Int J Adv Manuf Technol (1994) 9:362-368 Intm~ioMl Journal of

9 1994 Springer-Verlag London Limited

Advanced
manufactumg
Technologu
An Automatic Tolerance Assignment Approach for Tolerance
Charting
Ping Ji
Department of Mechanical and Production Engineering, National University of Singapore, Singapore

There are two dimensional sets in a tolerance chart: the Furthermore, the outcome from the approach is compared
blueprint (B/P) dimensional and the working dimensional with the original manual result and the solution from the LP
group. The interrelation between these dimensional groups model. Finally, the proportional smoothing approach is
makes it very difficult to assign appropriate tolerances to extended to a statistical tolerance model.
individual working dimensions. This paper deals with the
automatic tolerance assignment problem in a tolerance chart,
especially for computer-aided tolerance charting. After the 2. Background
manual methods of tolerance assignment are briefly discussed,
a linear programming model is analysed. Owing to the The detailed development and calculation procedure for
impractical solution from the linear programming ( LP) model, manual tolerance charting can be obtained elsewhere [9]. Fig.
a proportional smoothing approach is described with a numerical 1 shows a tolerance chart for a steel plug [9], which was
example, which can assign appropriate tolerances automatically modified with the representation of dimensional chains [8].
and systematically, based on the requirements of both blueprint The units in this example are inches (1 in = 25.4 mm).
dimensions and process capabilities. The result from this Tolerance assignment allocates appropriate tolerances to
approach is compared with that of the LP model and the each constituent link, arising from the tolerance of the closing
original manual procedure. Finally, the approach is extended link, in a dimensional chain9 For example, B/P dimension
to a statistical tolerance model. 1.000 -+ 0.020 has four constituent links: working dimensions
6, 7, 8 and 9. The tolerance summation of these working
Keywords: Dimensional chain; Linear programming; Pro- dimensions must not exceed the required tolerance value
portional smoothing; Statistical tolerance model; Tolerance 0.020 (the bilateral tolerance system is discussed in this
assignment; Tolerance chart
I -

1. Introduction
IS C D~l

Several computer-aided tolerance charting systems [1-5] have U c~/aon Wo~mg D~. Sli~k Rcn~,~al
been developed, and some theoretical research has been done NO. K,m~ -1-Tid. ~ C'lulm
I O,9"/9 003 Solid
on, tolerance charting [6, 7]. However, automatic tolerance I,q~4 .003 Solid
"%" I0 T~ml
assignment is still a problem, which delays the full automation ,003
4 4.031 .010 Solid
of a computer-aided tolerance charting system. A linear
3 20 T~,DS I.OO~l (io.I ,(1"~o (117 -3 4 .3
programming model was proposed [8], which considers the 6 30 1 ~.lm#
blueprint (B/P) design dimensions and process capabilities
i

7 4o I Onndml 1,000 .00~ ,017 .014 .7 4) $ 3 -2

x 50 r 1,1~0 (115 .u32 (KKI -1~.7.63 J .I
systematically, and can assign tolerances to working dimensions .l.q~l onl .Oll o~4 .q 3 J
6o Gn~m~
automatically. This paper analyses the linear programming
(LP) model, and more constraints will be added to the model
to make the solution more reasonable. Then a proportional
smoothing approach is introduced with a numerical example. B~Duncn~m
I +Tul.
I 4.I~
P.c~l~amDIr ~ ' ~ m
:t Tqd.
t~f y
1.000 ,0~ I (xK) o2U ,I .~ -7.6
I "
I
Correspondence and offprint requests to: Ping Ji, Department of le la
Mechanical and Production Engineering, National University of
Singapore, 10 Kent Ridge Crescent, Singapore 0511. Fig. 1. Tolerance chart for steel plug.
 Automatic Tolerance Assignment 363

