'DO NOT RETURN TO LIB R A R /
EVALUATION AND ANALYSES OF SOME FINITE ELEMENT AND FINITE DIFFERENCE PROCEDURES FOR TIME-DEPENDENT PROBLEMS
by
C. S. Desai, J. T. Oden, L. D. Johnson
Soils and Pavements Laboratory U. S. Army Engineer Waterways Experiment Station P. O. Box 631, Vicksburg, Miss. 39180
April 1975 Final Report
Approved For Public Release; Distribution Unlimited
TA Prepared for Office, Chief of Engineers, U. S. Army 7 Washington, D. C. 2 0 3 14 .W34m S-75-7 1975 PROPERTY OF SUREAU OF RECLAMATION BUAEA.y..?,ÌlECLAMAT,0N DENVER LIBRARY -TA 92031039 Unclassified ^ 5 ^ 0 , \ SECURITY CLASSIFICATION OF THIS PAGE (When Data Enter'd) s -7^7 BEFORE COMPLETING FORM 2. GOVT ACCESSION NO. 3. RECIPIENT’S CATALOG NUMBER h-?v 1. R E P O R T N U M B ER Miscellaneous Paper S-75-7 5. TYPE OF REPORT & PERIOD COVERED 4. T IT L E (end Subtitle) Final report EVALUATION AND ANALYSES OF SOME F] [NITE ELEMENT AND FINITE DIFFERENCE PROCEDURES I PO R T I M E - 6. PERFORMING ORG. REPORT NUMBER DEPENDENT PROBLEMS
8. CONTRACT OR GRANT NUMBERfe) 7. AUTHOR^; C. S. Desai J. T. Oden’
L . D . Johnson 10. PROGRAM ELEMENT, PROJECT, TASK 9. PERFORMING ORGANIZATION NAME AND AUU k ESS . AREA & WORK UNIT NUMBERS U. S. Army Engineer Waterways Experiment Station Soils and Pavements Laboratory P. 0. Box 631, Vicksburg, Miss. 39180 12. R E P O R T D A T E 11. CONTROLLING OFFICE NAME AND ADDRESS Office, Chief of Engineers, U . S. Army April 1975 Washington, D, C. 2031^- 13. N U M BER O F PA G ES 107 15. S E C U R IT Y CLASS, (of this report) 14. M O N ITO R IN G AG EN C Y NA M E & ADDRESS(7f different from Controlling Office) Unclassified
15a DECLASSIFICATION/DOWNGRADING SCHEDULE
16. D IS T R IB U T IO N S T A T E M E N T (of thia Report) Approved for public release; distribution unlimited.
17. D IS T R IB U T IO N S T A TE M E N T' (at the abstract entered ,n Block 20, It dltterent Iron, Report)
18. SUPPLEMENTARY NOTES
19. K E Y W O RDS (Continue on reverse side if necessary and identify by block number) Finite element method Finite difference method Time dependence Soil mechanics
------” . , r t ______nrrsst } riser, Hfv hv h lo c k num ber) A large number of scnemes using une i u u w --- - o (FD) methods have been proposed. Numerical and physical properties of these schemes, vital for their use as general procedures and significant from the viewpoint of the user, are often not examined adequately. One of the aims f this study was to examine numerical properties of a selected^number of schemes for problems relevant to geotechnical engineering. In this initial phase, the aim has been partially fulfilled by studying the problem of one-dimensional
EDITION OF t NOV 65 IS OBSOLETE DD 1473 Unclassified ______SECURITY CLASSIFICATION OF THIS PAGE < ™ « > uera Entered) Unclassified. SECURITY CLASSIFICATION OF THIS P»GE(TW.«. D.lm Bnt.fdi
