinic / Co-op Hours. MAC 224 Advanced CNC Milling MAC 229 CNC Programming

METHODS FOR ACHIEVEMENT UNIFORM STRESSES DISTRIBUTION UNDER THE FOUNDATION

LOGNORMAL MODEL FOR STOCK PRICES

Computer Graphics CS 543 Lecture 12 (Part 1) Curves. Prof Emmanuel Agu. Computer Science Dept. Worcester Polytechnic Institute (WPI)

Example G Cost of construction of nuclear power plants

The Relation between Two Present Value Formulae

Numerical Methods for Differential Equations

4.3 Results Drained Conditions Undrained Conditions References Data Files Undrained Analysis of

This unit will lay the groundwork for later units where the students will extend this knowledge to quadratic and exponential functions.

Lesson 4 Annuities: The Mathematics of Regular Payments

The advantages and disadvantages of dynamic load testing and statnamic load testing

SIMPLIFIED PERFORMANCE MODEL FOR HYBRID WIND DIESEL SYSTEMS. J. F. MANWELL, J. G. McGOWAN and U. ABDULWAHID

Entry Level College Mathematics: Algebra or Modeling

Transcription:

Rate-type creep law of aging concrete based on Maxwell chain Z.P. BAtANT (1) S.T. WU (2) t is shown that the linear creep law of concrete can be characterized. with any desired accuracy by a rate-type creep law that can be interpreted by a Maxwell chain model of time-variable viscosities and spring moduli. dentification of these parameters from the test data is accomplished by expanding into Dirichlet series the relaxation curves which in turn are computed from the measured creep curves. The identification has a unique solution if a certain smoothing condition is imposed upon the relaxation spectra. The formulation is useful for the step-by-step time integration of larger finite element systems because it makes the storage of stress history unnecessary.. For this purpose a new unconditionally stable numerical algorithm is presented allowing an arbitrary increase of the time step as the creep rate decays. The rate-type formulation permits establishing a correlation with the rate processes in the microstructure and thus opens the way toward rational generalizations to variable temperature and water content. The previously developed Kelvin-type chain also permits such a correlation but its identification from test data is more complicated. BASC NOT A nons EN [fju t) = instantaneous Young's modulus of concrete; pseudo-instantaneous Young's modulus in Eq. (18); () Professor of Civil Engineering Northwestern University Evanston. 621 Active Member RLEM. (2) Graduate Research Assistant Nonhwestern University (presently Staff Engineer Sargent ~nd Lundy. Engineers Chicago.). J(t t') f f' f' to relaxation modulus ~ stress in time t caused by a constant unit strain enforced in time t' (Eq. (»; given data on ER; modulus of uth spring in Maxwell chai n fig. Eq. (3);.~ ultimate relaxation modulus fig. Eq. (6»; coefficients of smoothing expressions (9) (1) (); creep function (or compliance) = stress at time t caused by a constant unit stress acting since time t' (Eq. ); number of units in chain m = n - ; Maxwell time from casting of concrete (in days); - time of application of constant stress or strain; weights in the penalty term in Eq. (8); strain and stress; prescribed stress-independent inelastic strain (Eq. (» and pseudoinelastic strain in Eq. (18); 45

VOL. 7 - N 37-1974 - MAT~RAUX ET CONSTRUCTONS hidden stresses in Eq. (3) = stress in the ftth spring in Maxwell chain (fig. Eq. (3»; viscosity of the ftth dashpot Maxwell chain (fig. Eq. (3»: parameter given by Eq. (16): relaxation time of the ftth unit in Maxwell chain (fig. ). Subscripts: r. s - for discrete times '. s in step-hystep analysis: rxj for selected values of ' and ( - ') used in the least square condition: for the ftth unit in Maxwell chain (fig. ). Dot stands for time derivative; f = drtil. e.g. NTRODUCTON A rate-type creep law. i.e. a stress-strain relation in form of a differential equation. has two important advantages over the integral-type creep law involving hereditary integrals. First. whereas the integral-type law requires storing the complete history of stress or strain in the structure (which overtaxes. in the case of larger finite element systems. the capacity of rapid access memory of even the largest computers). the rate-type law necessitates only the current values of stresses strains. and a few hidden state variables to be stored. Second the rate-type law may be brought in correlation with the rate processes in the microstructure associating the hidden (or internal) state variables with some sort of microstrains or microstresses. This allows the dependence of various material parameters upon temperature and water content to be predicted to some degree from a mathematical theory of the creep mechanism [3] reducing thus considerably the degree- of arbitrariness in the identification of material parameters from experimental data. Recently [6] it has been shown that the stress-strain law of concrete viewed as a linear time-variable viscoelastic material can be approximated with any desired accuracy by a rate-type creep law based on the Kelvin chain model with age-dependent properties. This approximation has enabled formulation of a very efficient new algorithm of step-by-step time integration of creep problems [6]. Unfortunately. however. the Kelvin chain has some disadvantages_ The Kelvin-type formulations with such relations between age-dependent viscosities and spring moduli for which the identification of material parameters from the test data is simple either yield negative spring moduli for some periods of time or lead to non-standard relations for the dash pots [3] either of which precludes the correlation with the rate processes in the microstructure mentioned before. Only one Kelvin-type formulation namely that with proportionate age-dependence of viscosities and spring moduli is free from both drawbacks [3]. But it is more complicated and the identification of its parameters from test data is more involved. n classical viscoelasticity it is proved that any material can be described with any desired accuracy either by a Kelvin chain or by a Maxwell chain_ These chains are therefore mutually equivalent [22J (and also equivalent to any other possible model). t is natural to anticipate this property to hold also in the case of aging. Thus in view of the drawbacks of the Kelvin chain. attention is focused in this study on the Maxwell chain despite the fact that its relationship to the creep tests may be conceptually less convenient. CREEP LAW OF A LNEAR AGNG MATERALS f stresses remain less than about.4 of the strength and no strain reversals occur the creep law of concrete may be assumed as linear [19 3]. n the integral-type formulation the uniaxial creep law may be introduced in either of the following two equivalent forms [3] : r(l) - f o(l) -.1'1 J(t ') tla(l') a(l).1'1 E(1 1') [dr(l') - (h-o(t')] () where the integrals are Stieltjes integrals; 1 time from casting of concrete: rt stress: E = strain: f 1 () given stress-independent inelastic strain: J ( ') creep function (called also creep compliance) strain at time 1 caused by a constant unit stress acting since time '; Ell ( 1') = relaxation function (c~led also relaxation modulus) ~- stress at time caused by a constant unit strain introduced attime': J(.) ()~' (t)=-cinstantaneous Young's modulus (usually taken as J for - 1'::::. mn). The relation between functions J and E may be obtained by considering the strain history to be a unit step function that is e = for ;;. 1 and t' ceo for < 1 EO = in which case the response is. by definition rt(t) n(t () for 1 ;" u. Substitution into the first of Eq. () yields the Volterra's integral equation : ' OER(t'o) f' J( () (1) +.to J(t ) a-'- cl = (2) The rate-type formulation may be based on the Maxwell chain model (fig. ) which gives the stressstrain relations: rt ~ L a" F - F O o'-c rt' + a11 ' (t ~ 1.2... 11) (3) where rt arc the hidden stresses (or partial stresse) in the i~dividual Maxwell units of the chain. EXPANSON OF RELAXATON FUNCTON N DRCHLET SERES Similarly as in the case of Kelvin-type chains [5] a unique characterization of the model requires some relationship between " (t) and " (t) to be assumed. The simplest assumption is : i" " 11' c= const. (ft O~ 1.2... n) (4) 46

