flac3D

  • Upload
    usosan

  • View
    93

  • Download
    0

Embed Size (px)

Citation preview

  • An Examination of the FLAC Software:

    Undrained Triaxial Test on Cam-clay

    Prepared by Jack Montgomery

    June 8, 2010

    ECI 280A Term Project

    Instructor: Boris Jeremi

  • Introduction

    Many problems in geotechnical engineering are too complex to be properly evaluated

    using analytical solutions or physical models. Complexity may be due to non-linear material

    properties, complex loading patterns or non-standard geometries. For these problems

    numerical simulations are often used to examine the forces and displacements from various

    types of loading. Many software packages are available for this type of simulation, but one that

    is commonly used in practice is FLAC. FLAC, Fast Lagrangian Analysis of Continua, is distributed

    by Itasca Consulting Group and has the ability to solve a wide range of geotechnical problems

    involving dynamic loading, multiple material models, structural elements and pore fluid. In

    order to better understand the inner workings of the software, the formulation will be

    examined, some of the methods will be discussed and finally an undrained triaxial test on a

    sample of Cam-clay will be simulated.

    FLAC: Fast Lagrangian Analysis of Continua

    FLAC is a commercially available, two-dimensional finite difference software program. In

    FLAC, the finite difference method is used, instead of the more commonly understood finite

    element method, to solve the differential equations associated with each problem. Each

    differential term is replaced with an algebraic equation called a finite difference approximation

    and this set of equations is solved in FLAC using an explicit integration scheme. The applied

    forces are then divided into a series of incremental forces, referred to as time steps. The explicit

    integration solves each equation of motion at each element for each time step with no

    iterations (FLAC 2001).

    After solving the equations for each element, the calculated velocities and

    displacements are sent to the constitutive model to calculate stresses and strains. These

    stresses and strains are then used to create new equations of motions. Because there is no

    iteration this method assumes that the changes in each element do not affect the neighboring

    elements within a time step. This assumption allows each element to be evaluated

    independently. To ensure this assumption is valid, a small enough time step must be used so

    that information would not passed between the elements within a time step. This method is

    in contrast to the implicit method which solves the equations of motion for all elements at

    once. This requires iteration to find the solution and may take more computational effort for

    each time step, but significantly larger time steps may be used compared with explicit

    integration scheme. Two of the most important features are the finite difference method and

    the explicit integration scheme. These will be explored further in the following sections.

    Finite Difference Method

    One important difference between FLAC and some of the more commonly understood

    finite element programs is FLAC uses the finite difference solution for solving differential

  • equations. The Finite Difference method is a numerical solution scheme for solving the

    governing equations of a continuum body (Bathe 1996). This technique was pioneered in the

    1920s as a method of solving nonlinear hydrodynamic equations (for more on development

    see Thom and Apelt, 1961). Differential terms in the equations are replaced by algebraic

    equations called finite difference approximations. These approximations are defined as the

    difference between field variables at two discrete points in space. The finite difference method

    has no shape functions, as finite elements do, so a linear change is assumed between the two

    points (Bathe 1996). This would be equivalent to a finite element with a linear shape function.

    This simplification can cause numerical errors in areas of high gradients since the

    approximation can average out the changing variables. In these areas many elements may be

    required to properly capture the response. One example of this can be seen in liquefaction

    modeling where using only one row of elements can cause excess pore pressures to be

    averaged out with the layers above and below. In an extreme case this could mask the

    liquefaction phenomenon completely. This can be avoided by using two or more rows of

    elements for any material.