paper). The problem here is how much tolerance each working dimension, the interrelation of the two dimensional groups
dimension should have. Traditionally, an experienced process must be considered systematically, that is, all dimensional
engineer assigns tolerances according to his knowledge and chains have to be studied together. This leads to the
experience, and then checks the limitation of B/P dimensions formulation of a linear programming model.
and process capabilities. Obviously, it is very difficult for a
computer to follow this approach, even for an expert system.
Besides this empirical approach, there are another two manual 3. Linear Programming Model
methods of assigning tolerances for a single dimensional chain:
the equal tolerance and the equal tolerance grade method. A linear programming model was formulated in [8], and is
The equal tolerance method assigns the required tolerance of presented as follows:
the closing link to each constituent link equally, that is, each
constitutent link has the same tolerance. For example, working Maximise ~'~ tj (1)
j=l
dimensions 6, 7, 8 and 9 have an equal tolerance 0.005
(=0.020/4) from the B/P dimension 1.000---0.020. This subject to
method is very simple, however, it is impractical if some
ti <--sri (i=n+l,n+2 ..... m) (2)
constituent links are rough machining operations, and some
j~-Sg i
are final precision machining. For example, the stock removal
of operation sequence no. 8 has constituent links 1, 3, 5, 6, tj <- bp~ (i = 1,2 ..... n) (3)
7 and 8. Working dimensions 1, 3 and 5 are turning operations, j~-BP i
while working dimensions 6, 7 and 8 are grinding operations.
In this case, the equal tolerance grade method is more tj->ej (j = 1, 2 ..... m) (4)
appropriate. It groups the constituent links according to their (LP1)
operations, and then assigns unequal tolerances to different where n number of B/P dimensions
groups, the links in the same group having equal tolerances. m number of working dimensions
However, how much tolerance is assigned to each group is tj tolerance of the # h working dimension
dependent on the process engineer. This method does not sr~ minimum allowed tolerance of ith stock removal
consider the nominal sizes of different dimensions. As seen
bpi tolerance of the ith B/P dimension
in Fig. 1, working dimension 9 should have a larger tolerance
than working dimension 6, 7 or 8, if possible. SRi constituent link set of the ith stock removal
The previous discussion was concerned about one dimen- BP~ constituent link set of the ith B/P dimension
sional chain only. In a tolerance chart, all B/P dimensions ej minimum economic tolerance for the #h working
and working dimensions are related to each other via dimension
dimensional chains. A dimensional chain cannot be studied and
without consideration of other dimensional chains. The
dimensions in a tolerance chart can be classified as two sets, sri = min{(SRmeani- SRmini),
e.g. the B/P dimensional and the working dimensional group. (SRmaxi - SRmeani)} > 0
The former is the requirement, while the latter results from
the former to ensure that all B/P dimensions (including ( i = n + 1,n + 2 ..... m) (5)
tolerances) are guaranteed, so the latter depends on the where SRmini minimum allowed stock removal of
former. In general, the process engineer cannot change the sequence no. i
B/P dimensions, while he can freely modify his process plan, SRmaxi maximum allowed stock removal of
so, the former is fixed, while the latter can be changed and sequence no. i
the two dimensional groups are interrelated. For example,
SRmeanl mean stock removal assigned for sequence
working dimension no. 9 is involved in the dimensional chains
no. i
of all B/P dimensions. On the other hand, B/P dimension
1.000 +- 0.020 should be guaranteed by working dimensions The objective function (1) is to maximise the cumulative
6, 7, 8 and 9 (the constituent links of the dimension chain). tolerance of each working dimension, which corresponds to
This interelation makes it more complicated to assign tolerances each tolerance assigned as being as large as possible. Constraint
for all dimensional chains than for just one dimensional chain. set (3) is derived from B/P dimensional chains to meet the
For example, B/P dimension 4.000 +- 0.005 has only one design requirements. Constraint set (4) is required from the
constituent link, working dimension no. 9, so it is intuitive minimum economic capability of each process sequence.
to assign a tolerance to the working dimension of "--0.005. Constraint set (2) and equation set (5) are derived from:
However, working dimension no. 9 is also involved in B/P SRmini -< SRmeani --+ ~ ti <- SRmaxi
dimensions 1.000 +_ 0.020, 2.000 --- 0.009, and 3.000 --- 0.002. jESR i
The tolerance ---0.005 is probably acceptable for B/P dimen-
sions 1.000 --- 0.020, and 2.000 --- 0.009, however, it is defi- (i=n+l,n+2 ..... m)
nitelyunacceptable for B/P dimension 3.000 +- 0.002. In this which reflects the dimensional chains of stock removal, and
case, B/P dimension 3.000 -+ 0.002 should be considered first. ha~ o n l y m - n because there are n solid stock removals.
In order to obtain a proper tolerance for each working The detailed derivation can be obtained elsewher(~ [8].
364 P. Ji

