Professional Documents
Culture Documents
C. G. Giannopapa
Department of Mathematics
and Computer Science,
Eindhoven University of Technology,
PO Box 513,
Eindhoven 5600, The Netherlands
e-mail: c.g.giannopapa@tue.nl
R. M. M. Mattheij
Department of Mathematics
and Computer Science,
Eindhoven University of Technology,
PO Box 513,
Eindhoven 5600, The Netherlands
e-mail: r.m.m.mattheij@tue.nl
Introduction
C 2014 by ASME
Copyright V
Problem Formulation
(1c)
Simulation Model
A Galerkin finite element method is employed for the discretization of the forward problem. The 2D axial-symmetrical finite
element model has been implemented in COMSOL 3.5 with MATLAB.
For the spatial discretization of the flow problem Mini-elements
are used [31]. For the spatial discretization of the level set and
energy problem linear Lagrange elements are used. For the temporal discretization an implicit differential-algebraic solver scheme
has been used [32], which employs a variable-order variable-stepsize backward difference formula method. The discretized system
Transactions of the ASME
of equations is solved using the BiCGstab method with a geometric multigrid method as a preconditioner [33]. A simulation stops
if either the mold wall is covered with glass or an upper limit of
the process time is reached.
A level set method is used to capture the glassair interfaces.
This approach allows for the computation of two-fluid flow solutions with moving interfaces while using a fixed finite element
mesh. Moreover, the level set function naturally deals with topological changes. A triangulated fast marching method is used as a
re-initialization algorithm to maintain the level set function as a
signed distance function [26,34]. Further details about the level
set method and fast marching algorithm used can be found in
Refs. [1,25].
Air is replaced by a fictitious fluid with a much larger viscosity
in the simulations, so that the flow of air can also be described by
the Stokes flow equations. The viscosity of the fictitious fluid
should still be much smaller than the viscosity of glass, so that the
pressure drop in the air domain is negligible compared to the pressure drop in the glass domain [1,35]. For further details the reader
is referred to Refs. [1,2,25,30].
The accuracy of the glass blow simulation model has been
successfully validated by the authors by means of verification of
conservation of volume [1,25] and comparison of the container
thickness with experimental data provided by industry [2].
(2)
Ci
h2i;;p dC
1
0
h2i;;p xi sx0i sds
(3)
nG
nint X
1X
xk x0 sjk h2i;;p xsjk
2 j k
(4)
U
_
1 T
r r
2
(5)
Constraints
l 1; nc
(7)
where nc is the number of constraints. The constraints are incorporated in the optimization method by introducing a weighted objective function, given by
~
Up
Up fp
nc
X
wl
;
c
l p
l
(9)
i
(10)
Ji TJi ki I Hf si Ji Tri rp fi
where J is the nres m Jacobian matrix of r, Hf is the m m
Hessian matrix of f, and rpf is the gradient of f, all with respect
to p. The solution s(i) is a parameter increment, which is used to
adjust the control points
pi1 pi si
(13)
where e is a unit basis vector in Rm . The forward difference formula requires the computation of m additional function values in
p de for each basis vector e.
The error in the finite difference approximation directly influences the error of the optimization algorithm. Expression (13) suggests that the step size d is chosen as small as possible. However,
also the error in computing the residual r should be taken into
account. Let this error be denoted by eF. Then the sum of errors is
bounded by
2eF 1
Md
2
d
r
p
4
eF eF
d
M
(15)
(11)
(14)
Jpe
(i)
rp de rp
Od
d
(8)
rp de rp de
O d2
2d
(16)
(17)
where
~
Up
Up
:
;
jCi j
jCi j :
nG
nint X
X
j
d
6
eF
L
1=3
1=3
eF
(18)
Since the error in the central difference formula is O(d2), the error
2=3
in the solution of the inverse problem is at least OeF , which is
not a considerable improvement considering the fact that the
method is twice as expensive. Still the central difference formula
can be used when the optimal accuracy is reached to gain a few
extra digits in the last few iterations.
5.2.2 Broyden Update. In Refs. [23,24] a secant method using
a Broyden update of the Jacobian matrix is applied [3941]. The
initial Jacobian matrix is approximated by a forward difference
method. A forward difference method is also used if no convergence is reached after several iterations with the secant method.