    An example will be used to illustrate how the finite difference technique can be applied

    to a uniaxial bar (see Figure 1). This example is worked out by Bathe (1996) for both finite

    difference and finite elements, but only the finite difference solution is presented here. In this

    case the governing equation of the bar is the equation of motion. This equation requires finding

    the second derivative of displacement. Consider three nodes on the bar spaced equally at a

    distance h and two sub-nodes located midway between each of the main nodes. The second

    derivative of the center node is approximated by considering the change in displacements over

    three nodes and two points half-way between these nodes. First, the first derivative of

    displacements at the two sub-nodes is found by considering the difference in displacements

    between the left node and the center node. This approximates the derivative as the change in

    displacements divided by the distance between the nodes. The second derivative at the central

    node is found by taking the derivative between the first derivatives at the sub-nodes divided by

    the distance between them. The first derivatives of the sub-nodes were found as described

    above in terms of displacements at all three nodes. When the equations are combined the

    result can be seen in Figure 1. One issue that can be immediately seen is the need for a virtual

    node outside the bar to properly impose boundary conditions on the bar.

    The finite difference method has been used for many years and has some distinct

    advantages and disadvantages for solving differential equations. One of the most important

    advantages is the simplicity of the formulations. This method can be easily coded into programs

    and requires little memory to perform calculations. This simplicity makes it easy to recalculate

    the equations of motions at each step, so more steps can be used without great penalty in the

    form of calculation time. This makes it an ideal method for combination with the explicit

  • integration scheme which requires small step sizes. There are also no shape functions in the

    finite difference method as there are in the finite element method. This means that variables

    are undefined within the elements, but values can be assumed to vary linearly between nodes.

    In FLAC, quadrilateral elements are specified by the user, but are subdivided into two triangular

    elements internally. Given the lack of shape functions, these are analogous to constant strain,

    triangular elements in finite elements. One distinct disadvantage of this approach is that many

    elements may be required to properly capture areas where variables are changing rapidly. This

    is especially important in problems like liquefaction. If only one row of elements is liquefiable,

    the excess pore pressures will be averaged out by the nodes above and below the liquefied

    elements. More details on the use of the finite difference method in FLAC can be found in the

    users manual, Theory and Background (FLAC 2001).

    Explicit Integration

    There are two main types of numerical integration used in finite element or finite

    difference solutions. These are the explicit and implicit methods, respectively. In numerical

    solutions integration of the equations is performed at both the global and constitutive levels.

    The global integration is concerned with the response of the entire system, while the

    constitutive integration is concerned with the material response. The main difference between

    the two methods is that the implicit method uses iterations to ensure equilibrium at each step.

    On the global level, iterations ensure that neighboring elements are all in equilibrium with each

    other and the applied loads. On the constitutive level, iterations are performed to ensure that

    when the material is yielding it finishes on the yield surface. These iterations are not concerned

    with the accuracy of the solution, but do ensure equilibrium at each step which may increase

    the accuracy of the solution when compared with the explicit method in which the solution may

    not be on the yield surface at all. The explicit method relies on small time steps to ensure any

    errors in equilibrium are small and can be neglected.

    On the global level, FLAC uses the explicit integration scheme which uses no iterations

    to find equilibrium among the elements. This means that changes in stresses within a single

    element will not affect the calculated displacements, and therefore the neighboring elements,

    until the next step. This assumption is valid as long as the time step is small enough to ensure

    that the calculation wave moves faster than the physical wave of the loading (FLAC 2002). This

    means if the system is being loaded by an earthquake, the time step must be small enough so

    the wave could not propagate through an element within the time step. Over multiple steps the

    loading would propagate upward just as it would physically. This time step is calculated

    automatically by the program to ensure stability and accuracy of the solution.

    For all material models included with the FLAC software explicit integration is used at

    the constitutive level. If a user was to define their own model they could include some sort of

    implicit integration. Doing this would ensure equilibrium at the expense of computational time

  • and memory requirements. Explicit integration ensures accuracy by using a small time step

    which is just as important at the constitutive level as it is at the global level. At the constitutive

    level, the KarushKuhnTucker conditions for constitutive models (Karush 1939, Kuhn-Tucker

    1951) require that the stress state of a material be on or within the yield surface at all times.