Z.P. BAZANT - S.T. WU (This ratio is called relaxation time.) Then considering again the strain history as a step function e ~-o for t <" t' and e = for t < ' (ell = ) the integral of the second Eq. (3) for the initial condition a" (') = E" (1') is (5) Comparison of the expression for a = L a. with () shows that " ER (1 t') = f. E" (t') e-(l-t')(r" + Eoo (') (6) 1=1 where 11 = 11 - ; Eoo (t') is written here under the assumption that for ft = n there is 11" -+ X or T" -+.. i.e. the last spring in figure is not coupled to any dashpot. Eq. (6) represents an expansion of the relaxation function into the series of real exponentials called Dirichlet series [14]. Fitting of curves by this series is a difficult mathematical problem because E" (t') are unstable functions of ER [17]. The problem is generally not tractable unless the T ' - of are suitably selected and certain smoothing conditions are applied. After much computing experience the procedure outlined below has been found to work satisfactorily. To ontain the Maxwell chain model lilting given relaxation data one can select a numher of t '-values (i.e. t' - t; oc _c 2 3... ; see figure 9 below) and expand ER (t t') as a function of t alone for each fixed (~ separately according to Eq. (6). As has been mentioned the T"values must be chosen in advance. With a sufficiently close distribution of T"values the relaxation curves can be fitted with any desired accuracy (provided the slope of the curve in log (t -- 1') scale is nowhere too large). For uniqueness of fit however the T"S may not be too close to each other (unless "'1 "'2 "'3 in Eq. (8) below are chosen very large). A pratically suitable choice is T. = (}H Tl (ft = J... m) (7) Here Tl should roughly coincide with the point where the relaxation curve begins to descend and T11 with the longest (t --. (') considered. (n comparison with the usual experimental error it even suffices to usc T" = 6" 1 T1 which leads to about- 1.5 times fewer T;'-S and a"s for the same time range to be covered (ef. fig. 19 below). The best method of fitting was found to be the method of least squares based on minimizing the expression _ 1 <l> = [ER(tp' ') - R(tri' 1')]2 +"'1 (E.1 - E)2 11=1 m-2 + "'2 (E+2-2 E"ll + EY '=1 m-a + 11'3 L (E+3-3 +2 + 3 E"q - E)2 (8)."= 1 Here E denotes the given values of the relaxation function 1 he fitted Eli stands L1r expression (6). The second sum represents a penalty term which is added as a smoothing device [6] intended to minimize the first second and third differences between 1 2 3... and also to reduce the sensitivity of " to the scatter of data. Suitable values for W W2 W3 can be assessed only by computing experience (see Table below). They should not be larger than necessary. (For very smooth data one can put W = W2 = "'3 =.) Subscript fj stands for a series of (t - ')-values. (t - f')ri' fj = 2... which are best chosen as uniformly distributed in log (t - ')-scale four values per decade (see fig. 9 below); lfi - t' = (t - n;'- o! (t - 1')1-1. The minimizing conditions are o<l>o " C~ (ft'. 2... ) and o<l>oe TJ ~~ which yields a system of (.- ) linear algebraic equations for E" and E.. Because the slope of relaxation curve is everywhere non-positive all E-5 are non-negative. Other types of Maxwell chain representation could be obtained if Eq. (4) is replaced by other relations between " and E" including their derivatives. But the above representation is certainly the simplest one and because it allows fitting the creep curves with any desired accuracy there is no need for examining other representations. (Such a need could possibly arise when creep at variable environment is considered. But the success in fitting the data at variable temperature in Ref. [7] suggests that this is probably not the case.) The computed values of l:.~ as functions of log. usually exhihit too much scatter (as is seen from fig. 11-17 and from the dependence of relaxation curves on t' in fig. 9 all discussed in the sequel. Smoothing with respect to age was therefore deemed to he appropriate. The expressions (' in days) : l:.~(n :! Eo! Elj[1 + 1'(3 3; )] (9) i= 1 E(t') : :'o! E" [log ( : nl' (1) ;::--1 were found best suited for this purpose (for day.' t' <' 1 days). Expression of the type (' in days) : E(1'). Eo" tel" (' H +- E2" t'l ~ 1 E3" (' 2 i E" (':l () (and also with other exponents) gives good fits too but usually not as e1se as Eq. (9) or (1). Hovewer. Eq. () seems to be best for extrapolation into very high ages t'. The constants Eo"... E3" are easily determined by the method of least squares. (Their values for various data are presented in Table below along with suitable values found for the coefficients 11'1 and»'2 of the penalty term in Eq. (8». A smooth change of E" with ' has been achieved by means of the weights W w2 and»'3 in Eq. (8). as has already been mentioned. (With W = W2 -'" the scatter of E" was sometimes so large that some of the E-values were even negative.) Changes in W and W:! can occasionally change E>values by as much as 1 "~ while the fit of the original creep data (fig. 2. 3 below) still remains quite good. This is. of course. within the range of dispersion of the original data hut indicates that for scattered data the solution of the fitting problem exhibits no strong uniqueness (even if T.-S are selected properly.) For the data in which J had to be extrapolated into large 1' different extrapolation curves have been examined to obtain the smoothest possible. ('). 47