20. ABSTRACT (Continued).
consolidation and wave propagation. Another aim of the research was to examine the concept of space-time finite elements for some time-dependent problems. A portion of Section I involves quantitative analyses of such factors as con- sta^ i t y , and computational efforts for a number of FD and FE scheme* for the consolidation problem governed by a linear parabolic equation Also examined are the effects of the order of the approximation models and heavier i ^ s o l u t i o n at different locations in space and in time. Mathematical stability criteria are also derived for two time integration schemes with two approximation models for the linear parabolic equation. In Section II, Prof. . T. Oden has presented mathematical derivations for error estimates of FE ap proximations to the diffusion equation. The numerical characteristics have been examined and error estimates have been obtained by using semigroup theo retic, energy, and L? methods. The work presented in this section is father fundamental in nature~and can provide basis for needed investigations towards numerical analysis of FE procedures. Part V of Section I contains formulations of the concept of space-time finite elements for dynamical systems, and for owo equations of fluid flow through porous media. A simple example of the srngle-degree-of-freedom system is solved quantitatively. Since the proiect had to be postponed, no quantitative analyses and definitive conclusions were derived for the concept of space-time elements. The topic of evaluation of different numerical schemes is of vital importance. This is particularly so because a large number of schemes have often been proposed without adequate guidelines for the user to assist hit, in selectinglhe o p t i ™ scheme his specific needs. It is felt that the work initiated in this study should be analysed^and^ev^!uatedf0r <* P ™ « - ! ¡«Prance he
- ______Unclassified______SECURITY CLASSIFICATION OF THIS PAGE(Tf7ien Data Entered) PREFACE
The study of the concept of finite elements in the time domain and evaluation of some finite element (FE) and finite difference (FD) schemes was initiated by the Soils and Pavements Laboratory, U. S. Army Engineer Waterways Experiment Station (WES), Vicksburg, Mississippi. This work was sponsored by the Research in Military Engineering and Con struction (RMEC) group of the Office, Chief of Engineers, U. S. Army. The investigations described in Section I were performed by Dr. C. S. Desai, Research Group, Soil Mechanics Division, now with the Department of Civil Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Va. Prof. J. T. Oden, consultant to the project, University of Texas, Austin, Tex., contributed Section II, and provided valuable advice and suggestions. Dr. L. D. Johnson, Research Group, Soil Mechanics Division, assisted in computational work related to the analyses of one-dimensional consolidation problems. Prof. R. L. Lytton, Texas A&M University, collaborated on the derivation pre sented in Part IV. Section I of this report was prepared by Dr. C. S. Desai and Section II was prepared by Prof. J. T. Oden. The project was postponed after completion of the initial phase described herein. During the investigation, Mr. J. P. Sale was the Chief, Soils and Pavements Laboratory, and Mr. C. L. McAnear was Chief, Soil Mechan ics Divison. Directors of the WES during the investigation were BG E. D. Peixotto, CE, and COL G. H. Hilt, CE. Mr. F. R. Brown was Technical Director.
iii
CONTENTS Page iii PREFACE ...... 1 SECTION I ...... • • PART I: INTRODUCTION...... 3 PART II: EFFECT OF ORDER OF APPROXIMATION MODELS ...... 5 Finite Element Formulation...... 5 Numerical Characteristics and Comparisons ...... 13 Applications...... 22 Comments...... 31 PART III: EVALUATION OF FINITE ELEMENT AND FINITE DIFFERENCE SCHEMES...... 32 Finite Difference Schemes ...... 32 Finite Element Method ...... 33 Comparisons ...... 33 Illustrative Example...... 1+0 Comments...... 1+1 PART IV: DERIVATION OF STABILITY CRITERIA FOR TWO TIME INTEGRATION SCHEMES FOR LINEAR PARABOLIC EQUATION .... 1+3
Linear Model...... 1+3 1+ 1+ Cubic M o d e l ...... Discussion...... 1+7 PART V: FINITE ELEMENTS IN THE TIME DOMAIN ...... 1+8 Dynamical System...... 1+9 Application ...... 52 Fluid Flow Problems ...... 51+ 59 SECTION I I ...... PART VI: SOME ASPECTS OF THE THEORY OF FINITE ELEMENT APPROXIMATIONS OF THE DIFFUSION EQUATION...... 61 6l Introduction...... • • ...... Finite Element Approximation and Interpolation. . .^...... 6 3 Finite Element Approximation of the Diffusion Equation. . . . 68 Error Estimates for the Diffusion Equation Using Semigroup Theoretic Results ...... 70
v Error Estimates for the Diffusion Equation Using the Energy Method ...... 83 Error Estimates for the Diffusion Equation Using L2 M e t h o d s ...... 87 PART VII: CONCLUSIONS AND RECOMMENDATIONS ...... 97 REFERENCES ...... 98 TABLES.1-5
vi SECTION I
by
C. S. Desai, L. D. Johnson
EVALUATION AND ANALYSES OF SOME FINITE ELEMENT AND FINITE DIFFERENCE PROCEDURES FOR TIME-DEPENDENT PROBLEMS
PART I: INTRODUCTION
