Tolerance Allocation Methods for Designers

ADCATS Report No. 99-6

Kenneth W. Chase
Department of Mechanical Engineering
Brigham Young University

1999

ABSTRACT

Tolerance allocation is a design tool. It provides a rational basis for assigning tolerances to
dimensions. Several algorithms are described in this paper for performing tolerance allocation,
which is defined as the re-distribution of the “tolerance budget” within an assembly to reduce
over-all cost of production, while meeting target levels for quality.

The task of placing +/- tolerances on each dimension of a CAD model or set of engineering
drawings may seem menial and of little consequence. However, it can have enormous impact on
cost and quality. On the engineering design side, it affects the fit and function of the final
product, which can cause poor performance and dissatisfied customers. On the manufacturing
side, tolerance requirements determine the selection of machines, tooling and fixtures; operator
skill levels and setup costs; inspection precision and gaging; and scrap and rework. In short, it
affects nearly every aspect of the product life cycle.

Using the new CAD-based tolerance analysis tools, designers can perform variation analysis on
the CAD model, before parts are made or tooling purchased. They can determine where
tolerance controls are needed and how tight the limits must be set, to assist in process planning
and tool design. Using the same CAD model, production personnel can use tolerance tools to
determine whether marginal or out-of-spec parts can still be used. Tolerance analysis tools
provide a quantitative basis for design for manufacture decisions, resulting in shorter product
development time and increased quality. There is probably no other design improvement effort
which can yield greater benefits for less cost than the careful analysis and assignment of
tolerances.
Tolerance Allocation Methods for Designers
Kenneth W. Chase
Mechanical Engineering Department
Brigham Young University

1.0 Improving Quality by Tolerancing Techniques

There are several advanced tolerancing techniques available to a designer to improve quality
levels in assemblies. They are primarily fine-tuning operations which can be applied to an
assembly distribution to reduce the number of rejected assemblies, sometimes dramatically. The
three methods to be presented here are:
1 Centering the Mean of the Assembly Distribution
2 Upper or Lower Limit Justification of the Mean
3 Reducing the Spread of the Distribution

When bringing a manufacturing process for a new product on line, small lots are produced to
qualify the tooling. Parts are assembled, critical assembly features are measured and the mean
and standard deviation of the resulting distributions are calculated. It is often found that the
mean of the assembly distribution is not symmetrically distributed with respect to the upper and
lower design specification limits (USL and LSL), resulting in large numbers of rejects at one or
the other design limit. This can be caused by tooling bias, setup error, shrinkage of molded or
cast parts, or other factors. If the errors are systematic, adjustments must be made to the tooling
or processes to center the distribution. By centering the distribution, the total number of rejects
can be minimized.

If only one limit is critical, as with a minimum clearance specification, the mean may be shifted
away from the constraining limit until the desired quality level is reached. These operations are
depicted in Fig. 1.

LSL USL

Centered

Original Distribution x

USL LSL

LSL USL
USL Justified

x x +Zasm σ

USL

LSL LSL Justified

-Zasm σ x

Figure 1. Effects of mean shift operations

1.1 Centering the Mean of the Assembly Distribution

Centering the mean puts the greatest distance between the tolerance limits and the mean, thus
minimizing the number of rejects for a given assembly tolerance distribution. By adjusting one
or more component nominals, the assembly mean may be moved to the center, halfway between
the USL and LSL. Clearly, the centered distribution will have fewer rejects.

Note that adjustments to the mean do not involve tightening any tolerances, nor do tightening
tolerances affect the mean. To tighten a tolerance may require switching to a more expensive
process or creation of special fixtures, in which case adjusting the nominal dimension of a
component may be less costly.

Deciding which component nominals to change is a design decision. Some components may be
vendor-supplied and not allowed to change. Others may be too costly to change due to tooling
modifications or other requirements. Some may simply require a change in the program of an
NC machine used in its production. Of course, if the assembly is in production, the fewer parts
you have to change, the better.

If more than one nominal is to be changed, a set of weight factors may be used for allocation of
component nominals. The designer sets the value of the weight factors corresponding to those
components he wishes to change. The relative weights determine how much change is assigned
to each component.

