Optimization Methods Applied to

R. J. Menassa*
Graduate Research Assistant.
Selecting Support Positions in
Assoc. Mem. ASME
Fixture Design
W. R. DeVries
This paper proposes optimization techniques to assist in the design and evaluation
Assoc. Professor.
Mem. ASME
of fixtures for holding prismatic workpieces. This formulation of the fixturing design
problem takes into account deflection of the workpiece subjected to assembly or
machining loads. Using the minimization of the workpiece deflection at selected
Department ot Mechanical Engineering,
points as the design criterion, the design problem is determining the positions of
Aeronautical Engineering and Mechanics,
Rensselaer Polytechnic Institute,
the fixture supports. The Finite Element Method is used for calculating deflections
Troy, NY 12180-3590 that are the basis for the design objective function, and the Broyden-Fletcher-
Goldfarb-Shanno optimization algorithm is used to determine the fixture support
positions. In this paper the proposed objective function is developed and the method
is illustrated with three numerical examples.

Introduction
With the current advances in technology, e.g., the emphasis cation of the fixture components to, for example, minimize
on flexible or computer integrated manufacturing systems, workpiece deflections.
fixtures have become more flexible in their applications at the
expense of being less rigid mechanically. In order to overcome Finite Element Analysis. To assist in the design of fixtures,
this difficulty, it may be possible to reach a compromise be- a method was developed by Lee and Haynes (1986) to minimize
tween flexibility and rigidity by developing better ways of se- various criteria that insure sound machining and assembly
lecting the position of supporting devices, so that the resulting tasks. In particular, one may want to minimize the fixturing
deflections are maintained within acceptable ranges. This is a forces, the maximum stress, and part deformations. Usually,
major part of the fixture design problem. large clamping forces and high stresses imply that excessive
machining work is being exerted on the part that could lead
In the past, little analysis work has been devoted to the to workpiece failure. The Finite Element Method (FEM) is
design of fixtures. More recently, the development of new used and treats the workpiece as an isotropic deformable body
techniques, software and mathematical theories have emerged based on linear elasticity and takes into account friction that
which should permit sound fixture designs. Some of the recent develops at the fixturing element/workpiece interfaces. The
research is described below. software uses a finite element program that is capable of graph-
Kinematic Analysis. In the approach of Asada and By ing the deformed and undeformed workpiece when subjected
(1985), the fixture design and the set-up procedure are per- to manufacturing forces. This study, while recognizing the
formed automatically by a computer-integrated system and a importance of part deformations and stresses and while de-
robot manipulator, whereby the layout of the fixture is de- termining the necessary number of fixturing elements, does
signed on a CAD system. The analysis derives the location of not provide a method that would optimally re-configure the
fixturing elements to locate the workpiece in a unique position. fixturing components in order to minimize the above stated
Accessibility to and detachability of the workpiece are incor- design criteria.
porated in the derivation of the final location of the parts and
the fixturing elements. This study, however, uses simply con- Fixture Layout and Planning. Mani and Wilson (1988) rely
nected surfaces to model the workpiece and fixture compo- on basic machining practice and technology and the geometry
nents, thereby treating them as rigid bodies. Kinematic of the part to provide a fixturing plan. The design criterion is
feasibility is a necessary condition to provide an initial con- tofully constrain the part, which sacrifices accessibility of the
figuration of the fixture components, i.e., insure that tool fixture. The workpiece geometry is first extracted from a solid
points are in contact with the workpiece and restrict sliding geometric modeler and then displayed in a two-dimensional
and rotation of a part when located and held in a fixture. This view so that the fixturing plan is conducted in the cross-sec-
analysis, however, does not provide a way to choose the lo- tional plane. The selection of passive and active fixturing ele-
ments, part loading, and clamping sequence are the machining
practice and technology aspects taken into account in the plan-
* Currently Senior Project Engineer, Advanced Engineering Staff, General ning process. Edges are rated according to three major con-
Motors Technical Center.
Contributed by the Production Engineering Division for publication in the siderations: accurate and repeatable part loading based on the
JOURNAL OF ENGINEERING FOR INDUSTRY. Manuscript received November 1990. 3-2-1 locating principle, rating the workpiece edges as to the

