You are on page 1of 11

Finite Elements in Analysis and Design 130 (2017) 1–11

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design

journal homepage:

Determining the reference geometry of plastically deformed material body MARK

undergone monotonic loading and moderately large deformation

Jia Lua, , Linlin Lib
Department of Mechanical and Industrial Engineering, The University of Iowa, Iowa City, IA 52242-1527, USA
Suzhou Industrial Park Institute of Vocational Technology, Suzhou, Jiangsu, China


Keywords: This article presents an inverse method for predicting the reference geometry of plastically deformed material
Inverse method body. The reference configuration is found by solving an elastic-plastic boundary value problem to determine an
Inverse elastoplastic problem inverse deformation that maps the spatial material points back to their reference positions. Rate-type
Reference geometry elastoplastic constitutive laws are employed in the inverse analysis. When the stress exceeds the yield limit,
Plastic deformation
the plastic flow is invoked and plastic variables are predicted. The ensuing stress field satisfies equilibrium and
yield condition. However, the loading history is replicated only approximately and therefore the reference
configuration is approximately recovered. The method is limited to a certain family of deformations. In this
work, we restrict the method to problems involving monotonic loading and moderately large deformations.
Numerical examples demonstrate that the method is effective and reasonably accurate for such problems.

1. Introduction membrane theory to describe the mechanical behavior of thin metal

sheets and implemented a one-step scheme to obtain the solution
Finding the reference geometry of a finitely deforming material [20,19,26]. Bending effect was incorporated by using shell theory
body is of great interest to many engineering applications. For elastic [21,25] and three dimensional constitutive laws [27]. These works
material, this problem has been well-studied. Yamada [1] and mostly adopted the deformation theory of plasticity which assumes that
Govindjee et al [2,3] pioneered an inverse method that directly solves each material point undergoes proportional loading and the axes of the
the equilibrium boundary value problem for the reference configura- strain are fixed. To better capture the history effect, multi-step schemes
tion, and this approach has led to finite element implementations that which admitted the proportional loading assumption stepwise were
are similar to standard forward elements [2–8]. Structural inverse introduced [18,21,25]. Lately this inverse approach has been utilized in
problems have also been investigated [9–12]. Recently, the inverse forging applications [28].
method finds applications in biomedical analysis to deal with problems The inverse formulation based on the deformation theory of
for which only deformed configurations are known at the onset plasticity was found to give good strain estimation but poor stress
[13,10,14,6,15]. It was reported that, for some biological systems, the prediction. To improve the stress estimation a pseudo-inverse ap-
inverse method also helps to address the issue of lack of information of proach was developed by Guo et al [29–33]. In this approach,
material properties [16,17]. geometrically realistic intermediate configurations were introduced.
Theoretically, elastoplastic deformations are history dependent and The inverse method was used to adjust the intermediate configurations
thus the inverse problem is not well-posed. The inverse solution is not succeedingly starting from the final configuration. The flow theory of
unique unless the loading history or the plastic strain in the deformed plasticity and also damage theory were utilized in the inverse update
state is known. Nonetheless, there has been a strong practical interest process.
in the inverse elastoplastic problems. In the sheet metal forming The contributions cited above demonstrated the usefulness of
community, a large body of work has been devoted to determining inverse analysis in forming applications. However, since the deforma-
the initial blank geometry of workpieces based on a target final tion theory was adopted mostly, the material was treated algorithmi-
geometry using inverse analysis [18–25]. The hallmark of these works cally as elastic. Also, certain prior information about the shape of the
is to solve the equilibrium boundary value problem inversely to initial and/or intermediate configurations was incorporated in the
determine the initial geometry. Early developments mostly employed analysis and these conditions are pertinent only to forming or forging

Corresponding author.
E-mail address: (J. Lu).
Received 22 June 2016; Received in revised form 27 November 2016; Accepted 10 February 2017
0168-874X/ © 2017 Published by Elsevier B.V.
J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