1.2 Upper or Lower Limit Justification of the Mean

Small changes may also be made to individual nominal dimensions to shift the nominal spec
towards either the upper or lower spec limit. This may be desirable if there is only one spec limit
or if one of the two spec limits is more important than the other. As with the centering
allocation, the distribution is simply shifted without changing its spread. For example, a
minimum clearance (LSL) may be critical to an assembly, but the maximum clearance may not
be important.

In USL justification, the center of the distribution is shifted such that the +Zasmσ limit matches
with the USL, where σ is the standard deviation of the assembly variation and Zasm is the
desired quality level expressed in standard deviations. Similarly, in LSL justification, the
distribution is shifted such that the -Zasmσ limit matches with the LSL.

1.3 Variance Adjustment

If the mean has been centered and the reject rate is still too high because the spread of the
distribution is too broad, it will be necessary to tighten tolerances on one or more assembly
component dimensions to reduce the standard deviation of the assembly, as shown in Fig. 2.

Original Distribution Reduced Variation

LSL USL LSL USL

x x
 Figure 2. Effects of reduced variation

All three nominal allocation methods may be used in conjunction with any tolerance allocation
method to simultaneously shift the mean and change the spread of the distribution.

2.0 Tolerance Analysis vs. Tolerance Allocation

The analytical modeling of assemblies provides a quantitative basis for the evaluation of design
variations and specification of tolerances. An important distinction in tolerance specification is
that engineers are more commonly faced with the problem of tolerance allocation rather than
tolerance analysis.

The difference between these two problems is illustrated in Fig. 3. In tolerance analysis the
component tolerances are all known or specified and the resulting assembly variation is
calculated. In tolerance allocation, on the other hand, the assembly tolerance is known from
design requirements, whereas the magnitude of the component tolerances to meet these
requirements are unknown. The available assembly tolerance must be distributed or allocated
among the components in some rational way. The influence of the tolerance accumulation model
and the allocation rule chosen by the designer on the resulting tolerance allocation will be
demonstrated.

Tolerance Analysis Tolerance Allocation

Component Component
Tolerances Tolerances

Assembly
Assembly Tolerance
Tolerance
Allocation
Assembly Scheme
Function

Acceptance Acceptance
Fraction Fraction

Figure 3. Tolerance Analysis vs. Tolerance Allocation.

Another difference in the two problems is the yield or acceptance fraction of the assembly
process. The assembly yield is the quality level. It is the percent of assemblies which meet the
engineering tolerance requirements. It may be expressed as the percent of acceptable assemblies
or the percent rejects. For high quality levels, the rejects may be expressed in parts-per-million
(ppm), that is, the number of rejects per million assemblies .

In tolerance analysis the assembly yield is unknown. It is calculated by summing the component
tolerances to determine the assembly variation, then applying the upper and lower spec limits to
the calculated assembly distribution. In tolerance allocation, on the other hand, the assembly
yield is specified as a design requirement. The component tolerances must then be set to assure
that the resulting assembly yield meets the spec.
The rational allocation of component tolerances requires the establishment of a rule for
distributing the assembly tolerance among the components. The following sections present
several examples of useful rules.
2.1 Allocation By Proportional Scaling

The designer begins by assigning reasonable component tolerances based on process or design
guidelines. The component tolerances are summed to see if they meet the specified assembly
tolerance. If not, the component tolerances are scaled by a constant proportionality factor. In
this way the relative magnitudes of the component tolerances are preserved.

This algorithm is demonstrated graphically in Fig. 4. for an assembly tolerance Tasm, which is
the sum of two component tolerances, T1 and T2. The straight line labeled as the Worst Case
Limit is the locus of all possible combinations of T1 and T2 which, added linearly, equal Tasm.
The ellipse labeled Statistical Limit is the locus of root sum squares of T1 and T2 which equal
Tasm.

Suppose the designer chooses initial values for T1 and T2 based on typical process tolerances for
the two component parts. This combination is the point labeled Original Tolerances in the
figure. By drawing a line from the origin through this point, then extending it until it intersects
the Worst Case Limit, the largest possible tolerances for T1 and T2 are obtained, which satisfy
the worst case condition and which still have the original ratio of T1 to T2. Extending this line
further, until it intersects the Statistical Limit curve, new values for T1 and T2 are obtained which
satisfy the assembly tolerance limit by root sum squares.

Although Fig. 4 illustrates 1-D tolerance accumulation models, the algorithm may also be
applied to 2-D or 3-D tolerance accumulation by pre-multiplying each component tolerance by
its tolerance sensitivity.