The Broyden update of the Jacobian matrix J(p(i1)) for iteration i 1 of the optimization method is
Ai1 Ai
i i
A s s T
si Tsi
(19)
where yi ri1 ri . Assume that the Jacobian matrix is Lipschitz continuous with Lipschitz constant c
i1
Jp
Jpi csi
(20)
In the following a quantitative analysis of the errors is performed. The analysis is based on the following assumptions:
In Ref. [41] the following upperbound for the error of the Broyden
update is given:
i1
3
A
Jpi1 Ai Jpi csi
2
(21)
(13). Then the initial error is O(d), where d eF . Since the forward problem is not sensitive to changes in the parameters, c is
small. Assuming that r is twice differentiable, it holds that
Hj p M
(22)
c
max
m
m & CI h2 1
p2R ; j1;;n
where
i Hj is the Hessian matrix of rj. The order of magnitude of
s is not smaller than O(d), so the error of the Broyden update
is at least O(d) and can have larger order of magnitude far away
from the optimum.
(23)
(24)
(27)
13
R2;0 u R3m u R3i u R31;0 u ; 0 u u0 (28)
Here u 0 corresponds to the symmetry axis and u u0 corresponds to the mold entrance.
The next step in the construction of the initial guess is the
choice of the angles of the control points. A suitable choice is an
equidistant distribution of the control points along the curve. Consider only the case in which the control points are on the parametric curve, e.g., a natural spline. The distance between the control
points is expressed in terms of the length of the curve
u0 q
R2 u2 R02 u2 du
(29)
The angles of the control points are determined by solving the following equations by an adaptive Simpson rule:
uj q
R2 u2 R02 u2 du
(30)
m
uj1
where uj is the angle of Pj, j 1,,m.
Fig. 6
Fig. 8
Results
j 1; ; m
(31)
j 1; ; m 1
(32)
j 2; ; m
(33)
mold shape computed with sagging for 0.5 s and Fig. 9(b) shows
the mold shape computed without sagging. Obviously, the bottle
computed without sagging is much closer to the designed bottle in
Fig. 6 than the bottle computed with sagging. This could be
expected as during sagging the mass flows in negative axial direction. Consequently, the mass flow in polar direction during sagging in clockwise polar direction is significant. On the other hand,
in the initial guess of the preform shape the mass flow in polar
direction is completely omitted. In contrast the approximation
without sagging is quite satisfactory and suitable to be used as an
initial guess.
The optimization method is used to find the optimal positions
of the control points of the outer preform surface as to minimize
the difference between the inner surfaces of the computed container without sagging and the designed container in the sense of
Eq. (3). Consider a finite element mesh with 40 elements along
the mold opening by 272 elements along the symmetry axis and a
mesh distribution as in Fig. 5. The LevenbergMarquardt method
with combined forward difference method and secant method is
used to find the optimal preform. The typical mesh size for this
example is h 0.059. Figure 10 shows the convergence results.
From the second iteration the Broyden method is used to update
the Jacobian matrix. However, in the sixth iteration no convergence is reached, so the Jacobian matrix is approximated by finite
differences. In the seventh iteration another Broyden update is
performed, but from the eighth iteration onward only forward difference approximations are used, since the parameter vector is
close to the optimum. Only 59 function evaluations are required
until convergence. The central processing unit (CPU) time for one
simulation is approximately 5.4 103 (1.5 h) and the total CPU
time for the optimization procedure is approximately 3.2 105.
The order of magnitude of the error of the optimization method
DECEMBER 2014, Vol. 136 / 061301-7
Fig. 11 Optimal preform and mold shape for beer bottle. (a)
Optimal preform and (b) optimal mold shape.
Fig. 9 Mold shapes for initial guess. (a) Computed with sagging and (b) computed without sagging.
(27) can be verified from the convergence results. With the current
error tolerance, it can be observed that convergence is slowing
down in the last few iterations. If the tolerance is set stricter then
O(h), the optimization method will converge extremely slow or
not at all. This phenomenon has been observed for different mesh
sizes and confirms the theoretical error estimates.
Figure 11 shows the result of the optimization method.
Figure 11(a) shows the optimal shape of the preform and
Fig. 11(b) shows the resulting container. The resulting container
in is in good agreement with the designed container in Fig. 6.
The mean distance between the inner container surfaces is
7.83 102 cm and the maximum distance is 1.78 101 cm. The
061301-8 / Vol. 136, DECEMBER 2014
Conclusions
References
[1] Giannopapa, C. G., 2008, Development of a Computer Simulation Model for
Blowing Glass Containers, ASME J. Manuf. Sci. Eng., 130(4), p. 041003.