applications. It remains unclear how well an elastoplastic deformation f≔∂xΦ(x , t ) is the inverse of the forward deformation gradient
can be inversely predicted in a general setting. An attempt towards F≔∂ Xφ(X, t ). If, during the inverse solution process the stress is found
solving inverse elastoplastic problem using the algorithmic framework to exceed a given elastic limit, the problem cannot be treated as elastic.
of [1,2] was reported in Germain et al. [34]. The authors demonstrated An elastoplastic constitutive law is then invoked.
that when the plastic variables are known a priori the inverse approach We focus on rate-independent plastic behavior described by the
can be used to recover the reference geometry. This finding under- finite strain elasto-plasticity theory presented in, e.g. [37–39]. This
scored the fact the elastoplastic inverse problem is ill-posted. The theory treats an elastoplastic material as a family of elastic materials
inverse solution is not unique unless the plastic variables, or alter- parameterized by plastic variables. In particular, we will utilize
natively the loading path, are given. In reality, the plastic variables are constitutive forms that take a metric-like plastic deformation tensor
unknown at the onset. To cope with this issue, the authors introduced a Cp as a primitive plastic variable [40,41]. For the model employed later
recursive process [35,36] consisting of iterative loops of inverse and in the simulation, the plastic variables include the tensor Cp and an
forward analyses using the inverse computation to predict the refer- equivalent plastic strain, ep. At fixed plastic variables, the stress is
ence geometry and the forward analysis to engage the plastic flow. In given by a (hyperelastic) function of Cauchy-Green deformation tensor
each loop, an inverse step is carried out first, with plastic variables fixed C and the plastic variables. The admissible stress lies in a (convex)
at their current value. This is followed by a forward step applied on the region in the stress space bounded by a yield surface. If the stress tends
predicted reference geometry to generate the plastic variables. In to penetrate the yield surface, plastic flow is activated to bring the
essence, this is an operator-splitting scheme which splits the inverse stress back to the yield surface.
problem into two sub-problems: determining the reference geometry In the numerical solution the elastoplastic constitutive equation is
and determining the plastic variables. The plastic flow is introduced in handled in essentially the same manner as in the forward analysis.
the forward step to recover the loading path. Given a predicted reference configuration at time step n + 1, the
In this work, we explore a direct approach of inverse analysis for deformation tensor is computed from
elastoplastic materials governed by the flow theory of plasticity. The
fn +1 = ∂xΦn +1, Cn +1 = f −n +1
T −1
f n +1 (2)
given information is the current geometry, applied forces and materi-
al's constitutive law. The unknowns are the reference geometry and A trial stress STr
is computed using the strain Cn+1 and the current
n +1
plastic variables in the current state. We propose to use the elasto- plastic variables. If the trial stress satisfies the yield condition, then
plastic constitutive law directly in the framework of [2,3]. When the Sn +1 = STr
n +1 and plastic variables remain intact. Otherwise, a return
stress is found to exceed the yield limit, the plastic flow is invoked and mapping is performed to project the stress back and update the plastic
the plastic variables are predicted. The analysis will yield a reference variables. Note that the plastic tensor Cp is a field variable defined in
configuration and a set of plastic variables. The ensuing stress satisfy the (iteratively determined) reference configuration and it predicted
equilibrium and yield condition. However, since the analysis starts with along with the latter. The yield condition is enforced at every step and
the current geometry, the actual strain history is not replicated and hence the stress satisfies the yield condition at the end.
thus the analysis can only yield approximate solutions. The premise is, Although the treatment of elastoplastic response is algorithmically
if the inverse “loading path” is somehow close to the actual loading the same as in the forward analysis, there is a fundamental difference:
history, the inverse solution is expected to be reasonably accurate. The the stain history (e.g. the loading path) is inferred from the inverse
method, therefore, is not a general approach but limited to a certain deformation. In general, the inverse strain history cannot be the same
family of elastic-plastic deformations for which the strain history can as the forward one, and thus, the actual loading path is not exactly
be reasonably replicated in the inverse process. Here, we focus on replicated. The difference in the strain history is the root cause for the
problems involving monotonic loading and moderately large deforma- inverse solution to be approximate. A pre-requisite for the method to
tions. The rationale will be explained later. work is that the inverse loading path remains somehow close to the
The remainder of the article is organized as follows. To set the forward one. Under this circumstance, we expect that the predicted
stage, the inverse elastoplastic boundary value problem is briefly reference configuration and plastic variables are reasonably accurate.
described in Section 2. A finite element formulation is outlined in Below, we speculate some conditions for the loading paths to be close.
Section 3. The formulation utilizes existing material models and A rigorous error analysis is beyond the scope of this work.
therefore only the element level computation is presented. Numerical
examples are presented in Section 4. 1. Monotonicity of loading. We first require the loading to be mono-
tonic. To the leading order, the strain path in a monotonic loading is
a line between the starting and end points in the strain space. This
2. Inverse elastoplastic problem path depends largely on the end points, and thus is more likely to be
reproduced in the inverse process. Physically, since there is no
We seek to find a reference configuration of a plastically deformed unloading or reverse plastic loading, the history influence is less
material body based on the knowledge of (1) a deformed configuration prominent. Note that although the monotonicity condition seems
of the body, (2) the applied body force, (3) boundary conditions restrictive, there is a wide range of practical problems that can fit
including Cauchy traction data and displacement data, and (4) the into this category. For example, most forming or casting processes
constitutive law of the material. As alluded earlier, the approach is to are essentially monotonic; the deformation increases continuously,
determine the inverse deformation by solving the following boundary bearing little or no reversal loading.
value problem: find the inverse motion Φ: Ω ↦ ) ∈ R3 such that 2. Moderately large deformation. Another restriction is that the
deformation cannot be arbitrarily large. In elastoplastic analysis
σij, j + bi = 0 in Ω
the solution is typically obtained incrementally. In the inverse
Φ=Φ on ∂Ωu approach, the predicted configuration at an intermediate step tn is
σijnj = ti on ∂Ωt . a partially recovered reference configuration, in contrast to a
partially deformed current configuration in the forward analysis.
Here σ is the Cauchy stress, bi is a component of the body force per unit The stress in the forward analysis satisfies equilibrium on the
current volume, ti is a component of the prescribed boundary traction, partially deformed current configuration, whereas the stress in the
Ω is the given current configuration, and ) is the sought reference inverse analysis always achieves equilibrium on the given, full
configuration. The inverse deformation Φ(x , t ) is the kinematic inverse current configuration. The intermediate stresses are different from
of the forward deformation φ(X, t ) at any fixed time. The gradient the equilibrium perspective. If the deformation is too large, the