Worst Case Limit

1.0 T asm = T1 + T2
Statistical Limit
T asm = T12 + T22

T2
Tasm
Allocated
Tolerances by
Proportional
0.5 Scaling
Original
Tolerances

0
0.5 T1 1.0
T asm

Figure 4. Graphical interpretation of tolerance allocation by proportional scaling.

2.2 1-D Example: Worst Case Tolerance Allocation by Proportional Scaling

The following example is based on the shaft and housing assembly shown in Fig. 5. Two
bearing sleeves maintain the spacing of the bearings to match that of the shaft. Accumulation of
variation in the assembly results in variation in the end clearance. Positive clearance is required.
CLEARANCE
-E

-G +F +D
-A Ball
Bearing

-C
Retaining +B
Ring Shaft

Bearing
Sleeve

Housing

Figure 5 Shaft and housing assembly.

Initial tolerances for parts B, D, E, and F are selected from tolerance guidelines such as those
illustrated in Fig. 6. The bar chart shows the typical range of tolerance for several common
processes. The numerical values appear in the table above the bar chart. Each row of the
numerical table corresponds to different nominal size. For example, a turned part having a
nominal dimension of 0.750 in. can be produced to a tolerance ranging from ±.001 to ±.006 in.,
depending on the number of passes, rigidity of the machine and fixtures, etc. Tolerances are
chosen initially from the middle of the range for each dimension and process, then adjusted to
match the design limits and reduce production costs.

RANGE OF SIZES
TOLERANCES ±
FROM THROUGH

0.000 0.599 0.00015 0.0002 0.0003 0.0005 0.0008 0.0012 0.002 0.003 0.005
0.600 0.999 0.00015 0.00025 0.0004 0.0006 0.001 0.0015 0.0025 0.004 0.006
1.000 1.499 0.0002 0.0003 0.0005 0.0008 0.0012 0.002 0.003 0.005 0.008
1.500 2.799 0.00025 0.0004 0.0006 0.001 0.0015 0.0025 0.004 0.006 0.010
2.800 4.499 0.0003 0.0005 0.0008 0.0012 0.002 0.003 0.005 0.008 0.012
4.500 7.799 0.0004 0.0006 0.001 0.0015 0.0025 0.004 0.006 0.010 0.015
7.800 13.599 0.0005 0.0008 0.0012 0.002 0.003 0.005 0.008 0.012 0.020
13.600 20.999 0.0006 0.001 0.0015 0.0025 0.004 0.006 0.010 0.015 0.025

LAPPING & HONING

DIAMOND TURNING
& GRINDING

BROACHING
REAMING

TURNING, BORING, SLOTTING,

PLANING, & SHAPING
MILLING
DRILLING

Figure 6. Tolerance range of machining processes [Trucks, 1987].

The box below shows the problem setup. The retaining ring (A) and the two bearings (C and G)
supporting the shaft are vendor-supplied, hence their tolerances are fixed and must not be altered
by the allocation process. Initial values for B, D, E and F are shown as selected from the table,
assuming a mid-range tolerance for turned parts. The critical clearance is the shaft end-play,
which is determined by tolerance accumulation in the assembly. The vector diagram overlaid on
the figure is the assembly loop that models the end-play. The average clearance is the vector
sum of the average part dimensions in the loop:

Table 1 Initial Tolerance Specifications

Required Clearance = .020±.015

Average Clearance = -A + B - C + D - E + F - G
= -.0505 + 8.000 - .5093 + .400 - 7.711 + .400 - .5093
= .020

Dimension A B C D E F G
Nominals .0505 8.000 .5093 .400 7.711 .400 .5093
Tolerances(±)
Design .008 .002 .006 .002
Fixed .0015 .0025 .0025

The clearance tolerance is obtained by summing the component tolerances by worst case:
TSUM = + TA + TB + TC + TD + TE + TF + TG
= + .0015 + .008 + .0025 + .002 + .006 + .002 + .0025
= .0245 (too large)
Now, solving for the proportionality factor:
TASM = .015 = .0015 +.0025 +.0025 + P (.008 + .002 + .006 + .002)
P = .47222

Note that the three fixed tolerances were subtracted from the assembly tolerance before
computing the scale factor. Thus, only the four design tolerances are re-allocated:

