2. 354 21stInternationalConferenceon Computersand IndustrialEngineering
above lines. Figure 1 shows the spring at various load conditions. To formulate this problem, following
additional notations, expressions and data are assumed based on [1, 2]:
fo •L! -L.
e
/ I I Installed
,, ,, L, ~- -----rrop,,,t,, ~ T
I I I I I I length, Sollid
i, ,i ,i L. 17,th.
ip w n
adjacentcoila)
Figure 1" Compression Spring at Various Load Conditions ( Adapted from Mott, R.L.)
Number of inactive coils, (Q) = 2; Pitch (p) = (Lf- 2d)/N; Shear modulus (G) = (1.15E+07) lb/in2;
Weight density of spring material ('1,) = 0.285 lb/in3 ; Gravitational constant (g) = 386 in/sec 2
Mass density of material (p = T/g); and p = (7.383 42E-04) lb-sec2/in4
Frequency of surge waves ta - 2riD 2N
Pitch Angle ~.= tan-I [__~__] (2)
Coil Clearance cc = (Lo-Ls)/N (3)
Spring Index C = D/d (4)
Load deflection equation P = K~ (5)
d4G
Spring Constant K = - - (6)
8DSN
8kPD
Shear stress z = ~ (7)
nd 3
Wahl stress concentration factor k= (4 D - d) + 0.615d (8)
(4D - 4d) D
Mass of the spring M=I / 4(N + Q)n 2Dd~p (9)
Mathematical Model for Optimization
For a spring under tension or compression, the wire experiences twisting which causes the shear stress.
This is one of the most critical factors in spring design. The cost of the spring is another important
factor and may be considered as directly related to the mass of the spring. Objective of the design
problem is to minimize mass and shear stress while satisfying other constraints. A mathematical model
for spring design optimization problem will be as follows:
Minimize Z:
Z=WI(1/4(N+Q)~2Dd2p}+W2( --~5-~8PD.(-~ --4"d )(4D - d) ÷ 0.615dD ) ]" (10)
Subject to:
8PDSN _ ~ (11)
d4G
d "12~ > (12)
2~D2N - 6J oI - r
D + (ll/10)d <- D (13)
tan-,[n__.~._]< ~0 (14)
3. 21st International Conference on Computers and Industrial Engineering
(Lo-Ls)/N > d/10
Did > C
8PD((4D-d) + 0.615d )< '~d
/td 3 (4D - 4d) D
dG(L/ - dN) ( (4D - d) 0.615d
r~D2N (4D - 4d) + --D~
1/4(N +Q)g2Dd2p <.085
where: N is an integer and d, D and N > 0
(15)
(16)
(17)
) -< Xa (18)
(19)
355
Integration of Robustness & Durability
In design of any component it is necessary to assign tolerances to all the dimensions and consider
variability of all the inputs and outputs. The combination of all the tolerances with variability of inputs
should guarantee, at least from a statistical basis, that the system will perform as expected. Usually
there is a conflict in design and manufacturing engineers' interest for tolerances. This conflict of interest
can be resolved if design engineer consider tolerances as a decision parameter in overall design
optimization problem. We need to understand how these tolerances effect the output of the system and
tolerances which has significant impact may be kept tighter while other ones may be kept loose.
In our example of spring design we will consider tolerances of mean coil diameter (D), spring wire
diameter (d), and the variability of applied load as uncontrollable or noise. Desired outcome of the
spring is deflection at the given load condition. We are considering; tolerances of mean coil diameter (D)
= .004, .005 and .006 in.; tolerances of spring wire diameter (d) = .001, .002 and .003 in.; variability
in load (uncontrollable) = 14.00 + 0.25 and variability in deflection (acceptable range) = 0.5_o~÷o5
We know the relationship of deflection with mean coil diameter (D), spring wire diameter (d) and
applied load (P), and will use Taylor's series expansion to fred out variability of deflection. (G ~ ), as
function of D, d and P and their variances. Using this relationship we can fred out the probability of
getting deflection with in acceptable range for a given value of D, d and their tolerances. We are
assuming that tolerances are equal to the standard deviation. Therefore standard deviations of D, d and
P (Go,G d and Ge), can be estimated using the respective tolerance. To calculate the out of
specification product or probability of getting acceptable deflection, we need to fred the value of
standard normal variate Z for upper and lower limit of deflection using the following relationship:
Z = x - ~t (20); R = q~(Zt: ) --q~(ZL) (21)
(Y
Here tp(Z) represent the area under standard normal probability distribution up to the point Z. In our
case kt = 0.5 in. and x= 0.4 and 0.7 in., for ZL and Zu respectively. We need to add terms for the cost
associated with tolerances in the objective function also extra constraints will be added to ensure
minimum acceptable probability of getting deflection with in acceptable range. New optimization
problem will have five decision variables D, d, P, To (Tolerance for D) and Td ( Tolerance for d). It is
assumed that, (1000*TD) and (1000*Td) are integer. Equation (22) represent the objective function for
this optimization problem. Equations (23) & (24) ensures proper selection of tolerances and equation
(25) ensures that more than 95% of springs provide deflection within specified limits.