The model LP1 can be written in a matrix format: recommended chart, as shown in Fig. 2 [10], so the m i n i m u m
economic tolerance can be obtained for each working dimen-
Maximise
sion from Table 1 [10]. The solution of the LP model is: h
IT (6) = 0.0274, t2 = 0.0126, t3 = 0.0006, t4 = 0.0160, t5 = 0.0004,
t6 = 0.0002, t7 = 0.0002, t8 = 0.0002, t9 = 0.0018, and
subject to
xg=~ tj = 0.0594. From this solution, several tolerances (e.g.
AT-<Q (7) t3, ts, tr, t7, /8) are included in the m i n i m u m economic
tolerances because the simplex method always gets the optimal
T-E (8)
solution at one of the corner points, that is, at the boundary of
(LP2) the constraints. This solution is acceptable, but unreasonable,
where 1 = [1, 1 . . . . 11 because it is difficult to machine those working dimensions
T = [t,, t2. . . . . t.,] T which have the m i n i m u m economic tolerances. This problem
can be illustrated by the tolerance ratios. For example, the
A =[aii]
original ratio of e9/ea is 1.5, however, the final tolerance ratio
{ 10 if tJ is a c~ link ~ the of I9/t8 (=tg/es) is 9, which means that 19 has been made 6
ith dimentional chain of (i,/' = 1, 2, times larger while ts is still the same. This is unavoidable
a0 = either stock removal or .... rn) owing to the simplex method once some constraints have
B/P dimension become redundant. For example, the constraint t9 -< 0.00500
otherwise plays no role because of the constraint t6 + t9 -< 0.00200.

Q = [qb q2 . . . . . qm] T, and qi(i = 1, 2 . . . . . m ) is the

corresponding value of the m i n i m u m allowed tolerance of a
stock removal, or a tolerance of a B/P dimension, and qi > 0;
5. Proportional Smoothing Approach
E = [e,, e2. . . . . em] r From the above discussion, it is necessary to add more
The condition of the existence of a feasible solution of LP1 constraints to the LP model. Naturally, the following con-
or LP2 is: straint:

Q - A E -> 0 (9) tl _ tz _ ... = t,,, (10)

el e2 era
where 0 = [0, 0 . . . . . 0] T.
The model LP1 or LP2 can be solved by the simplex can be used to modify the LP model. Thus the result is: tl
method. If the dual problem of LP1 or LP2 is formulated, it = 0.0016, t2 = t3 = 0.0024, t4 = 0.0028, t5 = 0.0016, t6 = t7
is easier to obtain the solution from the dual than from the = ts = 0.0008, t9 = 0.0012, and Y.9=~tj = 0.0120. The
primal [8]. formulation of this direct smoothing is not a linear program-
ming model any more, although obtaining the solution is
much easier and faster than by the simplex method. The
4. Example tolerances t6, tT, ta and t9 are smoothed, and more reasonable
than those of the LP model. However, the tolerances tl, tz,
Take the steel plug for an example, the problem is formed t3, 14 and t5 are very tight, compared with the solution of the
as follows: LP model or the manual result. In order to improve the
direct smoothing approach and the LP model, a proportional
Maximise t~ + 12 + t3 + 14 -} 15 + t6 + 17 + ts + t9 smoothing approach is proposed and illustrated with the
subject to previous numerical example.
First, a table similar to that used by the simplex method
t3 + t4 + t5 -< 0.01700
[11] is created (see Table 2).
t5 + t6 -< 0.00500
t2 + t3 + t5 + t6 + t7 -< 0.01400
MACIIINING TOLERANCE GRADES
tt + t3 + t5 + t6 + t7 + t8 -< 0.02900
OPERATION 4 5 6 7 8 9 10 1 12113
t3 + t5 + t9 -< 0.00800 1 1
LAPPING & HONING
t9 -< 0.00500 CYLINDRICAL GRINDING
t6+tT+ta+tg<.O.02000 SURFACE GRINDING
DIAMOND TURNING
t6 + t7 + t9 -< 0.00900 DIAMOND BORING
BROACI lING
t6 + 19 --- 0.00200 REAMING
h -> 0.0004, t2 --> 0.0006, t3 -> 0.0006, t4 -> 0.0007, t5 "I~IRNING
>- 0.0004, t6 - 0.0002, t7 -> 0.0002, ts -> 0.0002, t9 > BORING
MILLING
0.0003 PLANING & SHAPING
DRILLING
Here, the m i n i m u m tolerance grades 7 and 5 are selected
for turning and grinding operations separately from the Fig. 2. Relation of machining processes to tolerance grades.
 Automatic Tolerance Assignment 365