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 1. One-step results. (a) The reference (unshaded) and current configuration (shaded) in the loop analysis; (b) Configuration error plotted on the predicted reference; (c) Forward
prediction of J2; (d) Forward prediction of the Equivalent plastic strain, ep. The inverse solutions of ep and J2 are exactly the same as the forward counterparts.

inverse and forward stress histories could be too far apart to render a ΔG = ∫Ω δ E: jep : [ − C∇UF]S − 2δ E: [j∇UFS]S
good result.
+ (δ E: j S)tr(GradU) dv (4)
As a side note, if an elastoplastic problem can be solved in a single S
Here, [·] denotes the symmetric part of the tensor [·]. The linearized
load step, the inverse solution is expected to be exact, in the sense that
week form can also be represented in the following spatial form [8]
it exactly reverts the forward displacement and recovers the plastic
variables. This is because, algorithmically, the stress response is driven ΔG = ∫Ω ∇S δ u: ep : [ − F∇U]S − 2∇S δ u: [F∇Uσ ]S
by the end strain and there is no history effect. There is also no
limitation on the magnitude of deformation as long as the solution + (∇S δ u: σ )tr(GradU) dv (5)
converges. ep ep T T
where  is the spatial tangent tensor,  = jFF F F , and σ is the

Cauchy stress, σ = jFSFT . Introducing the following B-matrices:

Nel Nel Nel
3. Finite element formulation ∇S δ u = ∑ BI δ uI , [F∇U]S = ∑ BFI UI , [F∇Uσ ]S = ∑ BGI [σ ]UI
I =1 I =1 I =1 (6)
The inverse finite element formulation is based on the weak form we obtain the expression of the element stiffness matrix

G≔ ∫Ω δ E: j S dv − < fext, δ u > = 0 KIJ = ∫v − BTI epBFJ − 2BTI BGJ [σ ] + (BTI σ )GradT NJ dv
e (7)
⎡ ∂N ∂N ∂N ⎤
Here, E is the Cauchy-Green strain, S is the second Piola-Kirchhoff Here GradT NJ = ⎢ ∂XJ ∂XJ ∂XJ ⎥. The explicit expressions of the B-opera-
⎣ 1 2 3⎦
stress, δE is the variation of E induced by a kinematically
tors can be found in Ref. [8].
admissible variation δu to the current configuration, and
As a working plasticity model, the finite strain isotropic plasticity
<fext , δ u > = ∫ b·δ u dv + ∫ t ·δ u da . The stress power is integrated
Ω ∂Ω t theory in Ref. [42] was employed in this work. The constitutive
over the given current configuration Ω and thus the volume factor description is based on a multiplicative decomposition of deformation
j = detf is included. The operator ‘:’ stands for tensor contraction, gradient, leading to the elastic Finger tensor be = FC−1 T
p F . The strain
A: B = AIJ BIJ . The weak form yields a set of nonlinear algebraic energy is a quadratic function of the (logarithmic) elastic strain
equations for the nodal values of Φ , which is subsequently linearized 1
ϵe = 2 logbe . The basic equations are summarized below.
with respect to the increment of Φ . Let U: Ω ↦ R3 be an incremental
inverse displacement. The increment of G is computed using Gateaux Stress function τ = 2μϵe + λ trϵe1
differential ΔG(Φ, ·) = dε G(Φ + εU, ·), giving 2
ε=0 Yield condition f = ∥ τdev ∥ − σy
ΔG = ∫Ω δ E: j ∂∂ES : ΔE + Δ(δ E): j S + (δ E: S)Δj dv (3) or f = ∥ τdev ∥ +
trτ −
3 3
In writing ΔG as above it is assumed that the external forces do not ∂f 2
Flow rule 3 vbe = − 2γ nbe , n≔ e˙p = γ
depend on the reference configuration. If they do, their increments ∂τ 3
should be considered. The material tangent tensor ∂S is denoted as ep
∂E Hardening rule σy = σy0 + hisoep (8)
hereafter. Invoking δ E = FT ∇S δ uF and carrying out the computations,
˙ T
p F , τ = FSF is the Kirchhoff stress and τdev is the
Here 3 vbe = F C−1 T
it was found that [8]