    When a material attempts to cross the yield surface, plastic strains will develop and the stress

    state will change. In constitutive modeling, this is handled by using an elastic predictor to allow

    the stress state to cross the yield surface and then uses a plastic corrector to bring the material

    back to the yield surface (Figure 2). The implicit method would use iterations to ensure that the

    stress state ends on the yield surface. Explicit integration will draw a tangent line at the location

    where the material crossed the yield surface and will come back to that line. If the time step is

    small the error between the tangent plan and the yield surface will likely be small (Ortiz and

    Popov 1985). Errors will be most significant with irregular yield surfaces and at bifurcation

    points. This error is easy to examine at the element level, but can be lost in a large problem

    with many elements. This is one of the many reasons single element tests are important to

    check the accuracy and stability of the numerical solution.

    Numerical Simulation

    A single element test will be conducted on a sample of Cam-clay. The simulation will

    serve two purposes. First, the results will be compared to analytical solutions published by

    David Muir Wood (1990) to gauge the accuracy of the FLAC solution compared with what the

    model should be predicting. This process is called verification of the software. Verification is

    used to ensure that the model and software are functioning correctly numerically (Muir Wood

    2004). A complimentary process to verification is validation of the model. Validation compares

    the model response to some actual test result or case history to see if the model is capturing

    the desired behavior. A simple way to think of it is verification is checking whether the model is

    working right, validation checks whether it is the right model. Although validation is a crucial

    step in any numerical simulation, it is a test of the appropriateness of the model, not of the

    accuracy of the software program. For this reason it will not be explored further here. More

    about verification and validation can be found in Oberkampf et al. (2002).

    The second purpose of these analyses is to gauge the effect of different size time steps.

    As was discussed earlier the size of the time step directly affects the accuracy and stability of an

    explicit integration solution. In FLAC the minimum time step size is calculated automatically to

    ensure numerical stability. The calculated size is based on the relative stiffness of the materials,

    but an easy way to adjust the equivalent time step is to adjust the rate of loading so that more

    load is applied during a given time step. For this report an undrained triaxial test will be

    simulated using material parameters from Muir Wood (1990) and the Cam-clay material model

    implemented in FLAC. Each component of the simulation will be described in the following

    sections.

  • Cam-clay Constitutive Model

    The modified Cam-clay model was developed by Roscoe and Burland (1968) and is a

    modification of the original Cam-clay model developed by Roscoe et al. 1958. Because the

    original model is not considered in this report, the modified Cam-clay model will simply be

    referred to as Cam-clay. This model is an elasto-plastic constitutive model with a nonlinear

    hardening and softening law which depends on the pre-consolidation pressure of the soil. The

    model determines the response of the soil based on the specific volume or void ratio, a

    deviator stress and a mean effective stress. Cam-clay is an associated plastic flow model in

    which the yield surface is defined as an ellipsoid in q-p space with no strength at the origin and

    pre-consolidation pressure (i.e. isotropic consolidation). Within the yield surface the material is

    elastic and as the stress state crosses the yield surface both plastic volumetric and deviatoric

    strains will develop. The material is either incrementally contractive or dilative depending on

    whether it is dense or loose of critical state. The material will harden or soften depending on

    the volumetric strains. At critical state the sample will undergo only deviatoric strain and

    therefore will not harden or soften. This model has a relatively simple formulation and is

    compatible with critical state soil mechanics and certain idealized clays. The formulation is

    shown graphically in Figure 3. Further details about Cam-clay can be found in Muir Wood

    (1990).