Table I. ANSI Standard Tolerances, (ANSI B4.I-1967, R1979).

Nominal Size, inches Grade

4 5 6 7 8 9 10 11 12 13
Over To Tolerance in thousandths of an inch*

0- 0.12 0.12 0.15 0.25 0.4 0.6 1.0 1.6 2.5 4 6

0.12- 0.24 0.15 0.20 0.3 0.5 0.7 1.2 1.8 3.0 5 7
0.24- 0.40 0,15 0.25 0.4 0.6 0.9 1.4 2.2 3.5 6 9
0.40- 0.71 0.2 0.3 0.4 0.7 1.0 1.6 2.8 4.0 7 10
0.71- 1.19 0.25 0.4 0.5 0.8 1.2 2.0 3.5 5.0 8 12
1.19- 1.97 0.3 0.4 0.6 1.0 1.6 2.5 4.0 6 10 16
1.97- 3.15 0.3 0.5 0.7 1.2 1.8 3.0 4.5 7 12 18
3.15- 4.73 0.4 0.6 0.9 1.4 2.2 3.5 5 9 14 22
4.73- 7.09 0.5 0.7 1.0 1.6 2.5 4.0 6 10 16 25
7.09- 9.85 0.6 0.8 1.2 1.8 2.8 4.5 7 12 18 28
9.85- 12.41 0.6 0.9 1.2 2.0 3.0 5 8 12 20 30
12.41- 15.75 0.7 1.0 1.4 2.2 3.5 6 9 14 22 35
15.75- 19.69 0.8 1.0 1.6 2.5 4 6 10 16 25 40

19.69- 30.09 0.9 1.2 2.0 3 5 8 12 20 30 50

30.09- 41.49 1.0 1.6 2.5 4 6 1 16 25 40 60
41.49- 56.19 1.2 2.0 3 5 8 12 20 30 50 80
56.19- 76.39 1.6 2.5 4 6 10 16 25 40 60 100
76.39- 100.9 2.0 3 5 8 12 20 30 50 80 125
100.9- 131.9 2.5 4 6 10 16 25 40 60 1130 160
131.9- 171.9 3 5 8 12 20 30 50 80 125 200
171.9- 200 4 6 10 16 25 40 60 100 160 250

*All tolerances above heavy line are in accordance with American-British-Canadian (ABC)
agreements.

Gi in Table 2 is the s u m m a t i o n o f ej if working d i m e n s i o n Table 3.

j is a constituent link of the ith dimensional chain, that is:
i I I 12 13 14 15 17 t 8 ql Gi tti