TB = .47222 (.008) = .00378 TE = .47222 (.006) = .00283

TD = .47222 (.002) = .00094 TF = .47222 (.002) = .00094

Each of the design tolerances has been scaled down to meet assembly requirements as shown in
Fig. 7. This procedure could also be followed assuming a statistical sum for the assembly
tolerance, in which case the tolerances would be scaled up. Results are summarized in Table 2.
 STATISTICAL

ORIGINAL
TOLERANCES

WORST
DESIGN LIMIT
TOLERANCES

FIXED
TOLERANCES
SCALE FACTOR 1.40 1.0 .472

Figure 7 Tolerance allocation by proportional scaling.

3.0 Allocation by Weight Factors

A more versatile method of assigning tolerances is by means of weight factors. Using this
algorithm, the designer assigns weight factors to each tolerance in the chain and the system
distributes a corresponding fraction of the tolerance pool to each component. A larger weight
factor for a given component means a larger fraction of the tolerance pool will be allocated to it.
In this way, more tolerance can be given to those dimensions which are the more costly or
difficult to hold, thus improving the producibility of the design.
Fig. 8 illustrates this algorithm graphically for a two component assembly. The original values
for component tolerances T1 and T2 are selected from process considerations and are represented
as a point in the figure, as before. The tolerances are scaled, similar to proportional scaling, only
the scale factor is weighted for each component tolerance so the greater scale factors yield the
least reduction in tolerance.
 Statistical Limit
Worst Case Limit
T asm = T1 + T2 T asm = T12 + T22
1.0
Original
Tolerances
Proportional
Scaling (T ,T )
1 2
W2 T2
W1 T1
W1 T1
Scaled
0.5 Tolerances

T2
Tasm

W2 T2

0
0.5 T1 1.0
T asm

Figure 8. Graphical interpretation of tolerance allocation by weight factors.

Again, although the figure illustrates 1-D tolerance accumulation models, the algorithm may be
applied equally well to 2-D and 3-D stacks. It may also be applied to worst case, statistical, or
six sigma tolerance sums.

Note that any components which are vendor-supplied, or subject to other design considerations,
can be excluded from the allocation process by declaring them to be "fixed".

4.0 1-D Example: Worst Case (WC) Allocation by Weight Factors

The shaft and housing assembly of Section 2.0 will be revisited, using weight factors to allocate
tolerances. The three tolerances A, C and G are fixed. Design tolerances B, D, E and F are
assigned weight factors of 10, 20, 10 and 20, respectively. Weights were determined on the basis
of machining difficulty.

The clearance tolerance is obtained by summing the component tolerances by worst limits:
TSUM = + TA + TB + TC + TD + TE + TF + TG
= + .0015 + .008 + .0025 + .002 + .006 + .002 + .0025
= .0245 (too large)

Solving for the proportionality factor with weight factors:

Ti’ = PWiTi
TASM = .015 = .0015 +.0025 +.0025 +
P [(10/60)(.008) + (20/60)(.002) + (10/60)(.006) + (20/60)(.002)]
P = 2.31818
Note that only the four design tolerances are re-allocated:
TB = 2.31818(10/60)(.008) = .00309 TE = 2.31818(10/60)(.006) = .00232
TD = 2.31818(20/60)(.002) = .00155 TF = 2.31818(20/60)(.002) = .00155

5.0 1-D Example: Statistical (RSS) Allocation by Weight Factors

The clearance tolerance is obtained by summing the component tolerances by root sum squares:
TSUM = TA2 + TB2 + TC2 + TD2 + TE2 + TF2 + TG2
= .00152 + .0082 + .00252 + .0022 + .0062 + .0022 + .00252
= .01108 (too small)

Solving for the proportionality factor with weight factors:

Ti’ = PWiTi
TASM = TA2+ P2WB2TB2+ TC2+ P2WD2TD2+ P2WE2TE2+ P2WF2TF2+ TG2
P = 7.57238

Note that the three fixed tolerances are not re-allocated:

TB = (7.57238)(10/60)(.008) = .01010 TE = (7.57238)(10/60)(.006) = .00757
TD = (7.57238)(20/60)(.002) = .00505 TF = (7.57238)(20/60)(.002) = .00505

All of the preceeding examples of allocation are compared in Table 2. A graphical comparison is
shown in Fig. 9.