VOL. 7 N 37 1974 MAT~RAUX ET CONSTRUCTONS DWORSHAK DAM. 1968 1.1 ---- NPUT --FT o. Fig. 1 1.'r--------:::n" ROSS DAM 1958 12.1... Q.... Q.!: 1..6.2.1.1.1.1 t - t' in days 15~ '"' 5 1 1 loco Fig. 2.. 6... Q.... o Q.!:.4.2. 1 1 1 t - to in days Fig. 3..8.6.. Go... 'Q.4.!: CANYON FERRY > DAM 1958 1 NE u "- -.x <D ~ Q. '" ' eft..... Q.S.6.4 SHASTA DAM 1958 ~ bo.~c" ~ ~'O 8 "'e u "-CJ.JC 6 OJ Q. <D '.2.2 1 1 1 t-t' in days 1 1 1 t - t' in days Fig. 4. Fig. 5. 48

Z.P. BAZANT S.T. WU en a.... ' c r--------------------------------~ DESGN DATA for WYLFA VESSEL 1971.:: 14. 9. 7.5 FT (SMOOTHED). t'= 7 days.. ' 28 ~... 6 '"... 18 "". 4 2 years.' 3".' 11.3 4 2.~.-J L~-----L~--~ 1~ 1 5 1 4 1 t - t' in days Fig. 6. 5 L' HERMTE et 1. 1965 4.3 en a. ~ ' " C 11.4.3. 2 -' ( E u... -'" 'f <D ' 3.2 loa '-.. QJ O'OYS a. (q"'3) " w 2 E u a.....~... 1... ' S a: W N.1 -' ox 1 1 1 t -t' in days Fig. 8. o ~~--~---L---L---L--~~O.1.1.1 1 1 1 1 t - t' in days Fig. 9. 49

VOL. 7 N )7 1974 MATblAUX ET CONSTRUCTONS TABLE Figure 2 1 3 9 4 13 5 14 6 15 7 16 8 17 11 18 12 Reference [14J [15J [15J [16) [15 ] [18) [19J llil--1!1] [16J Temperature 7 F _ 7 F _ 7 F _ 7 F 68 F --~~ 75 F - Humidity sealed sealed sealed sealed sealed 94% 1 in water i 28-day cy1. strength (psi) 28 497 292 323 66 49-56 Stress-strength ratio.21 to S.33.385 Water-cement ratio.56.56 Aggr.-cement ratio 6.8 Size of cylinder (inch) 6 x 18 6 x 16 Type of cement.s;.33 S.33 S.36.42 to.25.46.5.58.42.7?.49 4.4 5.1 6 x 16 6 x 26 6 x 12 4 x 1 & other V Type of aggr. granite - gneiss Max. aggr. size (inch) i 1.5 1.5 ---- _.1 forami- crushed "Hera 1 greywacke i limestone 1.5 1.5 1.5.19 -------.J CONVERSON OF CREEP DATA NTO RELAX ATON DATA The least square procedure described above permits direct determination of Maxwell chain parameters E" and 'YJ 1l from stress relaxation data. Although several relaxation data have been published in the literature [ 23 12 8 16 26 24 15] none of them suffice for determination of En (1 ). The range of elapsed times measured was sufficient in the tests of Rostasy ef al. [24] and compan'lon creep tests were also made but only a single age at loading ' was used. n such a case the fitting problem has no unique solution (). Therefore one must resort to creep test data converting them to relaxation data. This conversion has been discussed in detail elsewhere [] and only a : 'ief description will be given here. Relaxation i'unction ER(t t') must be solved from the Volterra's (1) Assume that only the data on J(t fo) and Eu (.to) for a single fa - value are available. h2 can be computed from Eq. (12) for ' = 3. But then one needs to choose some value for J4.2 to be able to compute J4.3 from Eq. (12) for ' 4. For r = 5 one must choose J52 and J5.3 to compute 1.;.4. etc. Thus it is always possible to find such J(t. f') that matches a given single creep curve and a single relaxation curve exactly. n fact. infinitely many such J(t t') exist. Consequently. the attempts [24. 16 121 to demonstrate the validity of the superposition principle on the basis of such incomplete data are of little value. They would be conclusive only if aging were absent or its effect were known in advance. equation (2). This is best accomplished numerically. subdividing time f by discrete times 11.... fs into N time steps t1lr = fr - fr-1 (r = 2 3... 'i). t is convenient to put 1 = flo expressing the fact that in creep tests with (J =~ constant the first strain increment in time fll is instantaneous i.e. ~t1 -- [f the trapezoidal rule is used to approximate the integral in Eq. (2) and the forms of Eq. (2) for = f. and r 1 arc subtracted the following recurrent equation (whose error order is il( 2 ) is obtained [2] :!:!.Elr== -. 1 (Jrr t J r.r-1) 1!:!.Eu s (Jr.. s.<=1 - J'" _.- J. 1.8 - J r 1." r) (r = 234... N) (12) where t1elr = EuU. fo) - EuU. -1 n) J."' J (tr s) and!:!.eu -c E(to)'=' starting value. The time step!:!.1 1 is best chosen as increasing in a constant ratio. With regard to the scatter of creep data the subdivision (' - ()-= O}(t. 1 -- fo) (i.e. 4 equal steps per decade in log (t - lo)-scale) was found to give sufficient accuracy. At the same time times ' chosen in this manner are suitable for use as times til in the least square condition (8). The first time step tj.f~ should be chosen smaller than the value of the elapsed time t - 1 for which the creep curve begins to rise significantly (and also smaller than T1 the shortest retardation time). For conversion of the creep function into the relaxation function it is necessary to. know J (t t') for any 5