Gi=~aq~
/=l
(11)
1 0 0 1 1 1 0 0 0.0170 0.0017 10.00
2 0 0 0 0 1 0 0 0.0042 0.0004 10.50
and 3 0 1 1 0 1 1 0 0.0132 0.0018 7.33
4 1 0 1 0 1 1 1 0.0282 0.0018 15.67
qi 5 0 0 1 0 1 0 0 0.0068 0.0010 6.80
ui = a i 7 0 0 0 0 0 1 1 0.0180 0.0004 45.00
8 0 0 0 0 0 1 0 0.0070 0.0002 35.00
F r o m Table 2, it can be seen that the m i n i m u m value of
ui(i = 1, 2 . . . . . m ) is Ug. So constraint no. 9 is c o n s i d e r e d
first, and q9 is partitioned to t6 and t9 as: t6 = u9 e6 = 4.00 A new q~ is o b t a i n e d by subtracting t6 a n d / o r t9 from the
x 0.0002 = 0.0008, and t9 = u9 x e9 = 4.00 x 0.0003 = previous q; if working d i m e n s i o n 6 a n d / o r 9 is a constituent
0.0012. Certainly, t6 + t9 = 0.0020 = qg. N o w constraint no. link in the ith constraint.
6 is r e d u n d a n t . By deleting constraints no. 6 and no. 9, N o w the smallest u~(i = 1, 2, 3, 4, 5, 7, 8) is Us, and t3 =
removing columns t6 and t9 from Table 2, and calculating q , u5 e3 = 6.80 x 0.0006 ~ 0.0040, 15 = us x e5 = 6.80 x
Gi and ul, Table 3 is o b t a i n e d . 0.0004 ~ 0.0028. Continuing in this way, o t h e r tolerances are
obtained: t~ = 0.0132, t2 = 0.0048, t4 = 0.007, t7 = 0.0016,
Table 2. ta = 0.0066, and Eg=l ti = 0.042. If an integer set M = {1,
2 . . . . . m} is defined to r e p r e s e n t the constraint indices, an
i tt t2 t3 t4 ts t6 t7 ts t9 q, Gi u, algorithm can be s u m m a r i s e d for the p r o p o r t i o n a l s m o o t h i n g
a p p r o a c h as follows:
1 0 0 1 1 1 0 0 0 0 0.0170 0.0017 10,00
2 0 0 0 0 1 1 0 0 0 0.0050 0.0006 8.33 Step 1: i ~ - - 1 , and M ~ - - { 1 , 2 . . . . . m}
3 0 1 1 0 1 1 1 0 0 0.0140 0.0020 7.00
4 1 0 1 0 1 1 1 1 0 0.0290 0.0020 14.50 Step 2: q_L
5 0 0 1 0 1 0 0 0 1 0.0080 0.0013 6,15 G~ ~-- ~ aq ei, and u i <--
j=! Gi
6 0 0 0 0 0 0 0 0 1 0.0050 0.0003 16.67
7 0 0 0 0 0 1 1 1 1 0.0200 0.0009 22.22 If ui < 1, then stop ( t h e p r o b l e m d o e s not have a
8 0 0 0 0 0 1 1 0 1 0.0090 0.0007 12.86 feasible solution)
9 0 0 0 0 0 1 0 0 1 0.0020 0.0005 4.00 O t h e r w i s e , if i 4: m , t h e n i ~ i + 1, go to Step 2
O t h e r w i s e , go to Step 3
366 P. Ji

Step 3: d ~-- min{ui; i E M}, and let s be the constraint instead of (11). Also, Step 4.1 should be modified as "If a~i
index where the value u is the minimum, i.e. u, = = I, then ti ~ min(vi, max(e/, w i d))" instead of ti ~ eid to
d ensure both lower and upper tolerance bound. In other words,
Algorithm 1 uses the minimum economic tolerance e i as
Step 4 : F o r j ~ - - 1 t o m , do:
weight w#
Step 4.1: If a.~i = 1, then ti ~-- ej d
Step 4.2: ViEM, do:
If aii = 1, then aq ~-- 0, and qg ~-- q~ -
t; 6. Comparison
Step 5: M ~ - - M f q s
The results of four different approaches are listed in Table
Step 6: V i ~ M, do: 4. Comparing the result of the proportional smoothing
Step 6. 1: approach with that of the manual procedure, the tolerances
G~ ~ ~ aii e~
,. j=l of working dimension no. 4, 5, 6, 7 and 8 of the former are
Step 6.2: If Gi = 0, then M ~-- M rl i; Otherwise, tighter than those of the latter. For working dimension no.
qi 4, the tolerances from the approach are between tolerance
ui "~ Gi grade 12 and 13 (the tolerance of the former is reached at
Step 7: If M :# ~, then go to Step 3; Otherwise, stop tolerance grade 12, while that of the latter is greater than 12,
but less than 13). For working dimension no. 5, the two
In the algorithm, Step 6.2 is to find a redundant constraint. tolerances are between tolerance grade 11 and 12 (the
This algorithm is obtained by combining the simplex method tolerance of the former is larger than 11, but less than 12,
and the direct smoothing approach: Step 3 is to obtain a pivot while that of the latter is at 12). So, for these two working
equation, while Step 4 corresponds to the elimination step of dimensions for turning operations, there is no substantial
the simplex method. The big difference is that the proportional difference between the results of the two tolerance assignment
smoothing approach partitions the required value to several approaches. As far as the grinding operations 6, 7 and 8 are
individual variables according to the direct smoothing concerned, the smallest tolerance 0.0016 (-+0.0008) from the
approach, while the simplex method has only one entering proportional smoothing approach is greater than that of
and one leaving variable for each iteration. The algorithm tolerance grade 8. The tolerance grade range for grinding is
complexity of the approach is O(m 2) in the worst case. This from 5 to 8, so there is no difficulty in obtaining those
proportional smoothing approach can also be used for a dimensions. In other words, the propertional smoothing
unilateral tolerance system. approach is applicable in this example.
A general linear programming model for the tolerance Under normal conditions, the result from the proportional
assignment problem in a tolerance chart was presented smoothing approach is acceptable. However, if one of the
previously [8], and can be expressed as follows: tolerances of B/P dimensions is very tight, the tolerances
from the approach will be very close, and may seem
Maximise unacceptable. To make matters worse, condition (9) may not
be met in some cases. In other words, tolerances of working
dimensions are assigned as the minimum economic tolerances,
j=l
but they still exceed the required tolerance of a B/P dimension.
subject to In this situation, even an experienced process engineer finds
it difficult to assign appropriate tolerances. It is recommended
~ t~<-srl (i=n+l,n+2 ..... m) that a statistical tolerance model for tolerance assignment is
j ~ SR~ used in this case.