8PD ( (4D-d)Z=W1{I/4(N +Q)x2Od:p }+W2{ nd 3 x +
7-ff--47)
6 > (1000*TD) >4
3 _>(1000*Td) >1
R > 0.95
0.615d ) }+Wa{1/TD}+W4{1/Td} (22)
D
(23)
(24)
(25)
New objective function represented by equation (22) and constraints are reflected in equations (11) to
(19) and (23) to (25). As we do not have any information about preferences we have selected weights
for objective function such that each term is of the same magnitude. The solution # 1 in Table 1,is for
this model and it provides satisfactory mass and stress condition while ensuring that 95% of springs
provide deflection with in acceptable range.
4. 356 21st International Conference on Computers and Industrial Engineering
In previous formulations we have put an equality constraint on the deflection of spring to ensure that
final solution provides us target deflection at the mean value of decision parameters. However, in reality
our objective is to ensure functionality and not the target value. Taguchi has recommended the concept
of loss function. According to this concept we have no loss if outcome is at the target value. But as it
deviates from the target, the loss is depicted by a quadratic function [3]. We are recommending a loss
function represented by a step function instead of a continuos function. Advantage of step function is
that calculation of loss as in our case of spring deflection becomes very simple. Further, this will be
more representative of real life situations in many cases as there is always a limit on sensitivity of the
customers or of the system for which our output is an input. Spring design problem was solved by
applying the concept of step loss function. We assumed that the loss for 0.45 to 0.60 inch of
deflection as 0 units; for 0.40 to 0.45 and 0.60 and 0.70 inch of deflection a loss of 1 units and for
less than 0.40 and greater than 0.70 inch of deflection a loss of 2 units. Objective function for this
model will be:
Z = Wl{Quality Loss} + W2{ 1/TD}+W3{1/Td} (26)
The solution # 2 in Table 1 is for this model and it provides a good robust design with minimum quality
loss and best possible combination of tolerances selected. However, this solution fails to consider the
aspect of durability which depends upon the degradation. As product becomes older, its performance
degrades. To represent this real life scenario we should integrate the information on degradation of
design parameters with the changing customer expectations to calculate loss in quality. We calculated
loss in quality at the end of warranty period (3 Yrs) and also at the end of designed life (10 Yrs) of the
spring assuming degradation in spring wire and mean coil diameter. With usage and environmental
effects spring wire diameter will reduce at certain critical sections, and mean coil diameter will increase.
We are assuming an empirical relationship( C = A tB ) for degradation of D and d.. Here C is the
amount of degradation in t time units while A and B are empirically determined parameters [4]. The
specific relationship we have used are: d(t) = d - 0.0009t°65and D(t) = D + 0.001t°'65
Solution # 3, in Table 1 provides the results for this model. This solution considers degradation, and
interesting point to note is the value of deflection, which is towards lower side as we understand that
spring will provide more deflection with usage. It will be better to have little less deflection in the
beginning though it may mean more rejection at the manufacturing stage as tolerance is tight towards
lower side, but over the complete life of spring it will be a better choice.
Table 1: Summar of All the Solutions
Solution D d N TD Td Mass Deflec Stress Reliab
# inch inch # millie millie lb. inch Kpsi. %
1 0.513 0.069 9 6 2 0.050 0.500 62.47 95.00
2 0.516 0.076 13 5 1 0.081 0.523 49.21 99.90
3 0.513 0.078 13 4 1 0.085 0.463 45.57 95.00
Discussion
By comparing the models developed and solved in this example, many useful and interesting facts
comes out. First, model brings out the advantage of considering manufacturing options in terms of
design parameter tolerances. Second, model shows how concept of loss function should be integrated
in the design optimization model. Finally, model provides a Decision Support System for designer to
integrate the aspects of robust, and durable product design. Results of these models will change if we
select different weight factors. However, these models provide a decision support to the designer who
may try different combinations before making final choice.
References
1. Mott, R.
2.
3.
4.
L., 'Machine Elements in Mechanical Design', 2~d ed. Maxwell Macmillan
International,1992
Arora J. S, Introduction to Optimum Design., New York: McGraw-Hill, 1989
Taguchi, G. Introduction to Quality Engineering, Asian Productivity Organization, 1990
Albrecht P. , Naeemi A. H., Performance of Weathering Steel in Bridges, National Cooperative
Highway Research Program report 272, p. 65, July 1984