1. A large number of finite element (FE) and finite difference (FD) formulations has been proposed for solution of engineering problems The question facing the engineer who intends to use these methods is which procedure will provide sufficient accuracy most economically. A number of factors, such as numerical characteristics of the procedures, discretization schemes, physical requirements, geometry and properties of the media, order of approximation, and relative cost can influence selection of the proper scheme. The question has not been answered ade- 1 8 quately; only a few studies have been reported. 2. As a step toward the foregoing aim, eight FD and FE schemes were examined quantitatively from the viewpoints of convergence, sta bility, and computational times. The example of one-dimensional consoli dation governed by a parabolic equation was considered. Effects of such factors as spatial and temporal locations on the numerical solutions and of approximating models were also studied. 3. Often the user tends to establish the reliability of a numeri cal procedure by solving a number of example problems. This pragmatic approach is often necessary; however, it may not necessarily yield a general solution scheme. This is because a procedure may provide satis factory solutions for a subset of problems within the class of problems for which it is devised, and it may not provide consistent answers for another subset. One of the ways to establish generality of a numerical procedure is to examine its mathematical properties, such as convergence stability, and consistency. Hence, in addition to the quantitative study, mathematical expressions for numerical stability were derived for two FE schemes for the linear parabolic equation (Section I, Part IV), and for nonlinear hyperbolic equations, Section II. U. The concept of finite elements in the time domain was
3 investigated. A formulation was obtained for the general dynamical sys tem, and was applied in solving a single-degree-of-freedom system. Formulations based on this concept were derived for a parabolic, hyper bolic, and nonlinear fluid flow equation by using different interpola tion functions. ,5« The study involves rather a small step related to the general and significant problem of selection of the optimum numerical procedure. In the preliminary analysis, only problems governed by one-dimensional equations were considered. Further investigations will be required to answer in detail additional and more complicated aspects of multidimen sional problems. At this stage of the development of numerical methods, the analyst should devote attention to this question and evolve criteria to assist the user in selecting the optimum scheme for his specific needs.
u PART II: EFFECT OF ORDER OF APPROXIMATION MODELS
6. Some alternatives available for improving a FE solution are to use a finer mesh layout with a simple lower order model (or function) or to use a coarser mesh with higher order functions. For a given mesh, a higher-order model will usually yield a more accurate solution. However, a higher order model will give a greater degree of connectivity, thus increasing the band width of the resulting equation set.1, ’ ’ Because the number of operations, say in the commonly used Gauss-Doolittle pro- cedure, is proportional to NB , where N = number of equations and B = semiband width , the equations for the higher order model will take more computation time. Moreover, although the higher order assemblage may involve a smaller number of elements for a given accuracy, the com putation time taken in formulating the element matrices and forcing pa rameter vectors will be greater for the higher order models.
7 . There exist tradeoffs in computational time and cost as well as accuracy that must be examined if the above question is to be answered adequately. In the final analysis, what should govern the selection of an approach is the degree of accuracy per unit cost. This cost should include the expense of preparing the input, the charges for digital l I4. 8 computation time, and the amortization of software development. ’ 5 8. Some of the above aspects were evaluated by adopting two ap proximating models, linear and cubic, for pore water pressure in the parabolic equation governing one-dimensional consolidation^ (or heat flow). The approximation of one-dimensional consolidation can provide useful engineering solutions for many practical problems such as esti mation of vertical settlements of foundations and embankments.
Finite Element Formulation
Background 9. On the basis of Darcy's law, satisfaction of the condition of continuity, complete saturation of porous soils and linear elastic be havior of the soil skeleton, the governing equation for one-dimensional
5 (vertical) consolidation proposed by Terzaghi is
92 u _ 9u
Cv 8x „ 2 " 9t (i) where u = pore water pressure, x = space coordinate, and t = time co ordinate. The coefficient of vertical consolidation, c^ , is expressed as
k (l + e v c = ------in which k^ = coefficient of vertical permeability, e = void ratio of soil defined as the ratio between volume of voids and volume of solids, = unit weight of water, and a^ = coefficient of compressibility of soil. Terzaghi assumed that the total constant vertical stress, a , applied to the soil is carried jointly by the soil skelton and the water in the pores. Accordingly,
a = o' + u (3) in which o' = effective stress in soil and u = pore water pressure. Initially, the total stress a is carried wholly by the pore fluid, i.e. at t=0, o' = 0 , and a = u . As time elapses, the stress is transferred to the soil skeleton and finally at t = 00 , o' = a , and u = 0 . 10. The soil is assumed to deform as a linearly elastic material, and hence, the stress-strain law can be expressed as
da = -■ de a («0 v in which a^ equals the slope of the stress versus void ratio curve. Because of the relationships in Equations 3 and U, it is necessary to ob tain solutions for pore water pressure u only. Hence, solution of Equa tion 1 would also permit computations of deformations in the soil mass.
