Scribd Logo
Upload

Welcome to Scribd!

  • 16 views

    Project #2: Review: Tensor vs. Vector Notation

    Uploaded by

    Suraj Jayswal
    The document describes Project #2 for a computational solid mechanics course. It involves implementing material models in C++, including (1) a linear elastic model, (2) a compressible Neo-Hookean hyperelastic model, and (3) a function to numerically verify that the models compute stresses and tangent matrices correctly as derivatives of the strain energy density. The linear elastic and hyperelastic models are each to be implemented as a C++ class with methods to compute the energy, stress, tangent matrix, and strain from a displacement gradient vector. The verification function uses finite differences to numerically compute derivatives and compare to the analytical solutions.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd

    Project #2: Review: Tensor vs. Vector Notation

    Uploaded by

    Suraj Jayswal
    16 views3 pages
    The document describes Project #2 for a computational solid mechanics course. It involves implementing material models in C++, including (1) a linear elastic model, (2) a compressible Neo-Hookean hyperelastic model, and (3) a function to numerically verify that the models compute stresses and tangent matrices correctly as derivatives of the strain energy density. The linear elastic and hyperelastic models are each to be implemented as a C++ class with methods to compute the energy, stress, tangent matrix, and strain from a displacement gradient vector. The verification function uses finite differences to numerically compute derivatives and compare to the analytical solutions.

    Original Description:

    bcd

    Original Title

    Project 2 2017

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    The document describes Project #2 for a computational solid mechanics course. It involves implementing material models in C++, including (1) a linear elastic model, (2) a compressible Neo-Hookean hyperelastic model, and (3) a function to numerically verify that the models compute stresses and tangent matrices correctly as derivatives of the strain energy density. The linear elastic and hyperelastic models are each to be implemented as a C++ class with methods to compute the energy, stress, tangent matrix, and strain from a displacement gradient vector. The verification function uses finite differences to numerically compute derivatives and compare to the analytical solutions.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    16 views3 pages

    Project #2: Review: Tensor vs. Vector Notation

    Uploaded by

    Suraj Jayswal
    The document describes Project #2 for a computational solid mechanics course. It involves implementing material models in C++, including (1) a linear elastic model, (2) a compressible Neo-Hookean hyperelastic model, and (3) a function to numerically verify that the models compute stresses and tangent matrices correctly as derivatives of the strain energy density. The linear elastic and hyperelastic models are each to be implemented as a C++ class with methods to compute the energy, stress, tangent matrix, and strain from a displacement gradient vector. The verification function uses finite differences to numerically compute derivatives and compare to the analytical solutions.

    Copyright:

    © All Rights Reserved
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 3

    Computational Solid Mechanics (151-0519-00L) October 11, 2017

    Fall 2017 Prof. Dennis M. Kochmann, ETH Zürich

    Project #2

    assigned: Thursday, October 12th, 2017


    due: Thursday, October 26th, 2017, 17:00
    please hand in to one of the TAs in LEE N203

    Review: tensor vs. vector notation

    Since there is no straight-forward way to deal with higher-order tensors in code, we use vector notation. That
    is, here and in the following, our code will work with a displacement gradient vector and stress vector. For
    example, in 2D we convert tensors into vectors according to
       
      u1,1   σ11
    u1,1 u1,2 u1,2  σ 1,1 σ 1,2
     σ12 
    β = ∇u = → β̃ =  , σ= → σ̃ =  .
    u2,1 u2,2  u2,1  σ2,1 σ2,2  σ21 
    u2,2 σ22

    The finite-deformation version is analogous with F̃ and P̃ . The fourth-order incremental tangent tensor now
    becomes a simple matrix defined via
    ∂ σ̃
    C̃ = and, e.g., C̃11 = C1111 , C̃12 = C1112 , C̃13 = C1113 , etc.
    ∂ β̃
    That is, according to the stress and strain vectors where indices are ordered as {(11), (12), (21), (22)}, the
    indices of C̃ij (with i, j = 1, . . . , 4 in 2D) stand for a pair of indices according to the above order.
    For convenience, utility functions are available for the simple conversion between vector and standard notation.
    This way we can first define Cijkl in the classical way and then convert to the above vector form. Of course,
    you are welcome to use those utility tools in your code.
    Finally, recall that we had shown in class that (the finite-deformation version is analogous)
    ∂W ∂W ∂σij ∂ σ̃i
    σij = ⇔ σ̃i = and Cijkl = ⇔ C̃ij = .
    ∂(∇u)ij ∂ β̃i ∂(∇u)kl ∂ β̃j

    Problem 1: material model, linear elasticity (30 points).