412 / V o l . 113, N O V E M B E R 1991 Transactions of the A S M E

Copyright © 1991 by ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms
 quality of the surfaces which determines the type of contacts, Nodal
Point
and a rating of potential reference edges based on their length.
The position of locating points on the reference surfaces are
determined in order to arrest the motion of the part. More
recent work by Chou et al. (1989) extends some of these ideas,
and the problem of selecting and positioning locating points
for prismatic parts if the load directions are given is considered
by Menassa and DeVries (1989).
This paper assumes that the structure of the fixture is based
on the 3-2-1 locating principle and the fixture will be used to
hold a prismatic workpiece, which, using the methods of Asada
and By (1985), can be shown to yield an accessible fixture
design. It is assumed that the methods in Menassa and DeVries
(1989) have been used to select the positions for the locating
points on the secondary and tertiary reference planes. This
paper uses the FEM, similar to the approach of Lee and Haynes
(1986), to calculate workpiece deflections that are a function
of the location of fixture supports and the loads caused by the Fig. 1 Three-dimensional mesh
manufacturing process. The design problem is to determine
support positions that are kinematically feasible and minimize
workpiece deflections. The specific objectives of this paper are
to:
• Treat fixture design as an optimal design problem by for-
mulating an objective function to determine the location
of fixturing supports,
• Describe the FEM methods used in computing the de-
flections of the workpiece that are needed in evaluating
the objective function, and
• Demonstrate this methodology for three different loading
cases.

Formulation of Objective Function for Fixture Design

«m Support Locations
The most important technical principles common to all fix-
Fig. 2 View of support positions on bottom surface of a rectangular
turing applications are: rigidity, stability, accessibility, and workpiece
most important of all is deflection of workpiece/fixture com-
ponents (c.f. ASTE, 1950 or Lee and Haynes, 1986). Mini-
mizing deflection thus becomes the key point to the fixture
design process and if optimal design methods are to be used, deflections are important. In metal working applications that
it becomes necessary to formulate an objective function that are usually chosen for accuracy, fixture deflections are also
captures the worst effect of the applied manufacturing load important design constraints. Because of these practical con-
on workpiece deflections. siderations, the analysis in the following sections is aimed at
determining the positions of fixture supports so that they min-
This is true for almost all metal working and assembly ap- imize the deflections of certain key points on a workpiece,
plications. For example, in airframe assembly, limiting de- given the workpiece geometry and material and an initial fix-
flection is emphasized above other design factors (ASTE, 1950), ture layout.
therefore, predicting and minimizing resulting displacements
of the workpiece are key factors in successful design in this Objective Function. Prismatic workpieces are chosen for
application. In fabrication, particularly in drilling or riveting, this analysis. In particular, consider a workpiece that has a

Nomenclature
TV^ = number of function
(BOLDFACE VARIABLES DENOTE MA TRICES) evaluations
TV'" = number of mid-edge
Q = (xCi, yCi, zCj)T, /= 1, . . , TVC, = corner nodes nodes
T>"-(dxi, dyi, dzi)T, i= 1, . . , TV", = displacement vector TV" = number of nodal
at nth function eval- points
uation Np = number of nodal
F"(Dy, S*) = objective function at loads
nth function evalua- Ns = number of supports
tion (single load . Pi = (PXi, Pyi, Pzi)T,i=l, . . , TV", = loads
case) S"=(X/> y*h Zsif, i= 1, • • , Ns, = supports at «th
L/ = (*i/> y\h Zu)T, i= 1, . . , Np, — load point nodes function evaluation
Mi=(xmi, ymh zmi)T, «'= 1, • • , Nm, = mid-edge nodes W = weighting matrix
TVC = number of corner (3x3)
nodes $"(/v(D y , S,)), i=l N\
TVe = number of octant j = Nc + N'" + N", k=\, . . , TVS = objective function at
elements «th function evalua-
TV1 = number of load tion (multiple load
cases case)