6 Derivation of finite element equations
11. The following formulation is based on a variational principle for initial value problems by Gurtin,^ and Sandhu and Pister.1^ The boundary conditions considered herein are (Fig. l):
u(0,T ) = U-^x) xe(t - At,t) (5) u U, t ) = u2 (x) xs(t - At,t)
The initial conditions are
u(x,0) = u q (x ) 0 £ x £ l (6 )
Integration of equation 1 over a short time interval, At , gives
t 2 c dx = u(t) - u(t - At) (7) V 3x2 t-Atl which can be rewritten as
~2 I * o U (8 ) 8 cv 7 T = u “ uo dx
where the notation * denotes convolution product and g' * u denotes ,t I udx . The boundary conditions are transformed to t-At
g f * u(o,x) = g' * tLj^x) (9 ) g ! * u(&,x) = g' * u2(x)
12. The governing functional corresponding to Equation 1 can be
written as
32u = J ju * u - U * g' * cv — — - 2u * u q dx (1 0 ) dx
Introduction of nonhomogeneous boundary conditions gives
7 N O . O F MESH N O . NODES
1 6 2 11 3 21 4 26 A<7 A ct
i U(0, T) e Û, (T) r 0 U(0, T) - U^T) = 0 -DRAINAGE ©
U(X,0) = UQ (X) = A a T T © U (X , 0 ) = U q (X ) = A ( 7
ELEMENTS-*
tTm e ~m INTERFACE m u + & + NODES-* m m
u(1, t) = u2(r) = 0 V7/Y7777777777; x : DRAINAGE d u IM P E R V IO U S ------= 0 dx q. HOMOGENEOUS MASS AND BOUNDARY CONDITIONS b. VARIOUS MESHES c. LAYERED MASS
Fig. 1. One-dimensional consolidation of systems of porous soils ,2 J u * u - u * rf * C — ~ - 2U * U dx A2 V 3x2 l L
(u - 2ux ) * g' * cv |^ (u - 2u2 ) * S' * cv |i (11) x=o iJx=£
Finally, the governing functional in Equation 11 can be written as
A 3 = l /[a * u + * g' » cv I f - 2U « u J dx (12 ) * ry » * - 2 If] ♦ 2[(u-^)«g-cv|f] Hu.- u1) V J x=o x=£
If u satisfies identically the prescribed boundary conditions, the two
boundary terms in can be dropped. 13. If the integration over the interval (o,il) is replaced by sum of integrals over E elements, the discretized functional in Equa
tion 12 can be written as
(13)
where [N] = vector (matrix) of interpolation functions, [Nx ] = space derivation of the interpolation functions, and {qn > = nodal pore water
pressure vector. ll;. the spatial discretization is achieved by using linear and cubic pore pressure models as shown in Fig. 2. Derivation of equations for the linear pore pressure model follows. The linear pore pressure
model is (Fig. 2a)
u = [ |n - D | d * d ] {<4} (lUa)
hence, i-l. d£ dL = 1_ (N } (li+b) X dL dx a
9 o. LINEAR MODEL
NODE 1 NODE 2 •--- (L,.L2) —------—------a : ^
“= f-Ì(3"2Ll) L?L2° L2<3 -2L2> " LÌLà]^
L, = at/a , l2 = a2/ a
b. CUBIC MODEL
Fig. 2. One-dimensional element and various approximating models
10 , _ 2 , , , dL 2 where L = — (x - xQ) and — = — . Therefore e e
L=+l ± < V (V T d* = f / I fV {1,x)T dL ■ r U 1 (1 5 ) L=-l/ -1
L=+l J. / {N} {N}T = j w {N}T dL = l 0 (16) L=-l -1
Substitution of Equations 15 and l6 into Equation 13 and vanishing of variation of gives
5f*[K] {r} + [P] {r} = {R} (1 7 ) where 1 -1
[K]=I[kl, [k eJ = a e=l -1 1
E 2 1 [p] = I [Pj , [Pj = (18) e=l eJ 6 1 2 E 2 1
=1 (R > , iR } = 1 2 e=l J {<} 15. The equations for the cubic model, Fig. 2b, are
36 3a -36 e 3ae
2 ha2 -3a -a e e e (1 9 ) [ke] symm. 36 30a -3ae
Ua2 e
11 156 22ae 5 5^^ -13a
ka 13a -3ac e e (20 ) [Pel 1+20 symm. 156 -22 a
1+a^
Integration in the time domain
l6. By"writing
HH+ K4*}) ( 21 )
that is, {r^} = f2r ^ Equation 17 can he expressed as } |2rt_AtI " {rt-At} ’ t-2
(22) ([Kl + if [Pl) |rt_At [ = {E) + it [Pl {rt-4t}
17. Initial values (r(x,0)} are prescribed, hence, Equation 22 can be used to evaluate {r(x,t)} by selecting a suitable time increment At . The above integration scheme has been used since it has been often 6 7 12 13 employed for solution of similar initial value problems. 5 3 5 By making necessary rearrangements, Equation 22 is expressed in terms of increment of time factor, AT , as