    Let us implement the 3D isotropic linear elastic material model from Project #1, defined by the energy density
    λ 1h i
    W (∇u) = (tr ε)2 + µ ε · ε with ε= ∇u + (∇u)T , β = ∇u.
    2 2

    Let us implement the linear elastic material model as a C++ class that has the following components:

    • Use typedef to define the vectors/matrices DisplacementGradient, Stress, Strain, and TangentMatrix
    as Eigen::Matrix<double,?,?> with their correct dimensions in 3D.

    1
    Computational Solid Mechanics (151-0519-00L) October 11, 2017
    Fall 2017 Prof. Dennis M. Kochmann, ETH Zürich

    • The class constructor should receive two doubles as input (E and ν) and store those within the class
    as private members. (Recall the conversion between λ, µ and E, ν derived in Project #1.)

    • The class should have a method computeEnergy, which receives the DisplacementGradient vector
    β̃ and returns the energy density W as a double.

    • The class should have a method computeStress, which receives a DisplacementGradient vector β̃
    and returns a Stress vector σ̃.

    • The class should have a method computeTangentMatrix, which receives a DisplacementGradient


    vector β̃ and returns a TangentMatrix C̃.

    • The class should have a method computeStrain, which receives a DisplacementGradient vector β̃
    and returns the strain vector ε̃ as Strain.

    Problem 2: material model, hyperelasticity (35 points).

    Let us repeat the above implementation for the compressible Neo-Hookean material model from Project #1,
    which had the energy density
    µ λ
    W (F ) = (tr C − 3) + (ln J)2 − µ ln J.
    2 2
    Now working with β̃, P̃ and C̃, let us implement this material model as a new class having the same methods
    as the linear elastic one implemented above.

    Problem 3: derivative test (35 points).

    Let us verify the correct implementation of the above material models by checking if the stresses are indeed
    the correct derivatives of the energy, and analogously the tangent matrix the derivative of the stresses. To
    this end, we use the forward-Euler finite-difference approximation
    f (x + h) − f (x)
    f 0 (x) ≈ ,
    h
    where |h|  1 is a small perturbation. Let us implement a function that verifies if P̃ = ∂W/∂ β̃ and
    C̃ = ∂ P̃ /∂ β̃ are implemented correctly, as follows.
    Let your new function pick a random (reasonable) β̃ and numerically compute each component of P̃ and
    C̃ using the above finite difference approximation (by perturbing each component of β̃ by a small h, one
    component at a time). Compare those numerical values to the stress and stiffness values returned by the
    material model. The material model passes the test if the relative error between numerical and analytical
    derivatives (e.g., ||P̃analytical − P̃numerical ||/||P̃analytical ||) is below a specified tolerance tol.
    Use the test function to verify the correct implementation of your own linear elastic and Neo-Hookean models.
    For testing, use h = 10−6 and tol = 10−4 and pick some positive material constants.

    total: 100 points

    2
    Computational Solid Mechanics (151-0519-00L) October 11, 2017
    Fall 2017 Prof. Dennis M. Kochmann, ETH Zürich

    Hints for the derivative test:


    For example, the components of the stress vector σ̃ can be checked as follows:

    • Pick a random β̃ 6= 0.

    • Compute the energy W0 = W (β̃).

    • Next for each component i of β̃:

    Compute a perturbed β̃ 0 such that β̃ 0 = β̃ except for one perturbed component β̃i0 = β̃i + h.

    Compute the perturbed energy W 0 = W (β̃ 0 ).

    Compute the stress component


    W 0 − W0
    σ̃i,numerical = .
    h

    • Repeat the above procedure for all components i of β̃ to compute all components of σ̃.

    The calculation of C̃ follows analogously: compute σ̃0 , then perturb each component j of β̃ 0 and compute σ̃ 0
    for each j. Then Cij,numerical = (σ̃0,i − σ̃i0 )/h.

    You might also like