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 2. Multi-step results. (a) Current configuration (shaded) and the predicted reference (mesh); (b) Configuration error; (c) Forward ep; (d) Inverse ep; (e), Forward J2; (f) Inverse J2;
(g). ep error; (h). J2 error.

deviatoric part of τ . The flow rule leads to the following plastic update
Table 1
Errors of configuration and some state variables. algorithm in the logarithmic strain space:
Case err (σ ) err(ep) err(J2) err(conf) ϵen +1 = ϵen +1 − Δγn

Tr 1 −1 T
Cantilever, one step 0.00 0.00 0.00 0.00 where ϵen +1 = 2 log(Fn +1C pnF n +1). The return mapping algorithm was
Cantilever, multi-step 2.65E-2 6.47E-3 5.35E-3 6.85E-3 detailed in Ref. [42] and further discussed in Ref. [39]. Briefly, the
Bar, 90° rotation 3.21E-2 4.13E-2 1.81E-2 2.52E-2 return mapping is carried out in the principal space of be . In this space,
Bar, 180° rotation 7.26E-2 8.32E-2 4.48E-2 3.31E-2
the computational structure of the infinitesimal J2 plasticity is com-
Disk compression 2.76E-2 7.74E-2 9.76E-3 1.08E-2
Thermal loading 3.37E-2 4.83E-3 7.14E-3 7.34E-3 pletely preserved. The material function returns the Cauchy stress σ
and the algorithmic spatial tangent tensor ep , which are directly

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

passed to the element level computation.

4. Numerical examples

Loop tests were conducted to evaluate the inverse method. In a loop

test, a forward analysis was conducted to produce a current configura-
tion and a set of plastic variables. Subsequently, an inverse analysis was
performed using only the current configuration as input. The plastic
variables were initialized to zero, and re-predicted. The forward and
inverse solutions were compared. Scaled error norms were computed
to quantify the difference in state variables:

∫Ω ∥ Δ(·)∥ dv
err (·) =
∫Ω ∥(·)∥ dv

Here Δ(·) stands for the difference between the forward and the inverse
solutions. The configuration error was defined slightly differently by
normalizing the difference against the forward displacement u :

∫Ω ∥ XGiven − XPredicted ∥ dv
err (conf ) =
∫Ω ∥ u ∥ dv

The contours of the errors of equivalent plastic strain ep, the stress
invariant J2, and the configuration were plotted in each example. The
norms are reported in a single table at the end of this section.

4.1. Cantilever

A cantilever problem was introduced primarily to demonstrate the

difference between single-step and multiple-step solutions. This was
the only example wherein the solution converged in a single load step.
The length and height of the beam were 10 and 1 units of length,
respectively. A downward force of 0.02 units was applied at the right
Fig. 3. Histories of plastic strain and stress component σ1 at the Gauss point in the left- tip. The left bottom node was fixed, and other nodes on the left edge
upper corner.
were allowed to move vertically. The material parameters were set to be

Fig. 4. Twisting of the bar at two levels of deformation. Upper row: 90° rotation. Lower row: 180° rotation. Left column: deformed configurations; Middle column: Predicted reference
configurations; Right column: configuration error plotted on the reference configuration.

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 5. Comparison of forward and inverse solutions. (a) Forward J2; (b) Inverse J2; (c) J2 error; (d) Forward equivalent plastic strain, ep; (e) Inverse ep, (f) ep error.