[2] Giannopapa, C. G., and Groot, J. A. W. M., 2011, Modeling the BlowBlow
Forming Process in Glass Container Manufacturing: A Comparison Between
Computations and Experiments, ASME J. Fluids Eng., 133(2), p. 021103.
[3] Cesar de Sa, J. M. A., 1986, Numerical Modelling of Incompressible Problems
in Glass Forming and Rubber Technology, Ph.D. thesis, University College of
Swansea, Swansea, UK.
[4] Cesar de Sa, J. M. A., 1986, Numerical Modelling of Glass Forming Processes, Eng. Comput., 3(4), pp. 266275.
[5] Cormeau, A., Cormeau, I., and Roose, J., 1984, Numerical Simulation of
Glass-Blowing, Numerical Analysis of Forming Processes, J. F. T. Pittman,
O. C. Zienkiewicz, R. D. Wood, and J. M. Alexander, eds. Wiley, New York,
pp. 219237.
[6] Williams, J. H., Owen, D. R. J., and Cesar de Sa, J. M. A., 1986, The Numerical Modelling of Glass Forming Processes, Collected Papers of the XIV International Congress on Glass, H. C. Bharwaj, ed., Indian Ceramic Society, New
Delhi, India, pp. 138145.
[7] DeLorenzi, H. G., and Nied, H. F., 1987, Blow Molding and Thermoforming
of Plastics: Finite Element Modelling, Comput. Struct., 26(12), pp.
197206.
[8] Warby, M. K., and Whiteman, J. R., 1988, Finite Element Model of Viscoelastic Membrane Deformation, Comput. Methods Appl. Mech. Eng., 68(1),
pp. 3354.
[9] Chung, K., 1989, Finite Element Simulation of PET Stretch/Blow-Molding
Process, J. Mater. Shaping Tech., 7(4), pp. 229239.
[10] Cesar de Sa, J., Natal, R., Silva, C., and Cardoso, R. P., 1999, A Computational Model for Glass Container Forming Processes, Europe Conference on
Computational Mechanics Solids, Structures and Coupled Problems in Engineering, Munich, Germany, August 31September 3.
[11] Hyre, M., 2002, Numerical Simulation of Glass Forming And Conditioning,
J. Am. Ceram. Soc., 85(5), pp. 10471056.
[12] Diraddo, R. W., and Garcia-Rejon, A., 1993, Profile Optimization for the Prediction of Initial Parison Dimensions From Final Blow Moulded Part Specifications, Comput. Chem. Eng., 17(8), pp. 751764.
[13] Diraddo, R. W., and Garcia-Rejon, A., 1993, On-Line Prediction of Final Part
Dimensions in Blow Molding: A Neural Network Computing Approach,
Polym. Eng. Sci., 33(11), pp. 653664.
[14] Lee, D. K., and Soh, S. K., 1996, Prediction of Optimal Preform
Thickness Distribution in Blow Molding, Polym. Eng. Sci., 36(11),
pp. 15131520.
[15] Thibault, F., Malo, A., Lanctot, B., and Diraddo, R., 2007, Preform Shape and
Operating Condition Optimization for the Stretch Blow Molding Process,
Polym. Eng. Sci., 47(3), pp. 289301.
[16] Gauvin, C., Thibault, F., and Laroche, D., 2003, Optimization of Blow Molded
Part Performance Through Process Simulation, Polym. Eng. Sci., 43(7),
pp. 14071414.
[17] Hsu, Y. L., Liu, T. C., Thibault, F., and Lanctot, B., 2004, Design Optimization of the Blow Moulding Process Using Fuzzy Optimization Algorithm,
Proc. Inst. Mech. Eng., Part B, 218(2), pp. 197212.
[18] Yu, J.-C., Chen, X.-X., Hung, T.-R., and Thibault, F., 2004, Optimization
of Extrusion Blow Molding Processes Using Soft Computing and Taguchis
Method, J. Intell. Manuf., 15(5), pp. 625634.
[19] Lochegnies, D., Moreau, P., and Guilbaut, R., 2005, A Reverse Engineering
Approach to the Design of the Blank Mould for the Glass Blow and Blow
Process, Glass Technol., 46(2), pp. 116120. Available at: http://www.ingentaconnect.com/content/sgt/gt/2005/00000046/00000002/art00017
[20] Moreau, P., Marechal, C., and Lochegnies, D., 2001, Optimum Parison Shape
for Glass Blowing, XIXth International Congress on Glass, Society of Glass
Technology, Edinburgh, UK, July 16, pp. 548549.