    Simulated Triaxial Test

    An isotropically consolidated, undrained triaxial test was simulated on a sample of Cam-

    clay. In FLAC this was simulated with a single element using the axisymmetric option within the

    software. The boundary conditions selected are shown in Figure 4. The horizontal axis of the

    element is considered to be the radial direction and the vertical axis is the axial direction. The

    nodes are numbered clockwise from the lower right corner as node 1, 2, 3 and 4. Each node has

    two degrees of freedom in the axial and radial direction. Node 1 and 2 are fixed in the radial

    direction to simulate the axisymmetric element. Node 3 is free to move in both directions and

    nodes 1 and 3 are fixed in the axial direction to simulate the end platen. To summarize, Node 1

    is fixed, Node 2 is fixed radially, Node 3 is free and Node 4 is fixed axially. A constant confining

    pressure is applied between nodes 3 and 4 to simulate the cell pressure and a constant rate of

    axial strain was applied to the top of the element. As a consequence of the constant strain

    assumption the total axial load is not controlled directly. The effects of this will be examined in

    the results section of this report.

    The internal stresses in the element are initialized to the confining pressure to simulate

    an isotropically consolidated sample. The sample is initially saturated and the bulk modulus of

    water is defined to simulate the compressibility of water. Groundwater flow is turned off to

    simulate an undrained test. As plastic volumetric strains try to accumulate, the sample is unable

    to change volume due to saturation and the pore pressure will increase or decrease depending

  • on whether the sample is trying to contract or dilate. Two samples were simulated to capture

    this behavior. One was a lightly over-consolidated sample (OCR =1.5) which is loose of critical

    and a heavily over-consolidated sample (OCR = 20) which is dense of critical. The material

    properties were taken from published test results in Muir Wood (1990). These properties are

    summarized in Table 1.

    Analytical Solution

    To validate the results of the numerical solutions an analytical solution was used from

    Muir Wood (1990). This solution solves for the stresses in an undrained triaxial test at yielding,

    and critical state. A formula is also presented to calculate the stress path of the material

    between the initial yield point and critical state. The results of the analytical solutions for both

    the highly and lightly over-consolidated samples are presented in Figure 5. The predicted

    responses are consistent with general critical state theory in that the loose sample contracts

    and generates positive pore pressure, while the over-consolidated sample dilates and

    generates negative pore pressure. The calculations needed for the analytical solution were

    performed using the Mathcad software package and are shown in Appendix A.

    Numerical Results

    The numerical simulation was conducted using three different loading rates. The loading

    was applied as a constant rate of strain at the top of the model. The slowest loading rate was

    0.01% strain per step which was increased to 0.5% strain per step and then to 10% strain per

    step. This increase in loading was meant to raise the equivalent step size of the simulation and

    gauge the effects on the results. Figure 6 through Figure 8 show the response of the lightly

    over-consolidated sample to each of the three loading rates and Figure 9 through Figure 11

    show the response of the heavily over-consolidated sample. The figures show that the effective

    stresses predicted by the numerical models are in excellent agreement regardless of the step

    size. This is not surprising as Cam-clay has a very simple yield surface in which tangent lines will

    closely approximate the actual surface. This may not be the case if the loading crossed near the

    apex of the yield surface. Figure 12 shows the pore pressure response for both the lightly and

    heavily over-consolidated samples. Pore pressures are only shown for the smallest step size

    because the other paths were very erratic as suggested by the total stress paths shown in the

    other figures. As expected the lightly over-consolidated sample undergoes contraction and

    positive pore pressure generation. The heavily over-consolidated sample experiences

    contraction at first and then dilation. Selected points from the analytical and numerical

    solutions are shown in Tables 2 and 3 for comparison.

    Although the effective stress path is in good agreement regardless of the step size the

    total stress paths show how large steps can adversely affect the results. This triaxial test was

    simulated by applying and axial strain while holding the radial confining stress constant. This is

  • defined as a conventional triaxial test and the total stress path will rise at a slope of 3q to 1p