Journal of Engineering for Industry NOVEMBER 1991, Vol. 113/413

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

 finite element mesh consisting of Ne octant elements and N" The objective function in Eq. (2b) is set up to handle several
nodal points, Fig. 1. There are five tetrahedron elements load points, which constitutes a load case such as a gang-
embedded within each of the N€ octant elements. In assembly drilling operation. However most designs require several man-
processes (drilling, fastening, etc.), supports contact the bot- ufacturing steps, and in many cases each step requires another
tom surface of the part to provide support for the workpiece. fixture design. This is costly from a practical point of view
The 3-2-1 locating principle, which is a requirement for re- because separate fixtures have to be designed, built, stored,
peatedly locating a workpiece in a holding device, confines the and workpieces have to be refixtured for each operation. Even
six degrees of freedom in 3-D space of a part by providing six with fixtures that can be repositioned automatically, it would
tool points and clamping devices, (Colbert et al., 1986, and be undesirable to reconfigure the support positions for every
Pollack, 1976). This analysis uses the 3-2-1 locating principle load case because:- lead time is increased if a fixture is not
but concentrates on the three supports that form the primary flexible enough to undergo the reconfiguration as determined
plane, since three points define a plane. Arresting the remaining by the fixture design and analysis system, and the evaluation
degrees of freedom can be done with suction pressure, as is of optimal support locations for each load case requires the
done in aircraft assembly, or by clamping the top surface. The solution of a number of FEM problems that could prove to
boundary conditions for the finite element analysis are defined be costly.
such that the deflections D, at the supports, S,-, i= 1, .., Ns, A solution is to formulate an objective function that can
are zero (i.e., D(S,) = 0). Let us then consider the bottom optimize several load cases simultaneously, in order to generate
surface of a part, that rests on three fixturing supports, S,-, as a robust location for the supports. This compromise design
shown in Fig. 2. A point load Pi applied normal to the top objective function is to minimize the sum of the objective
surface is denoted by a vector functions resulting from the Nl load cases. The proposed func-
T T tion is given below.
Pi =~(Pxl,Pyl,Pzi) =(0.,0.,Pzl) (1)
The load is vertically translated down to the bottom node *(F,(D ; , Sk))=j^F,(Pj, S*) (3)
for the sake of the argument. When considering an arbitrary
edge, simulation results have shown that maximum deflections
occur at the corner nodes, C,, and at a node located along the
Weighting Function Selection. Assembly forces are in most
edge between the two corner nodes. When calculating these
cases applied normal to the workpiece surface, so that deflec-
deflections with FEM, a refined finite element mesh is desired
tions normal to the part are weighted, while the deflections in
to provide accurate deflections, but is economically infeasible
the workpiece plane are neglected, as indicated in Eq. (4a).
when considering a large number of tetrahedron elements.
Therefore, determining the exact location of the maximum 0 0 0
deflection at the middle node between the two corner nodes,
Q , along each of the four edges is not practical. Our experience W= 0 0 0 (4a)
using FEM simulations has shown that when a load is applied 0 0 1
away from an edge the maximum deflection is near the mid-
edge node, and when the load is close to an edge the maximum Machining processes, except perhaps for drilling, are applied
deflection shifts toward the load. Thus, rather than consider at an arbitrary direction to the top surface of the part. As a
all deflections, it is reasonable to minimize deflections at the result, deflections in all three coordinate axes are important,
corner and mid-edge nodes, and at the nodes where the loads as indicated by weights of one located on the diagonal in Eq.
are applied, which leads to the following: (4b).
• The net effect of arbitrarily positioned applied load points 1 0 0
on a prismatic workpiece causes critical deflections at the W= 0 1 0 (4b)
corner nodes and near the mid-edge nodes. 0 0 1
• Maximum deflections occur at the nodes at which the loads
are applied. Using the weights given by Eq. (4a) and Eq. (4b), the ob-
The formulation of the objective function for three dimen- jective function defined in Eq. (2a), and estimated machining
sional parts can be derived. In this case the design problem is: or assembly loads, the analysis determines how to position the
determine how to position the supports, S,-, ;'= 1, .., Ns. The supports.
point where a load vector is applied can be a point of large
deflections as identified by the above statements, so that the Objective Function Constraints. The objective function,
deflections at these locations, L,, for the Np loads along with Eq. (3), is comprised of terms given by Eq. (2b) expressed as
the deflections of the Nc corners, denoted by C,-, and the Nm the sum of the magnitudes of the displacement vectors at the
mid-edges of the workpiece edges, denoted by M,, are included critical nodes. Additional constraints to the support locations,
in the objective function given below. S,-, /= 1,.. Ns, are required otherwise an ideal case (no deflec-
tion) will develop by placing one of the S,'s under the load
Nc Nm
(note that a minimum of three supports are required, since
F;(Dj, St) = 2 (D (Q)WD(Q)) +2 (Dr(M,)WD(M,))"2
r 1/2
three points define a plane). Practically, however, this ideal
;'=1 1=1
case is undesirable when considering processes such as drilling
NP because it will result in a hole being made in the fixture support.
+ 2(Dr(L,)WD(L,))1/2 (2a) An over-hung structure problem could also develop in situa-
;=i tions where all the S,-'s are close to the load for large workpieces,
Nc Nm Np (i.e., airplane panels). This leads to Geometrically Constrained
= 2 IID'(C,-)II + 2 HD'(M,)II + J] UD'(L,-)II (2b) Regions (GCR's) for the supports dependent upon: the size of
/=i /=i 1=1 the structure and the accessibility of the supports to all the
nodes within the constrained region.
where D'(C,-), D'(M,-) and D'(L,) in Eq. (2b) are the weighted In circumstances where more than two supports exist within
displacement vectors at the corner, mid-edge and the load point the same GCR, inequality constraints are supplied to the op-
nodes, respectively. W is a matrix that weighs the importance timization software to keep the S,'s away from each other, as
of the deflections in the three orthogonal directions and will shown in Fig. 3. Therefore an inequality constraint is needed
be discussed later. and constraining only one degree of freedom for S2 and S3 is

