Professional Documents
Culture Documents
Finite Elements in Analysis and Design: Dong-Chan Lee, Chang-Soo Han
Finite Elements in Analysis and Design: Dong-Chan Lee, Chang-Soo Han
e=1
[k
0
]
e
(7)
Let the element displacement field be given by {u}=[N]{d} and {u}=
[N]{d}=[B]{d}. The element's stress stiffness matrix can be given by
[k
0
]
e
=
_
Ve
[B]
T
[S][B] dV (8)
where [B] is obtained from shape function, [N] by appropriate dif-
ferentiation and [S] contains the stress level due to the preloading.
[N] is the shape function matrix.
[S] =
S 0 0
0
S 0
0 0
S
, [
S] =
o
x0
t
xy0
t
zx0
t
xy0
o
y0
t
yz0
t
zx0
t
yz0
o
z0
(9)
The stored strain energy due to the preloading is given by
U
0
=
_
V
(
1
2
(u
2
,x
+v
2
,x
+w
2
,x
)o
x0
+ +(u
,x
u
,z
+ v
,x
v
,z
+w
,x
w
,z
)t
zx0
) dV
=
_
V
{c}
T
{o
0
} =
1
2
{u}
T
[K
0
]{u} (10)
where {c}
T
= {c
x
c
y
zx
}, {o
0
}
T
= {o
x0
o
y0
t
zx0
}.
To design a stiffening structure under a given loading, mean com-
pliance is chosen as the objective function, which is defined as the
least amount of displacement and the minimum mean compliance.
Thus, optimization involves not only minimizing the mean compli-
ance or elastic strain energy of the structure, but also minimizing the
effect of external forces. The compliance of a structure with stress
stiffness under a given loading can be written as
U =
1
2
{u}
T
([K] +[K
0
]){u}. (11)
Taking the derivatives of Eq. (11) with respect to the design param-
eter gives
dU
dj
= U
=
1
2
_
du
dj
_
T
([K] +[K
0
]){u}
+
1
2
{u}
T
__
dK
dj
_
+
_
dK
0
dj
__
{u} +
1
2
{u}
T
([K] +[K
0
])
_
du
dj
_
=
1
2
{u
}
T
([K] +[K
0
]){u} +
1
2
{u}
T
([K
] +[K
0
]){u}
+
1
2
{u}
T
([K] +[K
0
]){u
}
= {u
}
T
([K] +[K
0
]){u} +
1
2
{u}
T
([K
] +[K
0
]){u} (12)
Eq. (11) can be rewritten by
U
= {u
}
T
{F} +
1
2
{u}
T
([K
] +[K
0
]){u}. (13)
Assuming that the coordinates under a given loading are only
considered in the work done by the external forces and that the
coordinates in the free domain are not under loading, the following
may be defined:
{u
}
T
{F} =
_
du
T
f
dj
du
T
e
dj
_
_
0
F
e
_
=
_
du
e
dj
_
T
{F
e
}
= ({u
e
}
T
{F
e
})
= 2U
(14)
where u
f
is the displacement field that is not under a given loading
and u
e
is the displacement field under a given loading.
Using Eq. (14), the sensitivity of compliance from Eq. (11) is given
by the following expression
U
=
1
2
{u}
T
([K
] +[K
0
]){u} = U
e
+ U
0
(15)
where U
e
and U
0
are the strain energy sensitivities due to external
forces and preloading. [K
] and [K
0
] defined on an element level can
be given by the material densities of elements relative to the stress
ratio. The full stress scaling is used for the preloading. The deriva-
tive of the stress stiffness matrix depends on the initial stress {o
0
}.
If these stresses remain constant, [K
0
] is zero. But the topological
distribution or geometric dimensions of structure under preloading
may change on the design domain and in structural rigidities:
[K
0
] =
d
dj
e=1
_
Ve
[B]
T
[S][B] dV
e
=
e=1
_
Ve
[B]
T
[S
][B] dV
e
(16)
where [S
j[K]
jp
k
{u} (18)
where j{u}/jp
k
is the unknown displacement sensitivity coeffi-
cients, j{F}/jp
k
is the external loads not dependent on structural
D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332 327
Fig. 3. Design procedure of automotive structure through the model correlation.
properties, j[K]/jp
k
is the derivative of the system stiffness matrix
with respect to the parameter.
j[K]/jp
k
=
[K(p
k
+p
k
)] [K(p
k
)]
p
k
The geometric parameters can be defined by setting the basis vector
of grid point changes to the directions normal to the surfaces as
follows:
{G} =[T]{p} (19)
where {G} is the set of grid point changes, [T] is the set of shape
basis vectors and {p} is the set of scaled design parameter changes
in the shape dimension.