c At AT = (23) a e
where At = time increment and a^ = length of an element. The time factor T is defined subsequently in Equation 25« 18. Computer codes were developed by using linear and cubic models on the basis of the following formulations. The codes were pro grammed on the time sharing system of the GE U30 computer series. The
12 banded equation set was solved by using the symmetric Gauss-Doolittle procedure.
Numerical Characteristics and Comparisons
19. The homogeneous soil system considered for this analysis is shown schematically in Fig. 1 together with the boundary conditions used herein. The closed form solution of this problem obtained by using the 9 method of separation of variables is
u: (2h)
where M = tt/2 (2i + l) , H = length of the drainage path, Fig. 1, and T is called time factor,
c t v T = (25) H
For settlement analysis from the consolidation theory, a secondary parameter called average degree of consolidation, U , in addition to the primary unknown variable, u , is defined in terms of u ,
i=o°
U* (2 6)
20. For the finite element analyses, the soil mass was dis cretized into a number of sets of finite element meshes, Fig. lb. The nodal values of u were computed by using Equation 22, and then the values of U were evaluated on the basis of Equation 26. Accuracy and stability 21. Figure 3 shows plots of percent errors in u versus number of
13 ERROR (u*-u), PERCENT lb nodes in the discretized medium. The error was evaluated as the per centage of the difference between the closed form and the numerical so lutions, u* - u , where u* is the closed form solution. As expected, the accuracy of numerical solutions improves with refinement in the mesh for both linear and cubic models. The solutions from the cubic model converge faster than the linear model in the top regions (0.1£ and 0.2£) of the consolidating mass, whereas the solutions from the linear model seem to converge faster in the middle zones (O.hü) of the layer. This behavior may be explained on the basis of the distribution of pore water pressures at various time level, T . Figure h shows distribu tions of u* for T = 0.05, 0.10, and 0.20 in comparison with the com puted values of u from the linear model with the mesh consisting of
Fig. k. Distributions of pore water pressures at various time factors
21 nodes (Fig. lb) The two solutions agree very closely. The solutions from the cubic model with 21 nodes also show similar comparisons with the closed form solutions. Generally, the rate of change of u in the middle zones is small, whereas it is relatively large in the upper zones. Hence, additional unknown nodal gradients in the cubic model do not seem necessary for the middle zone and can make the cubic model solutions slightly less accurate than the linear model solutions, Figs. 3 and 5. Figure 5 shows the convergence of the degree of consolidation
15 better agreement with the closed form solutions than the cubic model. cubic the than solutions form closed the with agreement better oain vru A fr ehwt oe eul f2, i. b It lb. Fig. 21, of equal nodes with mesh a for AT versus locations verge from below. Overall, the linear model shows as good as and often often and as good as shows model linear the Overall, below. from verge s enta gnrly byn ie atr nrmn eul o about to equal increment factor time a beyond generally, that seen is the same convergence behavior. convergence same the bv, hra te ouin fr vrg dge o osldto con consolidation of degree average for solutions the whereas above, for the linear and cubic models. The two solutions for U yield about about yield U for solutions two The models. cubic and linear the for ERROR (U *-U ), PERCENT 23. Figure 6 shows plots of errors in pore pressures at various various at pressures pore in errors of plots shows 6 Figure 23. 22. It can be seen that the pore pressure solutions converge from from converge solutions pressure pore the that seen be can It 22. i. . ovrec o vrg dge o osldto at consolidation of degree average of Convergence 5. Fig. 01 n A 0.125 = AT and 0.1 = T 16 Fig. 6. Stability of pore water pressure solutions for 21 nodes and T = 0.1