414 / Vol. 113, NOVEMBER 1991 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

 Optimization of Support Locations Evaluate Objective Function, Eq (3)

I
Initialize the BFGS Algorithm: Setup for a set of supports:
1. Initial Support Positions, S ° 1. Apply constraints for all loads
2. Remesh
2. Initial Objective Function, O 3. Setm=l
i=0

Call Objective Function to Apply Load for m case

Estimate Gradients Numerically
Use FEM to solve for Deflections
Yes
Compute F in Eq (2)
for m load case

y ^ y Geometrically Constrained Region Do a Line Search: m = m+l

1. Increment S ' and evaluate <I>
C~i Region Determined By The SLFC Coefficient 2. Repeat until 3>is a minimum
along line of search
Fig. 3 Schematic of constrained regions
*
Update Support Positions
l.i . = i + l
satisfactory. Both supports in this case are allowed to have the 2. S = S Sum Fs to compute
same^-coordinate values but not the same x-coordinate values. 3. <D =0> Objective function, O
In general the constraint equations are defined as follows:
xsi< an xsj i, j = 1,2, .., Ns, i jtj (5a) Final Support Positions
. Return

ySi<(3ijySji,j=l,2, ..,Ns,i*j (5b) i.s ; = s !

2.<6 = <&'
where ay andftyare termed the Support Location Feasibility
Coefficients (SLFC's) between zero and one. In practice the
lower and upper limits are not desired since a value of zero
will force the support to exist on the boundaries of the GCR
and a value of one may overlay two or more supports. a. Optimization Alogorithm b. Evaluation of Objective Function
In summary, FEM computes the deflections, D,-, based on
a load case and constraints so that Eq. (3) can be evaluated Fig. 4 Flowcharts for picture optimization
for Af1 load cases. Optimization is then employed to solve the
following optimization problem:

minimize *(F,(Dj, S*)) = minimize' 2 *i(Dy, St) (6)

S»*=l../V* S„«r=l..A" '-'
< \-
subject to the constraints given by Eq. (5) and the GCR's.