4.2. Twisting of a prismatic bar

A square cross section bar fixed at one end and subjected to a

torque at the other end was considered. The length of the bar was 10
units and the cross section width was 1 unit. A forward analysis was
performed to generate a deformed configuration. This analysis was
displacement driven; a rigid rotation around the longitudinal axis was
applied on the rotating end with the longitudinal displacement con-
strained. The reaction forces at the peak twist were recorded. The
inverse analysis was force-driven; the rotational displacement at the
loading end was replaced by the nodal reaction forces recorded in the
forward analysis. The forces were proportionally applied keeping theirs
directions fixed. In effect, they generated a torque that was propor-
tionally applied to the bar. The other end remained fixed. In both
analyses the load was divided into 100 steps. The material parameters
were the same as the previous example. Two levels of deformation, a
90° and a 180° rotation, were considered. The maximum (logarithmic)
Fig. 6. Schematics of example 4.3. strains were found to be 0.0625 and 0.1260, respectively.
The given and predicted configurations are presented in Fig. 4. It
μ = 38.46, λ = 57.69 σy0 = 0.2, hiso = 5 can be seen that the reference configuration was accurately recovered
in both cases. The configuration error is slightly skewed towards the
All units are assumed to be consistent. loading end. The plastic strain ep and the invariant J2 are shown in
A single load step. When the problem was solved using a single load Fig. 5 along with the error contours. Note that forward solutions of
step, an exact reversion was obtained. Fig. 1 shows the configurations state variables are symmetric in the longitudinal direction whereas the
and contours of ep and J2. The inverse and the forward solutions are inverse solutions are not. This is expected because the boundary
exactly the same therefore only the forward results are plotted. The conditions are no longer symmetric in the inverse analysis. The error
maximum strain, ∥ 2 logC ∥max , is 0.0766. The vertical displacement at norms are reported in Table 1. The error norms at 180° rotation
the tip is 3.58 units of length. approximately double those at 90° rotation. When the bar was rotated
Multiple load steps. The same problem was solved in ten load steps. further, the error increased and a linear correlation was observed. The
The solution is not exact; the configuration error is shown in Fig. 2 derails are omitted.
along with contours of state variables and their differences. The error
norms are reported in Table 1.
To show the differences in solution paths, the histories of ep and σ1 4.3. Compression of a disk
at a Gauss point are plotted in Fig. 3. Notably, the plastic strains are
closer near the two ends of the loading path, but more distanced in the A circular disk of unit radius was compressed by two sets of rigid
middle. The stress exhibits a similar trend, although the difference at blocks to a maximum displacement of 5% of the radius (Fig. 6). Due to
the high strain end is larger than that of the lower strain end. symmetry, only a quarter of the disk was analyzed. The contact

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 7. Disk compression: (a) Current configuration (shade) and the reference configuration (mesh); (b) Configuration error; (c) Forward ep; (d) Inverse ep; (e) Forward J2; (f) Inverse
J2; (g) ep error; (h) J2 error.

condition between the blocks and the disk was enforced using an μ = 344.83, λ = 3103.45,
augmented Lagrangian method. Symmetry conditions were applied on
σy0 = 10.00, hiso = 1.00
the left and upper edges of the analysis domain. A plane strain
condition was imposed. The following material parameters were used:
In the forward analysis the displacement was applied (on the blocks)

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 8. Thermal plastic deformation. Upper row: Inverse prediction of the reference geometry. (a) The targeted configuration; (b) The intermediate configuration; (c) The predicted
reference configuration. Lower row: forward verification. (d) Reference configuration, the same geometry as (c); (e) The configuration after elastic-plastic heating; (f) The end
configuration after elastic cooling; (g) Configuration error – the difference in geometry between (a) and (f). The lower row is supposed to be the inverse process of the upper row.

in ten steps. At each step, the nodal contact forces exerted on the disk The reference configuration was recovered to a good accuracy. The
were recorded. The deformed configuration at the peak compression was predicted configuration is show in Fig. 7 along with some state
used as the input geometry for the inverse analysis. The inverse solution variables. Error norms are reported in Table 1. The maximum
was transformed into force-driven; the contact constraints were replaced logarithmic strain for this problem is 0.1810. The average strain is
by the contact forces retrieved from the forward analysis. The symmetry approximately 0.06.
boundary condition on the left and upper edges remained intact. The
contact forces were applied step by step in the sequence they were 4.4. Thermal plastic deformation
recorded. In this problem, it was proved critical to use the contact force
history instead of the reaction at the peak load. When the latter was used, A cylindrical tube under thermal loading was analyzed. The tube
the inverse solution was less accurate. was heated from a room temperature of T0 = 25 °C to 425 °C, and

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

Fig. 9. Comparison of state variables in the heated configuration. (a) Inverse ep; (b) Forward ep; (c) Inverse J2; (d) Forward J2; (d) ep error; (f) J2 error.