    (Powrie 2004). Only the step size of 0.01% produces a reasonable total stress path while the

    larger step sizes produce total stress paths which oscillate around the correct line. The reasons

    for this response will be explored next. In an undrained test pore pressures represent the

    sample attempting to change volume, but because the test is undrained this contraction or

    dilation occurs in the form of pore pressure. The pore pressure can be thought of as

    representative of the volumetric strains which would occur in a drained test. When the sample

    is loaded very quickly the pore pressure will suddenly spike or drop in response to the imposed

    strains. This would normally lower the effective stress of the sample, but in FLAC the

    constitutive model is only given displacements and velocities from which it produces stress and

    strains. Because the rate of strain is controlled by the loading, FLAC adjusts the total stress to

    maintain the proper strain rate. This adjustment occurs as the sample tries to maintain

    equilibrium under the large strains. In reality, the samples would likely form tension cracks

    when the effective stress dropped to zero, but the numerical solution prevents this and the

    error manifests itself in the total stress.

    Conclusions

    The formulation of the FLAC software package has been explored and the accuracy of

    the Cam-clay material model was examined by simulating an undrained triaxial test on a lightly

    and heavily over-consolidated sample. The results of these simulations show that FLAC does an

    excellent job in calculating the effective stress paths predicted by the analytical solution. It does

    this regardless of the step size chosen, but it can be seen from the total stress paths that the

    predicted strains are not accurate for the larger step sizes. These numerical errors are due to

    the large strains generated during each step.

    The dependence on step size shown in these simulations is not a surprising result. The

    accuracy of any explicit integration scheme is dependent on the size of the step chosen. This is

    a limitation of the integration method and does not speak to the accuracy of the software

    package. The implementation of the Cam-clay constitutive model appears to be functioning

    properly and the numerical methods in FLAC match well with the analytical solutions. As with

    all explicit integration schemes the step size must be small to ensure numerical stability and

    accuracy.

  • Table 1. Selected material parameters for Cam-clay

    Material Parameter Value

    Lambda 0.088

    Kappa 0.031

    Gamma 2.058

    N 2.097

    p0' 150

    M 0.882

    phi' 22.6

    G 3000

    Table 2. Results for lightly over-consolidated Cam-clay.

    Analysis Method

    OCR = 1.5

    qyield pyield qcs p'cs ucs

    Analytical 62.2 100.0 73.2 83.0 41.4

    Small T-Step 62.3 100.0 73.0 83.3 41.0

    Medium T-Step 62.2 100.0 73.1 83.1 36 to 46

    Large T-Step 62.3 100.1 73.1 82.9 ?

    Table 3. Results for heavily over-consolidated Cam-clay.

    Analysis Method

    OCR = 20

    qyield pyield qcs p'cs ucs

    Analytical 28.8 7.5 29.4 33.3 -16.0

    Small T-Step 28.8 7.5 29.4 33.1 -15.7

    Medium T-Step 28.8 7.5 29.3 33.3

    -12.5 to

    -19.5

    Large T-Step 28.6 7.5 29.3 33.3 ?

  • Finite Difference Example

    +

    = 0

    |1/2 = 1

    |+1/2 =+1

    | =|+1/2 |1/2

    2

    | =1

    2(+1 2 + 1)

    =

    (+1 + 2 1)

    Figure 1. Finite difference example for a bar in uniaxial tension (after Bathe, 1996)

    Figure 2. Schematic showing one large explicit step.

    Elastic Predictor

    Plastic Corrector

    m

    q

    P

  • Figure 3. Schematic of Cam-clay constitutive model.

    Figure 4. Boundary conditions for the axi-symmetric triaxial test

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    Cam-clay Constitutive Model

    CSL

    Yield Surface

    Dilative Contractive

    P0 = Pre-consolidation pressure = 150 kPa

  • Figure 5. Analytical solution to Cam-Clay loading.

    Figure 6. Numerical solutions of effective stress path for low OCR Cam-clay triaxial test.

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    Analytical Response

    Low OCR

    High OCR

    CSL

    Yield Surface

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    Low OCR - Effective Stress

    Low OCR

    High Velocity - Effective

    Medium Velocity - Effective

    Low Velocity - Effective

    CSL

    Yield Surface

  • Figure 7. Total and effective stress paths for low OCR Cam-clay

    Figure 8. Combined plot of all stress paths for low OCR Cam-clay.