1.0, the solutions become less accurate. This is true for both linear and cubic models• The solutions for the degree of consolidation U be comes less accurate beyond AT equal to about 1.0, Fig. 7* 2k. An examination of Fig. 6 reveals that the stability behavior of pore water pressure solutions is different at different locations in the medium. For instance, in the top zones (x = 0.1£ and 0.2l) the solutions become less accurate around AT = 1.0 , whereas in the middle zones (x = 0 .hi) the solutions seem to maintain accuracy up to a value
of AT of about 3.0. 25. From the analysis of accuracy, Fig. 3, and of stability,
IT Fig. 7- Stability of average degree of consolidation for 21 nodes and T = 0.1
Fig. 6, it is seen that for complete understanding, the behavior of a numerical solution for the primary quantity u should be examined at a number of typical locations in the discretized medium. Furthermore, if a secondary quantity such as the degree of consolidation, U , is de fined for practical use, it is necessary to examine the behavior of nu merical solutions for U also. Computational time 26. Figure 8 shows plots of computational time versus number of nodes for the linear and cubic model formulations. The time shown con sists of total time for execution of a given number of time increments which includes organization of input data in the computer, evaluation of element matrices, assembly of element matrices, introduction of boundary conditions and solution of the resulting equation set. The total time required by the cubic model is generally found to be about
18 Fig. 8. Comparison of total computational times
four times that required by the linear model. 2 27. Figure 9 compares plots of the error, u* - u versus NB It can be seen that this plot gives a different view than the plot of u* - u versus number of nodes , Fig. 3. It shows that if an evaluation of accuracy versus computational effort is desired, plots such as Fig. 3 do not yield an adequate picture. 28. Table 1 shows relative errors and computational times for typical values of nodes and locations for the linear and cubic models.
19 ERROR PERCENT 800 900 Any gain whatsoever in accuracy requires significant increase in compu tational effort, indicating the tradeoffs that exist between accuracy and computational effort. Formulation effort 29. This is the effort required in formulating the problem which includes selection of mesh layout and approximating models, preparation of data, and carrying out the operations involved in the derivation of element equations. It is difficult to obtain precise time required for these operations because they would depend upon many factors including human characteristics such as skill and the experience of the analyst who formulates the problem. On the basis of limited subjective experi ence, thé formulation time for the cubic model was found to be 3 or times that for the linear model. Satisfaction of continuity of flow 30. In the foregoing analyses, consolidation of a homogeneous soil was considered. In actual field situations, however, layered soil systems are a common occurrence. In this case, the physical situation requires that the following additional conditions be satisfied at the interfaces between the layers, Fig. lc.