cooled to the room temperature. The top and bottom surfaces were β 2
f = ∥ τdev ∥ + trτ − g(T )σy
fixed. During the heating phase plastic deformation occurred, resulting 3 3
in changes in both shape and volume. A thermal-elastoplastic consti-
tutive law was adopted. The stress function reads The return mapping algorithm of this model can be found [39]. The
following material parameters were used:
τ = 2g(T )μϵdev + g(T )λ(logJe − α(T − T0 ))1
μ = 357, λ = 1428, α = 0.0003,
where T is the temperature, α is the thermal expansion coefficient, σy0 = 5, hiso = 10, β = 0.05
Je = detbe , and g(T) is a function describing the temperature depen-
T − T0
dence of the elastic properties. Here, g(T ) = e− 800 . To accomondate Unlike the other examples which took a loaded configuration as given,
plasticity-induced volume change, a Drucker-Prager model was em- here we assumed that the configuration after the heating-cooling cycle
ployed: was known (or targeted). This configuration was plastically deformed
and residually stressed, but not loaded. The goal was to design the
reference geometry so that a target geometry was obtained after the

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

heating-cooling cycle. The target geometry was a perfect cylindrical suggests that the method may be applicable to up to 10–15% strain for
tube; the inner and out radii and the length were 1.2, 2.4, and 5 units, mechanical problems and higher for thermal strain problems.
respectively. The loading cycle consisted of an elastic-plastic heating
phase over which plastic deformation was accumulated, followed by an Acknowledgements
elastic cooling phase. The cooling phase was elastic because the stress
level reduced and thus stress points moved towards the interior of the The second author (LL) was supported by Jiangsu Overseas
elastic region. No new plastic deformation was generated in this phase. Research and Training Program for University Prominent Young and
A two-phase analysis was employed to predict the reference Middle-aged Teachers and Presidents. This work was initiated when
geometry. The first phase was a forward elastic heating from the target she was visiting The University of Iowa.
configuration, with plastic variables and residual stress set to zero. This
analysis was to revert the contracting deformation during cooling and it References
yielded an intermediate configuration which approximated the one
after the elastoplastic heating from the reference state. The second [1] T. Yamada. Finite element procedure of initial shape determination for rubber-like
phase was an inverse elastoplastic phase applied on the intermediate materials. Technical Report No. 20, Res. Lab. Eng. Mat. Tokyo Inst. Tech., 1995.
[2] S. Govindjee, P.A. Mihalic, Computational methods for inverse finite elastostatics,
configuration. The reference geometry was determined in this phase. Comput. Methods Appl. Mech. Eng. 136 (1996) 47–57.
To verify the inverse results, a forward analysis was carried out on [3] S. Govindjee, P.A. Mihalic, Computational methods for inverse deformations in
the predicted reference configuration with zero plastic variables. The quasi-incompressible finite elasticity, Int. J. Numer. Methods Eng. 43 (1998)
geometry and plastic variables at the end of the heating phase were [4] J. Lu, X. Zhou, M.L. Raghavan, Computational method of inverse elastostatics for
predicted. Subsequently, the body was elastically cooled, and the end anisotropic hyperelastic solids, Int. J. Numer. Methods Eng. 69 (2007) 1239–1261.
geometry was obtained. The end configuration was compared to the [5] Víctor D. Fachinotti, Alberto Cardona, Philippe Jetteur, Finite element modelling of
inverse design problems in large deformations anisotropic hyperelasticity, Int. J.
target, and the plastic variables were also compared to those predicted Numer. Methods Eng. 74 (6) (2008) 894–910.
in the inverse analysis. The heating and cooling phases were divided [6] M.W. Gee, C. Reeps, H.H. Eckstein, W.A. Wall, Prestressing in finite deformation
into 100 and 10 steps, respectively. At each step, the temperature was abdominal aortic aneurysm simulation, J. Biomech. 42 (2009) 1732–1739.
[7] Sandrine Germain, Philipp Landkammer, Paul Steinmann, On inverse form finding
applied uniformly over the body. Note that, although the entire loading
for anisotropic hyperelasticity in logarithmic strain space, Int. J. Struct. Chang.
cycle was not monotonic, the phase where plastic deformation took Solids 2 (2012) 1–16.
place was. For this example the maximum (logarithmic) strain in the [8] J. Lu, L. Li, On referential and spatial formulations of inverse elastostatic analysis,
heating phase was 0.2771. Comput. Methods Appl. Mech. Eng. 310 (2016) 189–207.
[9] X. Zhou, J. Lu, Inverse formulation for geometrically exact stress resultant shells,
Fig. 8 shows the configurations involved in the analysis and the Int. J. Numer. Methods Eng. 74 (2008) 1278–1302.
error contour. Notably, the target configuration was accurately recov- [10] J. Lu, X. Zhou, M.L. Raghavan, Inverse method of stress analysis for cerebral
ered. Fig. 9 shows the state variables ep and J2 and the error contours. aneurysms, Biomech. Model. Mechanobiol. 7 (2008) 477–486.
[11] Alejandro E Albanesi, Víctor D Fachinotti, Alberto Cardona, Inverse finite element