[21] Choi, J., Ha, D., Kim, J., and Grandhi, R. V., 2004, Inverse Design of Glass
Forming Process Simulation Using an Optimization Technique and Distributed
Computing, J. Mater. Process. Technol., 148(3), pp. 342352.
[22] Moreau, P., Lochegnies, D., and Oudin, J., 1998, An Inverse Method for Prediction of the Prescribed Temperature Distribution in the Creep Forming Process, Proc. Inst. Mech. Eng., Part C, 212(1), pp. 711.
[23] Groot, J. A. W. M., Giannopapa, C. G., and Mattheij, R. M. M., 2009,
Numerical Optimisation of Blowing Glass Parison Shapes, ASME Paper No.
PVP2009-77946.
[24] Groot, J. A. W. M., Giannopapa, C. G., and Mattheij, R. M. M., 2011,
Development of a Numerical Optimisation Method for Blowing Glass Parison
Shapes, ASME J. Manuf. Sci. Eng., 133(1), p. 011010.
[25] Giannopapa, C. G., and Groot, J. A. W. M., 2007, A Computer Simulation
Model for the BlowBlow Forming Process of Glass Containers, ASME Paper
No. PVP2007-26408.
[26] Sethian, J. A., 1999, Level Set Methods and Fast Marching Methods, Cambridge University, New York.
[27] Sussman, M., Smereka, P., and Osher, S., 1994, A Level Set Approach for
Computing Solutions to Incompressible Two-Phase Flow, J. Comp. Phys,
114(1), pp. 146159.
[28] Adalsteinsson, D., and Sethian, J. A., 1995, A Fast Level Set Method for Propagating Interfaces, J. Comp. Phys, 118(2), pp. 269277.
[29] Chang, Y. C., Hou, T. Y., Merriman, B., and Osher, S., 1996, A Level Set Formulation of Eulerian Interface Capturing Methods for Incompressible Fluid
Flows, J. Comp. Phys., 124(2), pp. 449464.
[30] Groot, J. A. W. M., Mattheij, R. M. M., and Laevsky, K., 2011, Mathematical
Modelling of Glass Forming Processes, Mathematical Models in the Manufacturing of Glass (Lecture Notes in Mathematics), A. Fasano, ed., Vol. 2010,
Springer, Berlin, pp. 156.
[31] Bathe, K., 1996, Finite Element Procedures, Prentice Hall, Englewood Cliffs,
NJ.
[32] Hindmarsh, A. C., Brown, P. N., Grant, K. E., Lee, S. L., Serban, R., Shumaker,
D. E., and Woodward, C. S., 2005, SUNDIALS: Suite of Nonlinear and Differential/Algebraic Equation Solvers, ACM Trans. Math. Software, 31(3),
pp. 363396.
[33] van der Vorst, H. A., 1992, Bi-CGSTAB: A Fast and Smoothly Converging
Variant of Bi-CG for the Solution of Nonsymmetric Linear Systems, SIAM
J. Sci. Stat. Comput., 13(2), pp. 631644.
[34] Kimmel, R., and Sethian, J. A., 1998, Computing Geodesic Paths on Manifolds, Proc. Natl. Acad. Sci. USA, 95(15), pp. 84318435.
[35] Haagh, G. A. A. V., 1998, Simulation of Gas-Assisted Injection Moulding,
PhD thesis, Eindhoven University of Technology, Eindhoven, Netherlands.
[36] Carroll, C. W., 1961, The Created Response Surface Technique for Optimizing
Nonlinear Restrained Systems, Oper. Res., 9(2), pp. 169184.
[37] Schnur, D. S., and Zabaras, N., 1992, An Inverse Method for Determining
Elastic Material Properties and a Material Interface, Int. J. Numer. Methods
Eng., 33(10), pp. 20392057.
[38] Gelin, J. C., and Ghouati, O., 1994, An Inverse Method for Determining
Viscoplastic Properties of Aluminium Alloys, J. Mater. Process. Technol.,
45(14), pp. 435440.
[39] Broyden, C. G., 1965, A Class of Methods for Solving Nonlinear Simultaneous Equations, Math. Comput., 19(92), pp. 577593.
[40] More, J. J., and Trangenstein, J. A., 1976, On the Global Convergence of
Broydens Method, Math. Comput., 30(135), pp. 523540.
[41] Dennis, J. E., and Schnabel, R. B., 1996, Numerical Methods for Unconstrained
Optimization and Nonlinear Equations, SIAM, Philadelphia, PA.
[42] Nocedal, J., and Wright, S. J., 1999, Numerical Optimization, Springer,
New York.