3. Sensitivity based correlations between test and simulation
3.1. Static response correlation
In the static analysis, the response error vector {R
n
} consists
of the n-th normalized displacement of configuration and DAC
(displacement assurance criterion), 1:
{R
n
} =
_
x
n
1
n
_
(20)
where
x
n
=
x
n
x
n
test
=
x
n
test
x
n
analysis
x
n
test
(21)
1
n
=11
n
=1
_
({[
n
test
}
T
{[
n
analysis
})
2
({[
n
test
}
T
{[
n
test
})({[
n
analysis
}
T
{[
n
analysis
})
_
(22)
[
n
test
and [
n
analysis
represent the n-th normalized static displacement
vector of the test and analysis. The sensitivity sub-matrices can be
given by
[Q
n
] =
_
jx
n
jP
_
_
j1
n
jP
_
(23)
3.2. Dynamic response correlation
In the dynamic response analysis, the response error vector {R
n
}
can be given by
{R} =
{R
1
}
{R
2
}
.
.
.
{R
n
}
y
1
T
1
y
2
T
2
.
.
.
y
n
T
n
(24)
where the error vector {R
n
} consists of the n-th independent vari-
able and DDAC (dynamic displacement assurance criterion), Tunder
328 D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332
unit dynamic loading:
y
n
=
y
n
y
n
test
=
y
n
test
y
n
analysis
y
n
test
(25)
T
n
=1 T
n
=1
_
({
n
test
}
T
{
n
analysis
})
2
({
n
test
}
T
{
n
test
})({
n
analysis
}
T
{
n
analysis
})
_
(26)
y
n
test
and y
n
analysis
represent the m-th dynamic response time or fre-
quency of behaviors under dynamic loadings.
test
and
analysis
rep-
resent the normalized dynamic displacements of test and analysis.
The sensitivity matrix consists of two parts: the sensitivity matrices
of normalized response times of behaviors and DDAC with respect
to design variables, {P}:
[Q] =[[Q
1
] [Q
2
] [Q
n
]]
T
=
_
jy
1
jP
_
_
jT
1
jP
_
_
jy
2
jP
_
_
jT
2
jP
_
.
.
.
_
jy
n
jP
_
_
jT
n
jP
_
(27)
Updating the stiffness modeling using static displacement tests
involves minimizing the error function:
Err =[e]
T
[w
e
][e] +[p]
T
[w
p
][p] (28)
where e is the difference between experimental and analytical
static displacements (e=j{u
j
}/j{p
k
}p={u
j
exp
}{u
j
ana
}) is the differ-
ence between updated and originally estimated parameters (p
k
=
p
ku
p
ka
) and w
c
and w
p
are the diagonal weighting matrices for the
selected updating the static displacements and the updating param-
eters, respectively.
4. Design optimization process
The overall engineering design roadmap and virtual prototypes
for the durability of the automotive structure [1117] are shown
in Figs. 3 and 4. The EMBS (elastic multi body simulation) is fre-
quently applied in order to determine the time depended load of
a flexible structure and predict the feasibly transferred load to the
components. The computation procedure which is presented, intro-
duces a modally based dynamic simulation for the maximum dura-
bility load. The numerical examples show that the applications range
from quasi-static behavior to vibration dominated problems, includ-
ing nonlinear effects. This paper presents the procedure of the model
correlation shown in Fig. 3.
The design process of mechanical structures can be based on the
optimization process [18,19] to find the feasible configurations that
fulfill certain quality requirements. At this point, engineers often find
that they require multiple interfaces to build the computer models
that will simulate a product's performance. In this paper, the design
roadmap based on process integration for nonlinear structural per-
formances is iteratively solved by the NewtonRaphson method and
sequential quadratic programming. The fully stressed design is used
for the parameter corrections in the iterations that are widely prac-
ticed for variably dimensioning the design parameters at its limit
Fig. 4. Virtual prototype for the durability load.