Optimization and FEM Software

—0
The objective function, application of both the GCR's and
SLFC's constraints and the methods for computing deflec-
FEM Node § Optimization Solution
tions, all described in the previous section, have been imple-
mented for assisting a tool engineer in the analysis of fixture Fig. 5 Remeshing
designs. This code, written in FORTRAN 77, has been de-
veloped to use a boundary representation of a workpiece and
carry out the fixture design optimization is termed FAST (Fix- parts and the Sky-Line and the Choleski-Factorization methods
ture Analysis and Synthesis Tool) and is described by Menassa that reduce storage and computational requirements within the
(1989). It consists of two major parts shown in Fig. 4. As Fig. software package (Cook, 1981, and Zienkiewics, 197). Deter-
4(a) and (b) indicate, the optimization algorithm calls the FEM mining displacements at the mid-edge nodes is easiest when
routine for each objective function evaluation. the FEM mesh has an odd number of nodes along the x and
The first part, shown in Fig. 4(a), consists of an optimization .y-axis of the workpiece, but the software linearly interpolates
software directed at determining the optimal position of the the mid-edge nodes displacements when the mesh generated
supports so that minimal deflections are achieved. The efficient has an even number of nodes.
feature to the optimization software is the implementation of Because the fixture supports are both constraints to the FEM
the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method, problem and the design variables for the optimization, the
known to be the best quasi-Newton method that avoids second optimization problem becomes nonlinear requiring the algo-
derivative calculations, (Reklaitis et al., 1983, and Papalam- rithms described before and the FEM problem requires re-
bros and Wilde, 1988). The advantage of the BFGS method meshing at each optimization step. Remeshing is needed because
is good convergence with low accuracy line searches which is the optimization solution seldom matches the coordinates of
particularly important for this optimization because each func- the nodes in the original finite element mesh. Remeshing, how-
tion evaluation involves solving a FEM problem. ever, is performed locally such that the FEM node closest to
The second part, Fig. 4(6), is used to evaluate the objective the optimization solution is moved as portrayed in Fig. 5, with
function, given by Eq. (3), and consists of three dimensional this applied along the z-axis to maintain the integrity of the
(3-D) Finite Element Methods analysis software directed at tetrahedron element.
determining deflections of 3-D parts and a remeshing routine. To aid in understanding the optimization, the workpiece
Additional features of this FEM code are the linear and par- geometry, applied loads and constraints, as well as the initial
abolic tetrahedron elements used to generate a mesh for the and final fixture support locations are displayed using a Pro-

NOVEMBER 1991, Vol. 113 / 415

Journal of Engineering for Industry

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

 Table 2 Load case 1 with two loads applied at 45 deg. angle

Load point nodes, NP = 2 Li = (2 cm, 2 cm, 0.5 cm) T

L2 = (2 cm, 6 cm, 0.5 cm) T
Loads Pi = (-200N, 0 , -200N) T
P2 = Pl
Initial objective function 05° = 0.413 mm
Final objective function * * = 0.091 mm
Final support locations, N s = 3 S*i= ( 4.00 cm, 4 cm, 0 cm) T
S*2= (12.36 cm, 2 cm, 0 cm) 1
S*i= (12.36 cm, 6 cm, 0 cm) T