Z.P. BAZANT - S.T. WU values of t' and - ' within the time range considered. When the information on J(f f') consists of a set of measured values interpolation must be used. For this purpose a FORTRAN V subroutine (listed for another purpose in Ref. [] has been written. t uses as input a two-dimensional array of J-values for selected discrete values of t' and (t - t') and assumes that between them the function J(f 1') varies linearly with log 1' as well as with log(t - 1'). Since this assllmption is dose to reality even for large time intervals relatively few values of J ( n are needed as input. n view of the large scatter of data on concrete the input values have been determined by smoothing visually the measured curves of J (f ') versus log ( - ') for various fixed f'. Because no creep data are known whose range of f' would match the range of(t - n an extrapolation into high values of ' has been necessary. t has been carried out in the plots of J(f f') versus log(t - f') by hand assuming a similar trend with l' as in the few known tests of old concrete [ 13]. This of course has introduced some degree of arbitrariness into the fitting problem. ALGORTHM OF STEP-BY-STEP NTEGRAT ON OF CREEP PROBLEMS f the usual step-by-step integration methods such as the Euler method Runge-Kutta methods or predictor-corrector methods are applied to Eq. (3) the time step cannot exceed the shortest relax<ltion time since otherwise numerical instability would occur. Such a restriction is practically unacceptable and a different algorithm must be developed. Let time f be subdivided by discrete times fo 11 f2... into time steps Afr = fr - fr-dr = 2... ). One can verify by back substitution that the integral of the second Eq. (3) satisfying the initial condition where e lid!) ' a. 1" elld!').1. "1 (t') [ddf')- tffl(t')] l (3) (13 a) For small time steps one may assume that clt-df dt: df and E are constant within each time step Af. and vary discontinuously at times f r. Furthermore using relation (4) E (t) _.. (t - f _})T. Putting f = fr Eq. (13) then becomes or ( 14) (jlr = alr- 1 e-atrit + A lr E llr - t (ilt:. - Af~) (ft =- 12... n) (15) where A lr = ( -e Atr'T') T!Y.f r (16) Here suhscript. stands for time r e.g. "r "1 (t.): and subscript s - 1- refers to the average value in the time step i.e. Elr--i = t { E lr- 1 + Elr }. n Eq. (15) onc has gained a recurrent formula for alr' According to Eq. (3) AUr = L D.u lr This yields ' ( 17) where E;:.l< C L E; = f. A lr E lr_ t + E"'.r-t (18) 1==1 { - e AlrlT } "r) E;'..f': (19) ' For generalizations to variable temperature (7] it is noteworthy that Eq. (15)-(19) also apply when E" ;- constant provided that T is replaced by Tlr! (llde)r-!' This follows from Eq. (13 a) assuming that fld E is constant within each time step. f in a given creep problem the stresses have already been computed up to the time f r - l the values E; and!y.f; may be determined from Eq. (18) and (19). Eq. (17) may then be regarded as a fictitious linear elastic stress-strain law in which D.t:; = pseudoinelastic strain increment and E; = pseudo-instantaneous elastic modulus. Thus creep structural analysis may proceed in each time step D.f as follows: ) Compute i'lr' E;' and.f ; for every point or finite element of the structure using the stress values for ' 1. 2) Solving the elastic structure with moduli E; compute ~ar and ". caused by D.f" (and the change of loads or boundary displacements during..llr) for every point or element. 3) Compute alr ({.7 2... ) for every point or element from Eq. (15) discard the values of a. t and alr and go to the next step..l.ll. The above algorithm can be easily generalized to the case of multiaxial stress writing analogous equations in terms of volumetric and deviatoric stress and strain components. To examine briefly whether the algorithm is prone to numerical instability attention may be restricted to the solution of stress produced by a prescribed strain history. Eq. (15) may then be viewed as a no1- homogeneous linear difference equation for the discrete variable a (' 2 3... ). The homogeneous part of Eq. (15) arising by deletion of the term with..f. has the solution of form "r ''''' a r whose insertion in Eq. (16) yields the characteristic equation a -"" e ~lrt. Since < a.< the solutions of the homogeneous part of Eq. (15) are of an exponentially decaying character and no numerical instability (unbounded magnification of numerical error) can occur. The nonhomogeneous equation (17) must then be numerically stable too. and noting that '. eeoc L "r it is seen that the same applies for (jr. (Excellent stability and accuracy of the algorithm has heen confirmed by computlltion of the fits in figures 2 and 3 below with time steps increasing in a geometric progression i.e constant in a logarithmic time scale.) (t is noteworthy that on the other hand the step-forward difference approximations of Eq. (3) are numerically unstable when!11 r exceeds the shortest relaxation time Tt). Note that i'!lr -+ for!y.frt1 -+ O. i'!l r -+ for..lr.ftl -+ - and always O i"'. Denoting by fj 'the value of for which TJ.'!Y.lr TJ- one may put i' llr for t -- 2... (p - ) pro- 51