~, t/<-bpi ( i = 1,2 . . . . . n) Table 4. Result comparison.

j ~ BP~

e,--- tj-- ~ ( j = 1,2 . . . . . m) Seq. Original LP Model Direct Proportional

no. (manual) smoothing smoothing
(LP3)
1 0.0030 0.0274 0.0016 0.0132
where wj > 0 tolerance weight of working dimension j 2 0.0030 0.0126 0.0024 0.0048
vj > ej upper tolerance bound of working dimension 3 0.0030 0.0006 0.0024 0.0040
4 0.0100 0.0160 0.0028 0.0070
J 5 0.0040 0.0004 0.0016 0.0028
The above algorithm can be applied to the general linear 6 0.0010 0.0002 0.0008 0.0008
7 0.0030 0.0002 0.0008 0.0016
programming model LP3. However, some changes are neces- 8 0.0150 0.0002 0.0008 0.0066
sary. First of all, in Step 2 and 6.1, 9 O.OOlO 0.0018 0.0012 0.0012

Sum 0.0430 0.0594 0.0120 0.0420

Gi =~ ai/ w i
j=l
 Automatic Tolerance Assignment 367

7. Statistical Tolerance Model Step5: M~--Mfqs

Step 6: V i E M , do:
From the previous dicussion, we can see that the tolerance m
summation of relevant working dimensions cannot exceed the Step 6.1: Gi <'-- ~ aq ej
required tolerance of a B/P dimension, or the minimum l-I
allowed tolerance of a stock removal in a dimensional chain. Step 6.2: If G~ = 0, then M <---M CI i; Otherwise,
This is guaranteed even in the worst case, or at the extreme qt
condition. However, if the number of constituent links in a ui <--- G---]
dimensional chain is large, or the required tolerance is close,
Step 7: If M ~ ~, then go to Step 3; Otherwise go to Step
the individual tolerances will become very narrow. In this
8
case, if it is possible to assume that each individual tolerance
has a normal distribution with a standard deviation, and the Step 8: For i <--- 1 to m, do:
mean value of a working dimension lies near the expected t i <---- ~ / l i
mean value, and the tolerance deviation does not exceed six
Step 9: Stop
times the standard deviation, a statistical tolerance model can
be applied for tolerance assignment. With the wide application In this algorithm, Step 1 is converting the required tolerances
of numerically controlled machine tools, the above assumptions and minimum economic tolerances to their squares. Step 8 is"
are reasonable. Suppose that there are k constituent links in converting the resultant tolerance square back to the original
a dimensional chain, the equation for the statistical tolerance tolerance.
model is as follows: Applying the above algorithm to the previous numerical
example, the tolerances for all working dimensions are: fi
t0 0.0244, t2 ~ 0.0109, t3 ~ 0.0065, t4 m 0.0150, ts ~ 0.0043, t6
0.0012, t7 ~- 0.0036, t8 = 0.0122, and t9 ~ 0.0016. Comparing
where to is the required tolerance, and ti (i = 1, 2 . . . . . k) is the result with the previous example, each individual tolerance
the individual tolerance of each constituent link in the is widened in this statistical model.
dimensional chain. Thus a nonlinear programming model can
be formulated as follows:
Maximise 8. Conclusion