Table 3 Load case 2 with two loads applied normal

Load point nodes, NP = 2 Li = (16 cm, 2 cm, 0.5 cm) T

L2 = (16 cm, 16 cm, 0.5 cm) T

* Initial Support Location, + Nodal Loads Loads Pi = (0 N, 0 , -200 N) T

P 2 = Pi
Fig. 6 FEM mesh of workpiece Initial objective function 0 ° = 0.164 mm
Final objective function <D* = 0.049 mm
Table 1 Values used in optimization example for all three load cases
Final support locations, N s = 3 SV(11.00cm,4cm,0cm)T
Workpiece Dimensions 18 cmx 8 cm x 0.5 cm S* 2 =(13.34cm,2cm, 0cm) T
Elastic Properties of Aluminum E = 69 GPa and \i = 0.33 S V (13.34 cm, 6 cm, Gcm)T
Comer nodes, N c = 4 Ci = ( 0cm,Q cm, 0 cm)r
C2 = ( 0 cm, 8 cm, 0 cm) T
Table A Load case 3 for combined load cases 1 and 2
C3 = (18cm,8cm,0cm) T
C4 = (18cm,0cm,0cm) T Load point nodes, NP = 4 Li = ( 2 cm, 2 cm, 0.5 cm) 1
Mid-edge nodes, N m = 4 Mi = ( 0 cm, 4 cm, 0 cm) T L2 = ( 2 cm, 6 cm, 0.5 cm) T
M2 = ( 9 cm, 8 cm, 0 cm) T L3 = (16 cm, 2 cm, 0.5 cm) T
M3 = (18 cm, 4 cm, 0cm) T L4 = (16 cm, 16 cm, 0.5 cm) T
M4 = ( 9 cm, 0 cm, 0 cm) T Loads Pi = (-200 N, 0 , -200 N) T
Initial support locations, N s = 3 S°i=( 8 cm, 4 cm, 0 cm) T P 3 = ( ON, 0 , - 2 0 0 N ) T
T
S°2= (12 cm, 2 cm, 0 cm) P 2 = Pl,P4 = P3
S^= (12 cm, 6 cm, 0 cm)T Initial objective function 4>° = 0.576 mm
Geometrically Constrained Regions (GCR's) x = 4 cm, y = 0 cm Final objective function <J>* = 0.350 mm
x = 14 cm, y = 8 cm Final support locations, N s = 3 S*i= ( 4.00 cm, 4 cm, 0 cm) 1
Support Location Feasibility Coefficients ai2 = ai3 = 0.9 S*2= (11.34 cm, 2 cm, 0 cm) T
(SLFC's) 023 = 0.9 s V ( H . 3 4 c m , 6 c m , 0cm)T

grammer's Hierarchical Interactive Graphics System (PHIGS) therefore, inequality constraints are needed to keep Si at a
as described by Abi-Ezzi and Kader (1986). This graphics in- distance away from the two remaining supports and to keep
terface is also capable of time simulation of the optimization S2 and S3 away from each other. Three SLFC's from Eq. (5a
process. and b), are assumed to be as follows: a12 = a13 = 0.9 will limit
In summary, the optimization technique (BFGS) selects fix- support S! in the x-direction to ninety percent of the x-coor-
ture support positions, S,-, that minimizes workpiece deflections dinate for S2 and S3, and /323 = 0.9 will limit support, S2, in the
by solving the optimization problem given by Eq. (6). j-direction to ninety percent of the distance to support S3.
(These data on SLFC's, as well as all the other data that remains
Optimization Example constant in the following examples, are in Table 1.) Here again,
selection of these constraints depends on practical consider-
To illustrate this proposed optimal design methodology, as- ations. For example, the size of the supports in the primary
sume that the problem is to design a fixture that will be used reference plane will limit how close the supports can be to each
to drill holes in a small aluminum part. A flat, relatively thin other, or if a subplate with locating holes is used, the closest
piece of aluminum is assumed, with an elastic modulus and positions of the supports will depend on the hole pattern.
Poisson's ratio of is = 69 GPa and ^ = 0.33, respectively, and
dimensions 18cm x 8cm x 0.5 cm. Furthermore, it is assumed Three cases: Case 1, Case 2, and Case 3 are used to illustrate
that the 3-2-1 locating principle is used to select the initial the optimization method. For Case 1, two loads are applied
configuration of the fixture, and, as is common in the aircraft at a 45 degree angle to the surface. Case 2 assumes a second
industry when drilling panels, suction cups are used as supports operation where holes have to be made on the opposite end,
so in these examples clamping is not required. The design but normal to the workpiece surface. Case 3 assumes that one
problem is to determine the position of these 3 supports; this fixture must be designed that will be used to handle the op-
means that Ns=3. erations in Case 1 and Case 2. Tables 2, 3, and 4 summarize
the numerical values used to represent these loads and their
Setting the constraints is also part of the design problem locations. The above optimization examples can be thought
that is related to many practical design considerations. Figure of as an idealization of cases where large thrust forces are
6 shows an FEM mesh of the example workpiece, with loads required in drilling or large upsetting forces are required on a
indicated on either end. Because of these assumptions about blind fastener.
the physical problem, i.e., loads applied at arbitrary positions
to the relatively thin surface, the weighting matrix in Eq. (4b) Load Case 1 and 2. Two loads are applied to one end of
is used when evaluating Eq. (3). Since these loads are assumed the part at a 45 degree angle to the top surface in Case 1, Fig.
to be the result of drilling holes through the aluminum, the 1(a). For Case 2, two loads are applied normal to the surface,
supports should not be directly under them. For this reason, but opposite the loads in Case 1, Fig. 8(a), as might happen
the GCR's are defined to be a rectangular area where the in a gang-drilling operation. The initial support positions des-
supports can be located, with the values given in Table 1. In ignated by the S°,'s in Table 1 were the same for both Case 1
this study the supports, S„ /'= 1,.., Ns, exist in the same GCR, and Case 2. The results for Case 1 are in Table 2, and indicate

