You are on page 1of 1

AUTOMATIC DIFFERENTIATION FOR SOLID MECHANICS

1∗ 2
Andrea Vigliotti and Ferdinando Auricchio
1 2
Italian Aerospace Research Center, Capua, Italy - University of Pavia, Pavia, Italy

Abstract Equilibrium as a free energy minimum Conclusions


In this study we use Automatic differen- Thermodynamics ensures that free energy is stationary for equilibrium configura- Automatic differentiation allows to solve
tiation (AD) for stating and solving solid tions,therefore, using a FE discretization we can evaluate the free energy of solid as a solid mechanics problem by finding the
mechanics problems. Given a Finite Ele- function of the nodal Degrees of Freedom (DoFs) of the FE model, u, as follows equilibrium configurations directly as the
k
ment (FE) discretization of the domain, Z
Finte Elements
N N
E XW
k
minima of the free energy of the body
Ψ= (φ − b0) dV0 ========⇒ Ψ(u) = wl [φ − b0]rlk ,
X
we evaluate the free energy of the solid V0 discretization k=1 l=1 subjected to the given boundary condi-
as the integral of its strain energy density and equilibrium configurations can be found by solving the following tion. With this approach its only re-
and we make use of AD for directly ob- quired to explicitly calculate the free en-
2
taining the residual force vector and the ∂Ψ ∂Ψ ∂r ∂ Ψ ergy of the body as a function of its de-
δΨ = · δu = 0 ∀ δu ⇐⇒ r= = 0 with Kt = =
tangent stiffness matrix of the problem as ∂u ∂u ∂u ∂u∂u grees of freedom, while the residual force
the gradient and the Hessian of the free vector and the tangent stiffness matrix
energy respectively. The result is a re- Examples are obtained automatically as the gradi-
markable simplification in the statement ent and the Hessian of this quantity.
and the solution of complex problems in- This approach is particularly suited for
volving non trivial constraints systems dealing with multiphysics problems. In
and both geometrical and material non fact, this type of problems are extremely
linearities, as well as multiphysics prob- simple from the thermodynamics point of
lems [1]. view, since the energy is simply additive,
and it suffices to add the different contri-
(a) (b) (c) bution to free energy, while the complex
AD through dual numbers Axi-symmetric problem with and without internal volume constraint interplay between the different physics is
Automatic differentiation allows to numer- handled, in the background, by the auto-
Atop f matic differentiation.
ically evaluate the derivatives of an expres- e
sion with the same accuracy of the expres- d References
sion itself. In this study we use forward
c
mode AD which is based on dual numbers. h [1] A. Vigliotti and F. Auricchio, “Automatic
i3
Dual numbers, or hypercomplex numbers, b differentiation for solid mechanics,” Archives of
i1 i2 i1
are a number system with multiple imagi- Computational Methods in Engineering,
Abtm i3
a vol. 28, no. 3, pp. 875–895, 2021.
nary parts that can store and operate on the
(d) (e) (f)
derivatives of the real part. Once the dual Large deformation of a three dimensional solid Contact Information
numbers algebra has been implemented in ∗
a.vigliotti@cira.it
a programming language, it is possible to The above examples have been produced by implementing a second order dual number
automatically evaluate the derivatives of a system in the Julia programming language. The surce code of the above examples and of
function without any modification to the all the examples from the paper [1] are available from https://github.com/avigliotti/
source code. AD4SM.jl

You might also like