— + u = u (27a) m m
(27b) where = the fluid flux and is proportional to the product k^G^ » 0 = the gradient of pore water pressure, 9 m = an interface, oX Fig. lc, and the superscripts - and + denote close vicinities on either side of the interface m . The condition in Equation 27a is satisfied in the finite element formulation, since the compatibility of nodal pore pressure is guaranteed in the assembly procedure. Equation 17* The con dition in Equation 27b represents the requirement of continuity of fluid flux across an interface, and in the case of the linear model, this con dition is not fulfilled precisely. It is found, however, that although the continuity condition is significantly violated in the initial time
21 levels, it is satisfied approximately for subsequent time levels. This is illustrated in Fig. 10 for the consolidation of the three-layered system considered subsequently. The ratio = Q^/Q^ is plotted in Fig. 10 for various values of time factors, T = c -nt/H^ , where the -D V JD £> subscript B denotes bottom layer. Ideally, the ratio should be unity. It can be seen that is different from unity up to a time factor of about 3.0, whereas beyond this value of time factor, that is, for the major portion of the consolidation process, the ratio is close to unity. The values of the coefficient of permeability k^ used in computing the fluid flux Q were those obtained from laboratory exper- ±k iments, Table 2, Barden and Youhan. The approximate values of 0 were obtained as 0” - (u - u n )/am 1 , 0+ - (u - u )/am . m v m m-l" e m m+1 m" e 31. In the case of the cubic model, Fig. 2b, both nodal pore pressures and nodal gradients are compatible. Hence, in Equation 27b, since 0 = 0+ , k~ should be equal to k+ . This implies that for use of the cubic model, the system should be homogeneous. Hence, for layered systems the cubic model with nodal gradients as unknowns could perhaps provide approximate solutions and may not be suitable. 32. From the foregoing analyses of accuracy, stability, computa tional effort and satisfaction of continuity of flow conditions, the linear model can provide acceptable accuracy with most economy. Hence, for one-dimensional settlement analysis of foundations, the linear-model formulation is recommended. Two applications of the linear-model formu lation are described below.
Applications
33. The two examples considered herein are adopted from the ex- ik perimental work by Barden and Younan. These examples involve a com prehensive experimental study of one-dimensional consolidation of two- and three-layered cohesive soil systems, Fig. 11. The experiments were conducted by using a number of Rowe consolidation cells connected in series. Special care was taken to satisfy accurately the continuity conditions, Equation 27b, at the interfaces. The pore pressures at the
22 Fig. 10. Satisfaction of continuity of flow at interfaces in three-layered system (Fig. lib) interfaces were measured by using transducers, and the systems were maintained deaired by applying sufficient back pressures. 3^. Three remolded saturated clays were used in the experiments. The effects of structural viscosity were minimized by maintaining the load increment ratio Aa/a = 1 . The basic material parameters, coeffi cient of consolidation, c^ , coefficient of compressibility, a^ , and the coefficient of permeability, k^ , were determined from detailed tests for various pressure increments, Aa = 5-10, 10-20, 20-^0, and 1*0-80 psi. Only those values of c^ and k^ for the load increment 20-1*0 psi used in this study are given in Table 2 for the two- and three layered systems. The thicknesses of the component layers for the load increment 20-1+0 psi are shown in Fig. 11. 35* Figure 12 shows a comparison between numerical, experimental, and closed form solutions for pore water pressures at the impervious base 2-2 and at the interface 1-1 for the two-layered soil system, lU Fig. 11a. The closed form solutions for the layered system were ob tained on the basis of Terzaghi!s theory. Equations 1 and 2k. The re sults from the finite element analysis show excellent agreement with the closed form solution over the complete range of consolidation, and with the experimental solutions beyond a time factor value of about 0.15. The correlation between the computed, experimental and closed form values of the average degree of consolidation is considered to be remarkably good, Fig. 13. The value of At = 0.000001 year was used for the two layered system. 36. Figures 1^+ and 15 show comparisons between the three solutions for pore water pressures and degree of consolidation for the three lay ered system, Fig. lib. Again the correlation between the solutions is found to be highly satisfactory. The value of At = 0.00001 year was used for the three-layered system. 37« The computational time for solutions for layered systems in creases approximately in direct proportion with the number of elements, Fig. 16 .