Fig. 5. Design flow for the model correlation of front rail structure.
under a given load. When performed iteratively, it usually converges
rapidly and yields a reasonable structural design. Commercial sim-
ulation program and process integration software are used for the
optimization design of model correlation in the nonlinear structural
specifications of automotive structure. A sensitivity-based correla-
tion algorithmrequires the computation of design variable variations
for each configuration based on the corresponding analysis data. The
correlation equation can be given by
[Q]
T
[Q]{P} =[Q]
T
{R} (29)
[Q
1
]
[Q
2
]
.
.
.
[Q
n
]
[Q
1
]
[Q
2
]
.
.
.
[Q
n
]
{P} =
[Q
1
]
[Q
2
]
.
.
.
[Q
n
]
{R
1
}
{R
2
}
.
.
.
{R
n
}
(30)
D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332 329
where [Q] and {R} are the overall structural sensitivity matrix and
response error vectors between test and analysis. {P} is the vector of
design variable variations. {R
n
} represents the corresponding error
vector of the n-th configuration response.
The changes to the model update can be represented by Eq. (31).
The design variables are simultaneously used in the static and dy-
namic response correlations.
{R(P)}
i+1
= {R(P)}
i
+[Q]
i
{P} (31)
where Q
i,j
=jR
i
/jP
j
.
Y
3
Y
2
Y
1
0
1
2
2
3
3
-p -p
k
= Effective Plastic Strain
-p
k = 3
k = 2
k = 1
H
3
H
2
H
1
Y (or s)
-
E
H
k
=
Y
k+1
- Y
k
k + 1
-
k
-p
Fig. 6. Stressstrain curve definition for the durability analysis.
Fig. 7. Road running simulation and crack zone from the durability test. (a) Road simulation, (b) Sampling Load history in the forward direction at RH tire, (c) Sampling
Load history in the lateral direction at RH tire and (d) Sampling Load history in the vertical direction at RH tire.
The design process for mechanical systems can be viewed as an
optimization process to find parts that fulfill certain quality require-
ments toward their functionality, appearance and economy. It can
be described as an iterative search process that uses the following
steps:
(1) Define an initial design p
(i=0)
.
(2) Analyze the nonlinear characteristics using a nonlinear solver
routine.
(3) Compare the results of the analysis with such requirements as
allowable plastic strain or residual deformation.
(4) If the requirements are not met, perform the optimization
routine in order to set op.
(5) Correct op based on the fully stressed design with [ = 0.9,
(op)
new
=(op)
old
_
o
actual
o
allowable
_
[
(32)
(6) Change the design variables using p
(i+1)
= p
(i)
+op.
(7) If the requirements are satisfied, complete the discrete design
with consideration of manufacturability. Otherwise, go to (2)
The formulation of an optimization problem in the model correlation
appears as
Objective : R(P) min
Constraints : C
i
(P) 0
Design space : P
L
PP
U
(33)
Objective R
i
(P) can be approximated for each design P
(i)
using the
series expansion,
R
i
=R
i
(P
(i)
) +
n
j=1
j(R)
i
jP
j
P
j
(34)
330 D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332
Fig. 8. Front rail structure model. (a) Finite element model for the model correlation
and (b) Weld points.
The gradient j(R)
i
/jP
j
can be obtained directly from the results
of finite element analysis. If the gradient is known, the search direc-
tion P can be obtained from the solution of an approximate opti-
mization problem. However, the optimization problem in the model
correlations should be formulated in the minimization of strains and
stress, subject to the welding locations, panel thickness and panel
configurations due to their variations in the manufacturing environ-
ments.
5. Simulations
In order to study the behavior of the vehicle system for the well-
correlated model, a dynamic nonlinear finite element simulation is
performed. This simulation allows for the evaluation of destructive
test events using a model made up of the body structure, interior
components such as the seat frames, instrument panel and steer-
ing system, and a suspension system and tire. Such destructive test
events are not only costly with respect to time, but require expen-
sive and scarce prototype vehicle, simulations of these events are
seen as the only way to accurately predict the vehicle behavior prior
to prototype vehicle availability. The destructive test events which
are chosen for simulation are a curb impact and a severe pothole
event. To perform the simulation, VPG is implemented. This tech-
nique makes use of vehicle FEA models in conjunction with the curb
road surface model.
The automotive body structure has the influences of the junction
boundaries and gradual variation shapes among the body panels.