416 / Vol. 113, NOVEMBER 1991 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

 a. Initial Support Locations
a. Initial Support Locations

b. Final Support Locations b. Final Support Locations

* Initial Support Location, \_j Final Support Location, + Nodal Loads * Initial Support Location, [J Final Support Location, + Nodal Loads
Fig. 7 Bottom surface of part for case 1 (c.f. Table 2) Fig. 8 Bottom surface of part for case 2 (c.f. Table 3)

about an 80 percent reduction in the objective function (com- onstrated in load Case 3. However, in situations where many
pare *° with $*) when the optimization algorithm is used to manufacturing loads are applied to a part, although the ob-
adjust the supports from the S°,'s in Table 1 to the S*,-'s in jective function is minimized, the required tolerances may not
Table 2. As expected, support S\ moves toward the load since be achieved for all the combined load cases at hand. This is
the load point nodes, L„ are nodes of large deflections while illustrated by Fig. 10 which shows the behavior of the objective
support S2 and S3 retract back in order to smooth out the function for these three cases. Figures 10(a) and (b) are designs
workpiece deflections, Fig. 1(b). Similarly, the results in Table for handling a single load case, while Fig. 10(c) illustrates a
3 for Case 2 show the optimization process achieved a 70 compromise design that can handle both cases. As can be seen
percent reduction in the initial objective function. The final in Fig. 10 the compromise solution has a larger objective func-
configuration of the supports, however, may result in an over- tion value than the two separate cases. A solution would be
hung structure if large parts are considered due to the fact that to assume that the part could be divided into regions and apply
the supports are close together, Fig. 8(b). In this instance, the the proposed techniques to each region, thereby increasing the
SLFC's ctn and an, may be reduced to avoid this problem. size of the OCR so that additional local minima can be found.
The strength of the BFGS algorithm in solving this nonlinear
Load Case 3. Case 3 represents a fixture design problem optimization problem lies in its good convergence as illustrated
where one fixture needs to be designed to handle Case 1 and in Fig. 10. In each of the three cases the objective function is
Case 2 in sequence, Fig. 9(a). As would be expected, this minimized in three or less optimization steps.
represents a design tradeoff. In this instance support Si moved
toward the inclined loads, Fig. 9(b), and the constraints are
active at the optimal solution which means that without the Conclusions
constraints, Si would be located under a drill. The remaining This paper has treated the fixture design as an optimization
supports, S2 and S3, moved slightly toward the inclined loads problem where support positions are the design variables and
since the normal loads are smaller in magnitude than the 45 the objective function is based on workpiece deflections. The
degree applied forces. In this case as indicated in Table 4, an specific conclusions that can be made based upon the work
optimal solution is found that minimizes the objective function are:
8
by 40 percent; less than for the individual load cases that An objective function for optimizing the design of fixture
comprise this problem. This is part of the design tradeoff and support locations was defined that used deflection of the
it should be remembered that the $ * value is the sum of the workpiece as the prime design criterion. Based on simu-
deflection for the two load cases so it is bound to be greater lation experience, only deflections of corner, mid-edge,
than the result for either Case 1 or Case 2. and load point nodes of a workpiece were used, and the
The advantage to this fixturing design methodology is that BFGS algorithm was used to minimize this objective func-
several load cases can be handled in one fixture setup as dem- tion.