    0

    10

    20

    30

    40

    50

    60

    70

    80

    -50 0 50 100 150 200 250

    q (

    kP

    a)

    p' (kPa)

    Low OCR - Numerical Stress Paths

    High Velocity - Effective

    High Velocity - Total

    Medium Velocity - Effective

    Medium Velocity - Total

    Low Velocity - Effective

    Low Velocity - Total

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    Combined Plot

    Low OCR

    High Velocity - Effective

    High Velocity - Total

    Medium Velocity - Effective

    Medium Velocity - Total

    Low Velocity - Effective

    Low Velocity - Total

    CSL

    Yield Surface

  • Figure 9. Numerical solutions of effective stress path for low OCR Cam-clay triaxial test.

    Figure 10. Total and effective stress paths for high OCR Cam-clay

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    High OCR - Effective Stress

    High OCR

    High Velocity - Effective

    Medium Velocity - Effective

    Low Velocity - Effective

    CSL

    Yield Surface

    0

    5

    10

    15

    20

    25

    30

    35

    -150 -100 -50 0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    High OCR - Numerical Stress Paths

    High Velocity - Effective

    High Velocity - Total

    Medium Velocity - Effective

    Medium Velocity - Total

    Low Velocity - Effective

    Low Velocity - Total

  • Figure 11. Combined plot of all stress paths for high OCR Cam-clay.

    Figure 12. Pore pressure response of both the high and low OCR samples

    0

    10

    20

    30

    40

    50

    60

    70

    80

    90

    100

    0 50 100 150

    q (

    kP

    a)

    p' (kPa)

    High OCR - Combined Plot

    High OCR

    High Velocity - Effective

    High Velocity - Total

    Medium Velocity - Effective

    Medium Velocity - Total

    Low Velocity - Effective

    Low Velocity - Total

    CSL

    Yield Surface

    -20

    -10

    0

    10

    20

    30

    40

    50

    0 500000 1000000 1500000 2000000

    u (

    kP

    a)

    Steps

    Pore Pressure Development

    High OCR

    Low OCR

  • References:

    Bathe, K.J. (1996). Finite Element Procedures. New Jersey. Prentice-Hall, 1996.

    FLAC User Manual. (2001). Itasca Consulting Group. Minnesota. 2001.

    Kuhn H. W. and A.W Tucker. (1951). "Nonlinear programming". Proceedings of 2nd Berkeley

    Symposium. Berkeley: University of California Press. pp. 481492.

    Karush W. (1939). Minima of Functions of Several Variables with Inequalities as Side

    Constraints. M.Sc. Dissertation. Dept. of Mathematics, Univ. of Chicago, Chicago, Illinois.

    Muir Wood, D. (1990). Soil Behaviour and Critical State Soil Mechanics. Cambridge University

    Press. 1990.

    Oberkampf W.L., Trucano T.G. and Hirsch C. (2002). Verification, Validation and Predictive

    Capability in Computational Engineering and Physics. Invited Paper: Foundations for

    Verification and Validation in the 21st

    Century Workshop. Laurel, Maryland: Johns Hopkins

    Univeristy. October 22-23, 2002.

    Powrie, W. (2004). Soil Mechanics: Concepts and Applications. Spon Press. 2004.

    Roscoe, K. H.; Schofield, A. N.; Wroth, C. P. (1958), "On the Yielding of Soils", Geotechnique 8:

    2253

    Roscoe, K.H. and Burland, J.B. (1968). On the generalized stress-strain behaviour of wet clay.

    Engineering Plasticity. Cambridge: Cambridge University Press. pp. 535-609.

    Thom, A. and C. J. Apelt, Field Computations in Engineering and Physics. London. Van Nostrand,

    1961

  • Appendix A: Mathcad Calculations