Their manufacturabilities have influences on the structural behavior
characteristics of active stress points. In other words, the origin of
localized stress concentrations and the high strains arise from the
coupling from the welded characteristics, design variations of body
Fig. 9. Stress contour of front rail structure. (a) Stress contour without the model
correlation, (b) Stress contour with the model correlation and (c) Crack location of
front rail structure.
panels and physical properties. As a sheet metal component is
formed through a stamping process, the material undergoes changes
from its initial conditions for thickness and residual stresses. Since
materials in finite element models are typically considered uniform,
there is a potential for the material to be drastically different from
the specified, nominal material thickness. The updated material
thickness, with local material thickness changes may then be used
in the durability simulation, yielding a more accurate prediction
of potential durability problems. Their differences from the manu-
factured products may allow many gradual variations of structural
rigidity and durability that can change the location and gradients of
localized stresses. Thus, a method for predicting the body deforma-
tion during the running operations creates a body model based on
the characteristics extracted by modal analysis of the results of a
D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332 331
Fig. 10. Relationship between geometric dimensions and welding location variations.
Fig. 11. Constraint trend in the simulation times.
vibration testing of an actual vehicle. In this example, the durabil-
ity loads were decided from the road running simulations based
on the actual experiments and through the maximum durability
conditions, the crack locations are found in the laboratory test. The
simulation model is limited to the front rail structure, which has
145,013 elements and 129,192 nodes. Fig. 5 shows the overall design
flow for the model correlation using the optimization design on the
geometric dimensions concerning the localized reinforced-beads
and panel thickness.
In particular, the welding locations and material thickness may
be used for minimizing the deviations of the required structural
specifications between test and simulation.
The basic concepts of the model correlation using the optimiza-
tion design as follows:
Minimize |W
TEST
W
SIM
|
Subject to
n
i=1
|c
TEST
i
c
SIM
i
|
|G
weld
| G
allowable
,
|T
shape
| T
allowable
,
|t
panel
| t
allowable
(35)
where W
Test
and W
SIM
are the respective weights for the designing
panes, c
TEST
i
and c
SIM
i
are the strain measured from the experiment
and simulations in the Von-Mises criteria, G
weld
is the location vari-
Fig. 12. Stress contour in the BIW (body-in-white) structure.
ations at the point of actual welding points, T
shape
is the geometric
variations of the reinforced bead on the panels and t
panel
is the panel
thickness. The linear constraints except the elasto-plastic material
properties [19,20] are explained as follows. The variations of weld-
ing locations have the 5mm radius variations of spot center on
reference of CAD data. The variations of geometric dimensions have
the width, height and longitudinal length with 2.0mm on refer-
ence of the measured stamped bead shapes. And, the panel thick-
ness has the 10% variations on reference of the initial CAD data.
The nonlinear constraints are the variations of material properties
(tangent modulus) of the stressstrain curve shown in Fig. 6, which
is concerned with the strain-hardening. The plasticity modulus (H)
shown in Fig. 6 is related to the tangent modulus (E
T
) by
H = E
T
/1
E
T
E
(36)
where E is the elastic modulus and E
T
= dY/dc is the slope of the
uniaxial stressstrain curve in the plastic region.
332 D.-C. Lee, C.-S. Han / Finite Elements in Analysis and Design 45 (2009) 324-- 332
Fig. 7 shows the road running simulation which represents the
proving ground and load time history. These loads were semi-
analytical loads and they were used as design and verification loads
to drive the design details and engineering design. Fig. 8 shows the
finite element model of front rail and weld points for the model
correlation, in the fully trimmed CAE model. In this model, all body
sheet an welds ere meshed in detail and all trim items attached to
body FEA model.
Fig. 9(a) shows the stress contour of the wheelhouse of front
rail structure without the model correlation and Fig. 9(b) shows
stress contour of the wheelhouse of front rail structure with the
model correlation under maximum pothole load. Fig. 10 shows the
estimation point between geometric dimensions and welding loca-
tion variations of the specific parameters in the 3-dimension space,
which shows the nonlinearities of durability performance from the
inter-acted design parameters. In this distribution contour, it is im-
portant that the body designer decides the feasible points and inter-
relates them with other structural specifications. Thus, in the model
correlations, the design parameter screening is also important.
Fig. 11 shows the constraint trends in the simulation times. The error
shows the summation of linear and nonlinear constraints. The most
impact factor is the strain constraints with 5060% of total error.
Fig. 12 shows the stress contour shown in the BIW structure under
the running condition of maximum pothole load. For the correlation
and nonlinear static analysis, the commercial softwares are used
[21,22].