VOL. 7 N 37 1974 MAT~RAUX ET CONSTRUCTONS 5r-------------------------------------------------------~ ROSS DAM. 1958.3 CD 'a... en.. c: 4 ~ 3 +> ~ W <D ' ~ CL> a. N E u... - -" 2 ( hour.1 1 1 t - t' in days Fig O. 2 DWORSHAK DAM.1968 Q. '" c:: w:t T~ in days Fig.. 52

Z.P. 8AZANT. S.T. Wu 3. ROSS DAM 1958 to Q c: ) 2 a. W'l... 1 4 Fig. 12. CANYON FERRY DAM 1958 Fig. JJ. 53

~OL 7 - N-17 - "" -... Tto.AUX.. CONSTRUCTONS SHASTA DAM 195a.5 T~ in days 1-1. '(1) 2 Q..S r WYlFA VESSEL. 197 Fig. 14.... r --- '.5.5 1'... in days Fig. 15 54

Z.P. BAZANT. 5.T. Wu -------- Q 1.5 ";;; 1. Q. c: -- O~ lu ~... GAMBLE a THOMASS 1969 "'...... ~--"'~~~. ~~---.. ' -t---~--." ~>«~ ~ " " " " ".5 T... in days r.5 Fig. 16. 4 L' HERMTE et a. 1965 T". in days Fig. 17. 55

VOL. 7. N 37 1974 MAT~RAUX ET CONSTRUCTONS TABLE -- 14E 14E 4 4 T 1 E2 1 E3 14E t' W w lj. 1 4 2 u lj. lj. lj. lj..5 -- 5.77 17.9 6.97.811 1.1.1 Fi s.2.5 2.33 23. - 9.66 1.18 1.1.1 ~" l =:':!.5 1.38 34.3 16. 2.6 1.1.1 5. - 2.6 46.9 24.2 3.21 1.1.1 [14) 5..755 48.9-25.8 3.49 1.1.1 Eq. 5. 6.16 31.9 (1)! - 8.86.869 5. 11.2-31. 9 27. - 4.8 CD 15.2-37.8 62.6-1.4 Fiss.3.5-115. 196. - 72.9 2.29 1-- ---_....5 1.5 81.6 147. 55.6 1. 79 1.2 Ref..5-76. 153. - 59.1 1.9 1.2 [15) 5. - 152. 291. -16. 3.i 1.2 [Hi) 5. - 143. 273. 86.2 1.53 1.2 Eq. 5. -12. 17. - 33.2.272 (t) i 5. - 15. 112. 7.21 1.5 f-- F.- ig :- 4 - ~- -112 85 4 28 1-1.36.5 21.3.65( 1.61.123 1.1.1 Ref..5 4.76 39.8 14.2.486 1.1.1 [15 ).5 31.4 93. 31.1.873 1.1.1 [17) 5. - 25.8 19. 34.4.647 1.1.1 [16 ) 5. 12.8 59.4 13.3.389 1.1.1 Eq. 5. - 17.3 75.8-6.94 -.75 (ll) 5. -138. 188. - 25.7.41 r CD - 265. 314. 48.9.98!.5 77.5-38.6 1.9.67 1.5 Fig. 5.5 39. 1.75 1.8.664 r 1.2.2 Ref..5-15.7 69.3-28.4.827 1.2 [S].2 5. - 3.6 96.6-33.7.633 1.2.2 [16 ] 5. 1.1 5.1 14.3.113 1.2.2 Eq. 5. - 11. 9 56.5 4.22.479 (11) 5. -12. lh. - 4.27.582 CD -132. _!_1. 21.4-1._~ Fig: (.5 69.4-36.7 6.57 -.353 1.1 U Ref. [18J.5 52.6-28.1 4.9 -.259 1.1.5 23.1 4.82-2.22.15 1.1 5. 17.7 66.2-14.4.816 1 ;1 5. - 28.1 85.5-15.2.783 1.1 Eq. 5. 47. 92.8-14.7.742 (11) 5. - 3.3 59.3-7.5.351 '" 47. - "LR 1 9.727 Fig. 7:.5 29.3 77.4 22.5 1.1! Ref. J.5 31.4 72.6-21.9! 1.1 [191 i.5-63. 119. 43.1 1.3 5-12 27. - 75.8 Eq. ; 5-13 167. 44.2 (11)! 5. -13. 153. - 8.9 '" -19. 156. 8.56 r Fig. 8 r.5 56.2 11.7-4.21.266 1.1.5 34.7 16. - 4.61.285 1.1 Ref..5 22.4 17. 3.85.243! 1.1 [2J 5. 29.7 9.29 -.141 -.24 1.1 5. 33.1 5.91 3.78 -.441! 1.1 Eq. i 5. 11.1 32.4 -.39 -.236 i (11) 5. 2.59 a 55.8-4.38 -.17 67.1-6.26 X).783.393 i t' are 1n days; ElJ. in Eqs. (1) (11) is in psi. Tl.' : : --'! vided that T = lohio. Then according to (18) and (19) m E;~ = L Alr Efr-t + Er.r-l: (2) 1'::::: P jj-l n E; Ue; = (lr-1 + L { 1 - rtl.trt' } (lr-1 1'=1 "=P + E; Ue? (21) (As a crude approximation. one may further put e tl.trt' = and A lr Co; 1 for ft > P +.) t is seen that those Maxwell units in the chain in figure whose relaxation times Tl are substantially less than ut r pose no appreciable resistance to the deformation and drop out of action. This of course may also be expected on the basis of an intuitive judgment. 56