The errors norms are on the order of 10−3–10−2, as recorded in Table 1. method for large-displacement beams, Int. J. Numer. Methods Eng. 84 (10) (2010)
5. Concluding remarks [12] Víctor D Fachinotti, Alejandro E Albanesi, José M Martínez Valle, Inverse finite
element modeling of shells using the degenerate solid approach, Comput. Struct.
157 (2015) 89–98.
Determining the reference geometry of plastically deformed mate- [13] J. Lu, X. Zhou, M.L. Raghavan, Inverse elastostatic stress analysis in pre-deformed
rial body is of significant interest to engineering applications. This biological structures: Demonstration using abdominal aortic aneurysm, J. Biomech.
40 (2007) 693–696.
inverse problem is challenging because it is not well-posed. The present [14] X. Zhou, M.L. Raghavan, R. Harbaugh, J. Lu, Patient-Specific wall stress analysis in
contribution tests an inverse method which utilizes elastoplastic cerebral aneurysms using inverse shell model, Ann. Biomed. Eng. 38 (2) (2010)
constitutive laws directly in the framework of [1,2]. The method works 478–489.
[15] M.W. Gee, Ch. Forster, W.A. Wall, A computational strategy for prestressing
under the premise that the loading path in the inverse analysis is close
patient-specific biomechanical problems under finite deformation, Int. J. Numer.
to the actual path. It is unclear in general how this condition can be Methods Biomed. Eng. 26 (2010) 52–72.
guaranteed. In this work, this condition is safe-guarded by restricting [16] J. Lu, S. Hu, M.L. Raghavan, A shell-based inverse approach of stress analysis in
the method to problems of monotonic loading and moderately large intracranial aneurysms, Ann. Biomed. Eng. 41 (7) (2013) 1505–1515.
[17] K. Miller, J. Lu, On the prospect of patient-specific biomechanics without patient-
deformations. specific properties of tissues, J. Mech. Behav. Biomed. Mater. 27 (2013) 1540166.
We used four examples to evaluate the method. Altogether they [18] S.A. Majlessi, D. Lee, Development of multistage sheet metal forming analysis
showed that the inverse method is effective and reasonably accurate for method, J. Mater. Shap. Technol. 6 (1) (1988) 41–54.
[19] Y.Q. Guo, J.L. Batoz, J.M. Detraux, P. Duroux, Finite element procedures for strain
the said family of problems. The first example showed that when the estimations of sheet metal forming parts, Int. J. Numer. Methods Eng. 30 (8)
load was applied in a single step, the exact inverse of the forward (1990) 1385–1401.
solution was obtained. When the load is applied in one-step, the [20] S.A. Majlessi, D. Lee, Deep drawing of square-shaped sheet metal parts, part 1:
finite element analysis, J. Eng. Ind. 115 (1) (1993) 102–109.
solution is driven by the end strain and thus, algorithmically, there is [21] Jean-Louis Batoz, Ying Qiao Guo, Frederic Mercier, The inverse approach with
no history effect. When the load is applied incrementally, the inverse simple triangular shell elements for large strain predictions of sheet metal forming
solution is approximate. The accuracy of the solution depends on the parts, Eng. Comput. 15 (7) (1998) 864–892.
[22] Y.Q. Guo, J.-L. Batoz, H. Naceur, S. Bouabdallah, F. Mercier, O. Barlet, Recent
magnitude of the deformation. In all example, it was shown that the
developments on the analysis and optimum design of sheet metal forming parts
errors increased proportionally with the deformation. The third using a simplified inverse approach, Comput. Struct. 78 (1) (2000) 133–148.
example also underscored the importance of prescribing the correct [23] Seung Ho Kim, Se Ho Kim, Hoon Huh, Finite element inverse analysis for the
design of intermediate dies in multi-stage deep-drawing processes with large aspect
force history. It was imperative to apply the contact force history
ratio, J. Mater. Process. Technol. 113 (1) (2001) 779–785.
instead of using the contact force at the peak contact stage. The last [24] C.H. Lee, H. Huh, Blank design and strain estimates for sheet metal forming
example suggested that the method can be well-suited to thermal processes by a finite element inverse approach with initial guess of linear
plasticity problems. For such problems the deformation is driven by the deformation, J. Mater. Process. Technol. 82 (1) (1998) 145–155.
[25] C.H. Lee, J. Cao, Shell element formulation of multi-step inverse analysis for
thermal strain; if the thermal strain is monotonic the method appeared axisymmetric deep drawing process, Int. J. Numer. Methods Eng. 50 (3) (2001)
to be effective even for fairly large deformations. 681–706.
The peak strains in these examples ranged from 6–27%. The thermal [26] C.H. Lee, H. Huh, Estimation of shape and non-shape parameters in sheet metal
forming processes with inverse finite element analysis. A.A. Balkema Publishers,
problem had the highest peak strain of 27%. The other three examples Simulation of Materials Processing: Theory, Methods and Applications(USA),
had strains between 6–18%. In all three problems, the accuracy of the (1998), 793–799.
solution was found to deteriorate when the deformation further increased. [27] C.H. Lee, H. Huh, Three dimensional multi-step inverse analysis for the optimum
blank design in sheet metal forming processes, J. Mater. Process. Technol. 80
The allowable strain range is certainly problem-specific, but these example