6. Conclusion
The integration of simulation with physical test will accelerate
the product-development process. The bi-directional flow of infor-
mation between these two functions is critical and important for
the new design successes. Therefore, engineers should compare test
data from previous models or components against simulation results
and calibrate them to increase confidence in their simulation predic-
tions for current designs. In this point, this paper presents a design
methodology to account for nonlinear structural rigidities by con-
sidering such issues as plastic strain and residual deformation. Such
geometric dimensions as panel structure, thickness and shape can
be corrected through the optimization based correlation design pro-
cess and can consider the environment uncertainties. The advantage
of the presented procedure increases such accuracies of durability
designs as the damage locations and fatigue life. This procedure is
considered as an important tool to realize the integrated computa-
tion of dynamics and durability in the virtual prototype parts such
as the control arm and knuckle in the chassis parts. The integration
process of CAE in the design, engineering and development process
is a critical success factor for successfully delivering a vehicle with
the product integrity.
References
[1] J.C. Aguinaldo, T.K. Jorge, C.C. Claudiomar, Nonlinear considerations in shock
tower durability analysis, SAE 2002-01-3483.
[2] G.F. Wallace et al., Structural optimization of automotive components applied
to durability problems, SAE 2003-01-3547.
[3] Y. Charles, N.A. Jon, T. Maolin, An integrated system for durability and reliability
synthesis using iSIGHT and FE-Fatigue, SAE 2003-01-1220.
[4] T. Mahesh, Y. Xiaobo, Vehicle cradle durability design development, SAE 2005-
01-1003.
[5] G.A. Wempner, Discrete approximations related to nonlinear theories of solids,
Int. J. Solids Struct. 7 (1971) 15811599.
[6] E. Riks, An incremental approach to the solution of snapping and buckling
problems, Int. J. Solids Struct. 15 (1979) 529551.
[7] H.S. Irving, L.D. Clive, Energy and Finite Element Methods in Structural
Mechanics, McGraw-Hill, New York, 1986.
[8] D.C. Robert, S.M. David, E.P. Michael, Concepts and Applications of Finite
Element Analysis, third ed., Wiley, New York, 1989.
[9] M.A. Crisfield, Nonlinear finite element analysis of solids and structures, basic
formulations, vol. 1, Wiley, New York, 1991.
[10] D.C. Lee, J.H. Jang, C.S. Han, Design consideration of a mechanical structure
with geometric and material non-linearities, Proc. Inst. Mech. Eng. Part D
J. Automob. Eng. 220 (3) (2006) 281288.
[11] S.S. You, S.G. Joo, Virtual testing and correlation with spindle coupled full
vehicle testing system, SAE 2006-01-0993.
[12] S.H. Lin, C.C. Cheng, C.Y. Liao, J.M. Chang, Experiments and CAE analyses for
suspension under durability road load conditions, SAE 2006-01-1624.
[13] W. Guoguan, K. Xiaodi, A virtual test approach for vehicle ride comfort
evaluation, SAE 2004-01-0376.
[14] M.P. Suyog, S.G. Santosh, Integrated structural durability test cycle development
for a car and its components, SAE 2004-01-1654.
[15] S.G. Joo, S.S. You, A.F. Joseph, C. Leser, Integration of physical and virtual tools
for virtual prototype validation and model improvement, SAE 2003-01-2813.
[16] S.B. Lee, W. S. Han, H.J. Yim, Fatigue analysis of automotive suspension system
considering dynamic effect, SAE 2003-01-2814.
[17] Z. Koos, Integration of physical and virtual prototypes, SAE 2002-01-1290.
[18] E.I. Haug, J.S. Arora, Applied Optimal DesignMechanical and Structural
Systems, Wiley, New York, 1979.
[19] D.C. Lee, C.S. Han, A frequency response function-based updating technique for
the finite element model of automobile structures, Proc. Inst. Mech. Eng. Part
D J. Automob. Eng. 220 (10) (2008) 281288.
[20] D.S. Lee, Y.H. Woo, S.H. Lee, C.S. Han, Design consideration of the nonlinear
specifications in the automotive body, Finite Elem. Anal. Des. 44 (14) (2008)
851861.
[21] SIEMENS Software, NX.NASTRAN V5.0.
[22] ENGINESOUS Software, iSIGHT V6.0.