A Z.P. BAZANT - S.T. WU.;~ k j(.... f. t ~E(T..=.95 days).... ok J 1. J'---'jt-...:~xrE-Z"'-:-T"":-=""'2""5""""") --;ll1---. ---"':f-..'ly'--"k -<y-:.5.2~.~ x j( ~.. ~ E..:.3_("!~'::"-=""' 15_) :::-=-~x~:x~:~!(~=:-«2..~ )( E7 (T... 5 days) x x x XX~ E.(~ y PC. x 1. E.. (-z;.. =) ~ j(.5 A : " t x DWORSHAK DAM 1968 ;. E. (<.. oo d.y).o.l<- --.J"-- -L- --... ---' 1 1 2 1 3 1 1 1 2 d t' in days t' in days Fig. 18. The algorithm presented above is analogous to the algorithm for the Kelvin chain model with timedependent properties. published in Ref. [4] (Eq. 7-79). A similar algorithm for Kelvin chains which has been developed earlier by Zienkiewicz et al. [27] does not apply for concrete because the relation considered for springs «(f =." El f) is thermodynamically unacceptable when E. increases with age. Another similar algorithm for Maxwell chains. applicable to the special case of time-invariable viscoelastic materials and thennorheologically simple materials has been presented by Taylor ef al. [25]. ORGANZATON OF THE PROGRAM FOR MAXWELL CHAN DENTFCATON FROM GVEN DATA. Values of J(t. 1') for selected values of t and t' are read. (f the given creep data do not cover the whole desired time range in both f and f-f' intuitive graphical extrapolation from given data must be done beforehand.) 2. Calling the interpolation subroutine for J(t t') values the stress relaxation curves are computed from Eq. (12) for selected '-values such as t'- 1 1 1 1 days (as in fig. 9 below). )n the rare cases that the relaxation curves for the same specimens are available for some f'-values the relaxation curves for these t'-values are also computed from the creep data and a weighted average of both curves is then taken. 3. Each relaxation curve is expanded in Dirichlet series (6). calling the least-square subroutine based on Eq. (8). 4. Moduli E as functions of t' are smoothed by expression (1) or (9) calling another least square subroutine. 5. Stress relaxation curves are evaluated from E (t') and plotted on Calcomp plotter to be compared visually with the relaxation data (if available). 6. Creep curves are computed from 11 (t') using Eq. (17) with Lla. ~ and Eq. (15). Then they are plotted on Calcomp plotter and compared visually with the given creep data. )f the fits of the given creep data as well as of the relaxation data are unsatisfactory the program is run again for a different extrapolation of creep data and for different weights W 11'2. W3 in Eq. (8) After some experience. however. repeated runs hae usually been unnecessary. NUMERCAL RESULTS AND FTS OF THE BEST A V ALABLE DATA All of the extensive data that could be found in the literature on concrete at constant temperature and nearly constant water content [9 - - 13 - ~ - 19-21] 57

VOL. 7 N 37 1974 MATERAUX ET CONSTRUCTONS were analyzed (using computer CDC-64) with the above program (written in FORTRAN V). The data points are shown in figures 2-8 and 1 and further information on the data used (including the references) is compiled in Table. The solid lines in figures 2-8 represent the fits by Maxwell chain model computed from Ell as defined in Table. Figure 1 shows the fit of the relaxation data measured on the same specimens as the creep data from figure 3. (The fit in figure 1 may be regarded as a confirmation of the superposition principle because the creep function was characterized almost uniquely by the creep data in figure 3.) For other data relaxation measurements are not available. As is seen all of the fits are quite satisfactory. Most of the data had to be plotted in the log (t - 1') - scale because the actual time scale obscures the misfit for all but one order of m~gnitude of time. (Unfortunately many hypotheses have been supported in the literature by fits in the actual time scale. Seemingly good fits are then easily obtained. But this practice is misleading and usually in the logarithmic time scale such fits would appear as unacceptable.) Whenever possible the replotting of data points from the original papers was done with the help of a table of values [4 24 8]. Whereas all data points represent test results the data in figure 6 as an exception represent the design curves recommended for prestressed concrete reactor vessels. n figures 3-8 the values of the instantaneous elastic modulus had to be judiciously estimated for lack of information. The relaxation data in figure 1 were originally presented as the ratio a (t )a(to) where a (to) was the stress a short delay to - t' after time t' of strain enforcement; a(to) was 4 psi (for both t' shown) but the time lag to - t' was not reported. t was assumed as one hour and taking the E-modulus for one hour lag ER-values were computed from the original data points and plotted in figure 9. (f this assumption were inaccurate the data points in figure 9 would all be shifted equal distance up or down). The values obtained for the spring moduli Ell of the Maxwell chain model (before smoothing by expressions (1) or (9» are plotted in figures 11-17 against Til and log t'. Each series of E-values for a fixed l' corresponds to what is called relaxation spectrum in classical viscoelasticity. The diagrams in figures 11-17 thus represent the relaxation spectrum as a function of age at loading t'. Note that the surfaces formed by the relaxation spectra have similar forms for different concretes. Figure 18 is an example of the fit by expression (1) (solid lines) with the Ep-values from Table. To check the quality of fit in detail the relaxation curves have been computed in this case for four uniformly spaced t '-values per decade as is seen from figure 18. Such dense point values of Ell were necessary to decide which of expressions (1) (9) or ( ) was the best one. Without using the smoothing term with W and W2 in Eq. (18) the sequence of data points would scatter much more than in figure 18. R ;... ' c.....;.. 1.4 1..6 ) DWORSHAK DAM SS6.2UL ~ ~ ~ ~ ~ ~~.1.1.1.1 1 1 1 t - t' in days Fig. 19. 25 i5 1 5 -C ~ ~ cd ' Function J (t (') representing the unit creep curves for various t' in figure 2-8 was computed from the expression for Ell (t') given in Table by means of the algorithm defined by Eq. (15)-(19). Four equal steps per decade in the logarithmic time scale were used. t is noted that the fits shown by solid lines in figures 2-R could have been even closer to the original data points if the various smoothing operations had not been applied. The smoothing is inevitable however. due to the high scatter of all data with regard to ('. (This scatter can be best demonstrated in plots of f' versus t' at constant (.) The solid lines in figure 19 show how the same data as in figure 2 are fitted when more sparsely spaced TlrS are chosen namely Til =.5. 5111. As compared with the experimental error the fit is still acceptable though.. bumpy". A crucial assumption in the present analysis is the principle of superposition which is applicable when (a) t he stress is less than about.5 of strength so that no significant microcracking occurs and (b) the water content is constant [3 5]. The principal experimental data in its support. are in addition to those in figure 1. the data of Ross [23]. n figure 2 they are replotted in log-time scale. Based on the smoothing of creep data by dashed-lines shown in figure 2 the stress relaxation curve given by a solid line was computed on the basis of the superposition principle. n spite of the fact that specimens in these tests were mildly drying (at 91 :' relative humidity) the relaxation data conform well with the principle of superposition in fact even better than Ross found originally [23] without the smoothing in log-time and by computations which in 1958 could not have been made accurately because electronic computers were unavailable. (Fig. 2 also shows the approximate predictions by the effective modulus method and the rate-of-creep method as given in Ref. [23]. 58