Journal of Engineering for Industry NOVEMBER 1991, Vol. 113 / 417

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms

 2 3
Iteration Step
a. Case 1

a. Initial Support Locations

1 2
Iteration Step
b. Case 2

.8 (p"

.8
b. Final Support Locations O
* Initial Support Location, [jj Final Support Location, + Nodal Loads
Fig. 9 Bottom surface of part for case 3 (c.f. Table 4) 0 1 2
Iteration Step
c. Case 3
• Deflections were calculated using FEM, with local re- Fig. 10 Minimization of objective function for three load cases
meshing performed as the support positions were moved.
8
This method was illustrated using three simple examples, American Society of Tool Engineers (ASTE), Dec. 1950, Practical Design of
where in each case the reduction in the objective function Manufacturing Tools, Dies, and Fixtures, First Edition, McGraw-Hill.
was at least 40 percent. Solutions required about 300 CPU Asada,H.,andBy, A., June 1985, "Kinematic Analysis of Workpart Fixturing
seconds per case on an IBM 4341. for Flexible Assembly with Automatically Reconfigurable Fixtures,'' IEEE Jour-
nal of Robotics and Automation, Vol. RA-1, No. 2.
While the examples with prismatic parts were simple, they Colbert, J., Menassa, R., and DeVries, W., 1986, " A Modular Fixture for
illustrate the basic principles of this proposed method of op- Prismatic Parts in an FMS," Proceedings of the Fourteenth North American
timizing a fixture design and how a compromise design might Manufacturing Research Conference, pp. 597-602, May 28-30.
be achieved. This emphasis of this fixture design optimization Chou, Y-C, Chandru, V., and Barash, M. M., 1989, " A Mathematical Ap-
proach to Automatic Configuration of Machining Fixtures: Analysis and Syn-
was on deflections due to loads applied during manufacture; thesis," ASME JOURNAL OF ENGINEERING FOR INDUSTRY, Vol. I l l , No. 4, pp.
however, some of the other fixture design methodologies con- 579-584.
sider the surface finish of workpieces and the selection of a Cook, R. D., 1981, Concept and Applications of Finite Element Analysis,
variety of workholding elements. Incorporating these design 2nd Edition, Wiley-Interscience.
criteria and design variables is an area for future research in Lee, J. D., and Haynes, L. S., 1987, "Finite Element Analysis of Flexible
Fixturing System," ASME JOURNAL OF ENGINEERING FOR INDUSTRY, Vol. 109,
optimization and the application of expert system techniques. pp. 579-584.
Mani, M., and Wilson, W., 1988, "Automated Design of Workholding Fix-
tures Using Kinematic Constraint Synthesis," Proceedings of the Sixteenth North
Acknowledgment American Manufacturing Research Conference, pp. 437-444.
Menassa, R. J., 1989, "Synthesis, Analysis and Optimization of Fixtures for
The authors would like to acknowledge the financial support Prismatic Parts," Ph.D. Thesis, Rensselaer Polytechnic Institute.
of RPI's Center for Manufacturing Productivity through its Menassa, R. J., and DeVries, W. R., 1989, "Locating Point Synthesis in
Automated Fastening Technology Program for this work. Also, Fixture Design," CIRP Annals, Vol. 38/1, pp. 165-170.
technical contributions of Dr. G. A. Gabriele and Dr. M. Papalambros, P. Y., and Wilde, D. J., 1988, Principles of Optimal Design:
Modeling and Computation, Cambridge University Press, pp. 338-343.
Georges are greatly appreciated. Pollack, H. W., 1976, Tool Design, Reston Publishing Comp., Inc., Reston,
Virginia.
Reklaitis, G. V., Ravindran, A., and Ragsdell, K. M., 1983, Engineering
References Optimization, Methods and Applications, Wiley-Interscience.
Abi-Ezzi, S., and Kader, S., July 1986, "Phigs in C A D , " Computers in Zienkiewics, O. C , 1977, Finite Element Methods, 3rd Edition, McGraw-
Mechanical Engineering, Vol. 5. No. 1, pp. 28-36. Hill.

418 / Vol. 113, NOVEMBER 1991 Transactions of the ASME

Downloaded From: http://manufacturingscience.asmedigitalcollection.asme.org/ on 08/15/2013 Terms of Use: http://asme.org/terms