J. Lu, L. Li Finite Elements in Analysis and Design 130 (2017) 1–11

(1998) 76–82. Institute of Physics, vol. 1353, pp. 1169, 2011.

[28] Ali Halouani, YM Li, Boussad Abbès, and Ying-Qiao Guo. An axi-symmetric inverse [35] Sandrine Germain, Michael Scherer, Paul Steinmann, On a recursive formulation
approach for cold forging modelling. in: Lecture Notes in Engineering and for solving inverse form finding problems in isotropic elastoplasticity, Adv. Model.
Computer Science: Proceedings of The World Congress on Engineering, 2010. Simul. Eng. Sci. 1 (2014) 1–10.
[29] Y.Q. Guo, Y.M. Li, F. Bogard, K. Debray, An efficient pseudo-inverse approach for [36] Philipp Landkammer, Sandrine Germain, Paul Steinmann, On inverse form finding
damage modeling in the sheet forming process, J. Mater. Process. Technol. 151 (1) for orthotropic plasticity, Comput. Assist Methods Eng. Sci. 20 (2013) 337–348.
(2004) 88–97. [37] P.M. Naghdi, A critical review of the state of finite plasticity, Z. für Angew. Math.
[30] Abel Cherouat, Y.Q. Guo, Khémaïs Saanouni, Y.M. Li, K. Debray, G. Loppin, und Phys. ZAMP 41 (3) (1990) 315–394.
Incremental versus inverse numerical approaches for ductile damage prediction in [38] P.M. Naghdi, J.A. Trapp, The significance of formulating plasticity theory with
sheet metal forming, Int. J. Form. Process. 7 (1–2) (2004). reference to loading surfaces in strain space, Int. J. Eng. Sci. 13 (9) (1975)
[31] Ali Halouani, Yuming Li, Boussad Abbès, Ying-Qiao Guo, Simulation of axi- 785–797.
symmetrical cold forging process by efficient pseudo inverse approach and direct [39] J.C. Simo, T.J.R. Hughes, Computational inelasticity 7, Springer Science &
algorithm of plasticity, Finite Elem. Anal. Des. 61 (2012) 85–96. Business Media, 2006.
[32] Ali Halouani, Yu Ming Li, Boussad Abbès, and Ying Qiao Guo. A pseudo inverse [40] C. Miehe, A constitutive frame of elastoplasticity at large strains based on the
approach with kinematically admissible intermediate configurations for the axi- notion of a plastic metric, Int. J. Solids Struct. 35 (30) (1998) 3859–3897.
symmetrical cold forging modelling. in: Advanced Materials Research, vol. 399, pp. [41] J. Lu, P. Papadopoulos, A covariant formulation of anisoptropic finite plasticity:
1832–1837. Trans Tech Publ, 2012. theoretical development, Comput. Methods Appl. Mech. Eng. 193 (2004)
[33] Ali Halouani, Yuming Li, Boussad Abbès, Ying-Qiao Guo, Efficient pseudo-inverse 5335–5358.
approach for damage prediction in cold forging process simulation, Int. J. Damage [42] J.C. Simo, Algorithms for static and dynamic multiplicative plasticity that preserve
Mech. 23 (8) (2014) 1168–1188. the classical return mapping schemes of the infinitesimal theory, Comput. Methods
[34] Sandrine Germain, Paul Steinmann, and Gary Menary. On inverse form finding for Appl. Mech. Eng. 99 (1992) 61–112.
anisotropic elastoplastic materials. in: AIP Conference Proceedings-American