Z.P. BAZANT - S.T. WU ROSS.1956 4 x x RELAXATON TESTS + for E =.36 (CT(.1l-2p (CT(O.OOil"" ~oo p;).3 D '... a. '".~ superposit ion rote -of-creep 2 ~ a. N E u... -C'.:.t: L 1 1 1 1 t - to in days Fig. 2. t - t' in days CONCLUSONS. The linear creep behavior of concrete as an aging linearly viscoelastic material can be approximated with any desired accuracy by the Maxwell chain model (fig. ) with age-dependent properties. 2. This model permits close and smoothed fits of test data as figures 2-8 demonstrate. 3. The Maxwell chain representation can be obtained by expanding the relaxation function in Dirichlet series with age-dependent coefficients (Eq. 6) which is best accomplished by the method of least squares with a penalty term on first to third d~fferences of coefficients as a smoothing devise (Eq. 8). The relaxation function needed may be computed from the creep data. The program outlined identifies the material parameters from any given test data very economically (in less than 15 sec. CPU time on CDC-64oo). 4. Since conclusion has previously been made [3] also for the Kelvin chain both types of chains are mutually equivalent even for aging materials. But the Maxwell chain is more convenient for identification of material parameters from test data. (Both types of chains are also equivalent to any other possible model.) 5. The Maxwell chain representation proposed guarantees positiveness of all spring moduli 11 and dashpot viscositie~ 1111 (if scatter of the data is not excessive). This permits giving the individual units of the chain a physical interpretation associating them with various diffusion processes in the microstructure (see [4]) and opens the way toward rational generalizations to variable temperature [7. 3] and water content [5 3]. 6. For the Maxwell chain representation a new algorithm for creep structural analysis is presented (Eq. 15-19). t permits an arbitrary increase of the time step and uses only a few internal state variables (hidden stresses) for characterization of the whole history of the material. The resulting substantial reduction of machine time and storage requirements is valuable in the case of large finite element systems. ACKNOWLEDGMENT The present results have been achieved under the U.S. National Foundation Grant GK-263. The authors wish to thank Mr. M. Mamillan C.E.B.T.P. Paris for providing a numerical table of the creep data in figure 8. RESUME Vne loi de ftuage de type differentiel d'apres UDe chaine de Maxwell et relative au betod en coors de vieillis~ement. - On montre que la loi de.ffua~e ineaire du hholl pt'llt etrc caracth i.'('e ar('c tollle la preci.ioll 1'111111 p:'r iii(' loi de.lllage de ype iliprellliel qu'on pelll illlerpreter par un modele de Maxwell ell chaine comhinanl les riscosile.l en rmclioll dll lemps el des modulcs de ressorl. On idenlitie ces parametres d'apre.l les rhullals d'essai en dereloppanl en series de Dirichlel les courbes de relaxalion qui sont elles-memes calculees d'apres les courbes de ftuage experimentales. L'identification ne comporte qu'une seule solution ~i une certaine condition de regularisation est imposee aux.pectres de reaxatiol1. La formulaliol1 est importante lorsqu'on prochle a une integration graduele dans e temps e systhnes a ethnent5.tin is plus grand~ car aimi il n'est pas besoill de slocker les donnees de 'erolufion des colllrailltes. A cefte fin on propose un nolll'eau alf!.orilhllc H 'riqlie illconditionllelel1lellt.labe qlli 1'(''11 '1 1111 accroi..'(' '1l arllitraire tic 'illl{'' '( ell' telllps a mel'll'e que a l'ilesl"e de.fua~(' (Mcroil. La fljrfllulatiol de t11u! d(tjerentiel permet d'etablir line correlation al' '( les processus de flolll'emellt au nil'eau de la microstl'llctllr ' et mene ainsi re's des geu! ralisations rationnelles a temperatures et teneurs ell eau rariables. La chaille de ype KeMn prechlemmellt etudih: pennet elle allssi line telle correlation mais 'ident(fication des resultats d'e5sai est alors compfiquee. 59