This paper has discussed the tolerance assignment problem

j=t in a tolerance chart. A linear programming model for the
subject to problem was analysed. Because the solution of the LP model
may not be applicable, a proportional smoothing approach
F/ <-S~ ( i = n + l , n + 2 ..... m) was described, which combines the simplex method and the
] E SR i
direct smoothing approach. Like the LP model, this approach
can still systematically solve the tolerance assignment problem
~/<<-bp2i ( / = 1,2 . . . . . n)
in a tolerance chart with consideration of both B/P dimension
j E OPi
requirements and process capabilities. The result from the
ti>-e j ( j = 1,2 . . . . m) approach is acceptable and reasonable, compared with that
This nonlinear programming model can be solved as a linear of the LP model and the manual procedure. The proportional
programming formulation if each ~ is considered as a smoothing approach was also extended to the statistical
independent variable. For the same reasons as for the linear tolerance model, which widens individual tolerance range.
programming model LP1 or LP2, an algorithm similar to However, the application of the statistical tolerance model
the previous algorithm is provided to solve the nonlinear can be used only under certain conditions.
programming model as follows:
Step 1 : F o r i < - - - 1 t o m , do:
References
qi~--q~,ei~--~
Step 2: M * - - { 1 , 2 . . . . . m} and f o r i ~ - - 1 to m, do: I. R. S. Ahluwalia, and A. V. Karolin, "CATC - a computer aided
m
q, tolerance control system", Journal of Manufacturing Systems,
Gi ~'- ~ aq ei, and ui ~-- 3(2), 153-160, 1986.
/=1 Gi 2. P. Ji, Computer Aided Operational Dimensions Calculation,
Step 3: d ~-- min{ul; i E M}, and obtain s(u, = d) Master thesis, Beijing Institute of Aeronautics and Astronautics,
P. R. China, 1984.
Step 4: For j * - - 1 t o m , do: 3. T. Xiaoqing and B. J. Davies, "Computer aided dimensional
Step. 4.1: I r a , i = 1, t h e n t i ~ - - e f l planning", International Journal of Production Research, 26(2),
pp. 283-297, 1988.
Step 4.2: " v ' i E M, do: 4. D. Fainguelernt, R. Weill and P. Bourdet, "Computer aided
If aq = 1, then aq *--- O, and q,. *-- q~ - tolerancing and dimensioning in process planning", CIRP Annals,
tj 35(1), 1986.
368 P. Ji

5. D. Duret, "Simulation de gamme d'usinage", L'lng#nieur et le a tolerance chart", International Journal of Production Research,
Technicien de I'enseignment Technique, 220, 230, March and 31(3), pp. 739-751, 1993.
April 1991. 9. O. R. Wade, 1983, "Tolerance control", in T. J. Drozda and C.
6. P. Ji, "A tree approach for tolerance charting", International Wick (ed.) Tool and Manufacturing Engineers Handbook, vol.
Journal of Production Research, 31(5), pp. 1023-1033, 1993. 1, Machining, Dearborn, Michigan: Society of Manufacturing
7. K. Whybrew, G. A. Britton, D. F. Robinson and Y. Sersuti- Engineers, pp. 2-1-2-60, 1983.
anuwat, "A graph-theoretic approach to tolerance charting", 10. E. Oberg, F. D. Johns and H. L. Horton, Machinery's Handbook,
International Journal of Advanced Manufacturing Technology, 23rd edn, Industrial Press, Inc., New York, 1988.
5(2), pp. 175-183, 1990. 11. H. A. Taha, Operations Research: an Introduction, 2nd r
8. P. Ji, "A linear programming model for tolerance assignment in Macmillan, New York, 1976.