2k 20 PSI 20 PSI
N O . O F N O . O F NODES NODES ELEMENTS ELEMENTS 1 r n
D E R W E N T 8 KAOLINITE 0 .8 0 ' 8 0 .8 5 " +1.25% BENTONITE v 9 1 ------1 9A D E R W E N T 4 0 .3 9 ' + 1.25% BENTONITE 13 ' iV7777777?z 8 DERWENT 0 .86" ro IMPERVIOUS VJ1 17 2 a. TWO-LAYERED SYSTEM
8 KAOLINITE 0.82*
25 3 r77/////ZA r - r 7 " " r NOTE: THE NODES WITHIN A LAYER IMPERVIOUS ARE EQUIDISTANT. b. THREE-LAYERED SYSTEMS
Fig. 11. One-dimensional consolidation tests on layered systems TIME FACTOR Tg
Fig. 12. Comparison of pore water pressure dissipation for two-layered system AVERAGE DEGREE OF CONSOLIDATION Fig. 13. Comparison of average degree of consolidation for two-layered system two-layered for consolidation ofdegree average of Comparison 13.Fig. PORE-WATER PRESSURE DISSIPATION Fig. ik. oprsno oewtrpesr dissipation pressure water pore of Comparison for three-layered system three-layered for 28
AVERAGE DEGREE OF CONSOLIDATION TIME, SECONDS 300 250 200 150 100 50 0 0
i. 1 Fig. / / © □ r ------6 10 . Computational time for layered systems layered for time Computational . -• TWO TWO -• 0 L E E R H T -0
LEGEND / / LA'S MBR F NODES OF BER UM N 20 (ER / / AYER /
30 / 30
/ • / 40 i / /
/ / / / l 50 Comments
38. The present study has demonstrated a useful finite element procedure for one-dimensional deformation and seepage in porous elastic media. A number of factors that influence the choice of a formulation scheme from viewpoints of accuracy and economy were considered. On the basis of the analyses, the following remarks are offered. 39. For the class of problems considered herein, the strategy of using lower order approximating models with reasonably fine mesh is pref erable. Specifically, the linear model is found to be better than the cubic model. 1*0. As far as engineering accuracy is concerned, the solution procedure with the proposed integration scheme would permit selection of a fairly coarse mesh. For instance, meshes with 11 and 21 nodes involve maximum errors in pore pressure solution of about 3 and 1 percent (Fig.3) and in degree of consolidation of about 20 to 10 percent (Fig. 5 ), respectively. Ul. The final selection of the mesh would, however, be guided by considerations of accuracy and economy. The pore water pressure solu tion at 0 .1 £ , 0 .2£ , and 0 .1* £ seem to become less accurate at different values of AT, Fig. 6 . The solutions for degree of consoli dation diverges beyond AT equal to about 1.0. Overall, a value of AT less than unity seems to yield a suitable mesh layout.
31 PART III: EVALUATION OF FINITE ELEMENT AND FINITE DIFFERENCE SCHEMES
1+2. Seven finite difference and finite element schemes were for mulated for the problem of one-dimensional consolidation governed by the parabolic equation (Equation l). They were evaluated quantitatively from the viewpoints of numerical convergence, accuracy, stability, and computational efforts.
Finite Difference Schemes
1+3. A number of explicit and implicit schemes were considered. The finite difference analogs according to these schemes corresponding to Equation 1 are: (i) Simple explicitiL^,1^,^,''7--l(FD)
A T ^ u ) " 0 = ut+1 - u* (28) l i i
2 "t where the symbol (6 u)^ denotes
(62u )^ = u^ - 2u^ + u* (29) i l-l l l+l
O AT = c At/a , and a = length of an element. (ii) Implicit— Crank-Nicholson^5516— 2(FD)
AT /,~2 xt+1 / «2 ,t\ t+1 t <(6 u)± + (5 u ) ^ = n ± - (30) 2
(iii) Implicit^5 — 3(FD)
Am/ j.2 ^t+1 t+1 t AT(<5 u). = u. - u. (31) 1 1 x
(iv) Special explicit— Saulev scheme. ^ ^--1+A(FD)
/ t+1 t+1 t t \ _ „t+1 t AT I u. - u. - u. + u. ) - u -u (32) V l-l l l i+V 1 1
32 (v) Special explicit— alternating direction^ ~^--^B(FD) Here the expression in Equation 32 is used at time t for mesh points starting say from left to right, and then at time t + 1 , the follow ing expression is used starting from right to left:
/ tt+2 t+2 t+1 ^ t+l\ tt+2 t+1 AT - u. - u. + u - u. (33) (ui1+1 1 1 l-i/• n J = U a . 1
kk. For consolidation in layered systems, the pore water pressures at the nodes in the vicinity of an interface were modified to satisfy the continuity of flow condition:
k 0 m m km+lÖm+l (310 where k is the coefficient of permeability, m denotes a layer, and e = f3x .
Finite Element Method
1+5- The FE formulation was covered in Part II. Two time integra tion schemes were considered. The first and perhaps the: most simple time integration scheme, l(FE), can be obtained by using the forward difference scheme {r } — ir } 1 t+At 1 tS {r} At (35)
Substitution of Equation 35 in Equation IT yields the assemblage equa tion as:
([k]+ii[P!){W }=[R]+Si[P]