VOL. 7 - N 37-1974 - MATERAUX ET CONSTRUCTONS REFERENCES [] BAZANT Z.P. - Numerical determination of longrange stress history from strain history in concrete. Matcrials and Strcutures (RLEM) Vol. 5 135-141 1972. [2J BAZANT Z.P. - Prediction of creep elf 'ct. sing the age-adjllsted ellecth'e modllllls method. J. Amcr. Concrete nst. Vol. 69 212-217 April 1972. [3] BAZANT Z.P. - Theory o{ creep and shrinkage ill concrete structllre : A precis of recent de~ elopments. Mechanics Today Vol. 2 Pergamon Press (in press). [4J BAzANT Z.P. - Thermodynamics of interacting continua with surfaces and creep analysis o{ concrete structures. Nuclear Engineering and Design Vol. 2 477-55 1972. [5J BAZANT Z.P. Wu S.T. - Creep and shrillkaf.{e tl' for concrete at variable humidity. J. Eng. Mech. Div. Am. Soc. of Civil Engrs. (under review). [6J BAZANT Z.P. Wu S.T. - Dirichlet series creep {lnction for af.{ing concrete. Proc. ASCE J. Eng. Mech. Div. Vol. 99 EM2 Apr. 1973 367-387. [7J BAZANT Z.P. Wu S.T. - Thermol'iscoelasticity of' aging concrete sent to Proc. ASCE J. Eng. Mech. Div.; and Pre print No. 211 ASCE Annual Meeting New York Oct.-Nov. 1973. [8] BEREs L. - La macrostructure ef e comportement du beton SOilS {'effet de solicitations de longlle dlm:e. Materials and Structures (RLEM) Vol. 2 p. 13-11 1969. [9] BROWNE R.D. BURROW R.E.D. - An example of' the utilization of the complex muitiphase material behavior in engineering design. nt. Conf. on.. Structure Solid Mechanics and Engineering Design in Civil Engineering " p. 1343-1378 Editor M. Te'eni J. Wiley - nterscience 1971 (Univ. of Southampton April 1969). [1] GAMBLE B.R. THOMAS L.H. - The creep ofmatllring concrete subjected to time 'aryin!! stress. Proceedings of the 2nd Australasian Conf. on the Mechanics of Structures and Materials Adelaide Australia Aug. 1969 Paper 24. [] HANSON J.A. - A 1-year study of creep properties of concrete. Concrete La boratory Report No. Sp-38 U.S. Department of the nterior Bureau of Reclamation Denver Colorado July 1953. [12] HANSEN T.e. - Estimating stress relaxatioll from creep data. Materials Research and Standal ds (ASTM) Vol. 4 1964 12-14. [13] HARBOE E. M. et al. - A comparison of'the instantaneolls alld the slstained modlls o(elasticity olcollcrete. Concrete Laboratory Rcport No. C-1l54 Division of Engincering Laboratories U.S. Department of the nterior Bureau of Reclamation Denvcl Colorado March 1958. [14] HARDY G.M. RESZ M. - The f.{eneral theory of Diricltlef series (Cambridge Tracts in Math. & Math. Phys. No. 18) Cambridgc Univ. Prcss 1915. [15] K~.NNE()Y T. W. PERRY E.S. -' An experilllenwl approaclr to tire stlldy ol tire creep helwrior of' plain colcrete sllhjecled to triuxi"f stresses alld elnared temperatllres. Res. Report 2864-1 Dept. of Civil Engineering Univ. of Texas Austin 197. [16] KLUG P. WTTMANN F. - Tire correfarion hetll'n creep deformation and stress relaxation in collcrete. Materials and Structures (RLEM) Vol. 3. 75-11 197. [17] LANCZOS C. - Applied Analysis Prenticc Hall 1964 (p. 272-28). [18] L'HERMTE R. MACMLLAN M. LEFEVRE C. - NO 'NX resllltats de recherches sllr la deforlllatioll 't la rupture dll beton. Annales de 'nstitut Techniquc du Biitiment et des Travaux Publics Vol. 18 No. 27-21l p. 325-36 1965. [19] NEVLLE A.M. - Creep (Jlconcrete: plain reinijrced prestressed North Holland Publ. Co. Amsterdam 197. [2] NEVLLE A.M. - Properties o{ concrete. J. Wi cy New York 1963. [21] PRTZ D. - Creep characteristics or mass cocrete.iijr Dll'Orslwk Dam. Report No. 65-2 Structural Engineering Laboratory Univ. of California Berkcley California Oct. 1968. [22] ROSCOE R. - Mechanical models for the represelllation oll'iscoelastic properties. British J. of Applied Physics Vol. 195 171-173. [23] Ross A.D. - Creep of concrete nder 'ariahle stress. Amer. Concrete nst. Journal Proc. Vol. 54 739-758 1958. [24] ROSTASY F.S. TECHEN K.T. ENGELK~. H. ReitraJ: Zlr Kliirung des ZSallltnenhall!!('S 1' Kriechen lind Relaxation bei Normalbeton. Bericht Otto-Graf-nstitut Universitat Stuttgart Strassenbau und Strasscnverkehrstechnik Helf 139 1972. [25] TAYLOR R.L. PSTER K.S. GOUDREAU G.L. Therlllolechanical analysis (f 'iscoeiastic solids. ntern. J. for Numerical Methods in Engineering Vol. 2 45-6 197. [26] WTTMAN F. Uher den Zsalllmenlrang 1' Kriechl'er.iilYll!! lind Spalllllllg.'Yelaxatioll des hetol.l'. Betonund Stahlbctonbau Vol. 66 p. 63-65. [27] Z.NKEWCZ O.c. WATSON M. KNO LP. - A nlllllerical method or riscoelastic stress analysis. ntern. J. Mech. Sciences Vol. 1 87-827. 1968. 6