Table 2 Comparison of Allocation Methods

Proportional Weight Factors
Original Worst Stat Weight Worst Stat
Dim Tolerance Case ±3σ Factor Case ±3σ
A .0015* .0015 .0015 0 .0015 .0015
B .008 .00378 .01116 10 .00309 .0101
C .0025* .0025 .0025 0 .0025 .0025
D .002 .00094 .00279 20 .00155 .00505
E .006 .00283 .00837 10 .00232 .00757
F .002 .00094 .00279 20 .00155 .00505
G .0025* .0025 .0025 0 .0025 .0025
Assembly Tolerance .0150 .0150 .0150 .0150
Scale Factor (P) .47222 1.39526 2.31818 7.57238
*Fixed tolerances
 Allocation Results

Original Tol

Prop Scale: WC

B
Prop Scale: RSS D
E
Wt Factor: WC F

Wt Factor: RSS

0.000 0.002 0.004 0.006 0.008 0.010 0.012

Tolerance

Figure 9. Tolerance allocation by proportional scaling and weight factors

6.0 Tolerance Allocation Using Least Cost Optimization

A promising method of tolerance allocation uses optimization techniques to assign component

tolerances such that the cost of production of an assembly is minimized. This is accomplished by
defining a cost-vs.-tolerance curve for each component part in the assembly. The optimization
algorithm varies the tolerance for each component and searches systematically for the
combination of tolerances which minimizes the cost.

Fig. 10 illustrates the concept simply for a three component assembly. Three cost-vs.-tolerance
curves are shown. Three tolerances (T1, T2, T3 ) are initially selected. The corresponding cost of
production is C1 + C2 + C3. The optimization algorithm tries to increase the tolerances to reduce
cost; however, the specified assembly tolerance limits the tolerance size. If tolerance T1 is
increased, then tolerance T2 or T3 must decrease to keep from violating the assembly tolerance
constraint. It is difficult to tell by inspection which combination will be optimum, but you can
see from the figure that a decrease in T2 results in a significant decrease in cost, while a
correponding decrease in T3 results in smaller increase in cost. In this manner, one could
manually adjust tolerances until no further cost reduction is achieved. The optimization
algorithm is designed to find it with a minimum of iteration. Note that the values of the set of
optimum tolerances will be different when the tolerances are summed statistically than when
they are summed by worst case.

Cost

Tolerance

Total Cost:

[Worst Case]

[Statistical]

Figure 10. Optimal tolerance allocation for minimum cost.

A necessary factor in optimum tolerance allocation is the specification of cost-vs.-tolerance

functions. Several algebraic functions have been proposed, as summarized in Table 3. The
Reciprocal Power function: C = A + B/tolk includes the Reciprocal and Reciprocal Squared rules
for integer powers of k. The constant coefficient A represents fixed costs. It may include setup
cost, tooling, material, prior operations, etc. The B term determines the cost of producing a
single component dimension to a specificed tolerance and includes the charge rate of the
machine. Costs are calculated on a per part basis. When tighter tolerances are called for, speeds
and feeds may be reduced and the number of passes increased, requiring more time and higher
costs. The exponent k describes how sensitive the process cost is to changes in tolerance
specifications.

Table 3. Proposed Cost-of-Tolerance Models

Cost Model Function Author Ref
Reciprocal Squared A + B/tol2 Spotts [Spotts 1973]
Reciprocal A + B/tol Chase&Greenwood [Chase 1988]
Reciprocal Power A + B/tol k Chase et. al. [Chase 1989]
Exponential A e–B(tol) Speckhart [Speckhart 1972]

Little has been done to verify the form of these curves. Manufacturing cost data are not
published since they are so site-dependent. Even companies using the same machines would
have different costs for labor, materials, tooling and overhead.

A study of cost vs. tolerance was made for the metal removal processes over the full range of
nominal dimensions. This data has been curve fit to obtain empirical functions. The form was
found to follow the reciprocal power law. The results are presented in the appendix to this
chapter. The original cost study is decades old and may not apply to modern N/C machines.

A closed-form solution for the least-cost component tolerances was developed by Spotts [1973].
He used the method of Lagrange Multipliers, assuming a cost function of the form C=A+B/tol2.
Chase extended this to cost functions of the form C=A+B/tolk as follows [Chase et al. 1989]:

• •
•Ti (Cost function) + λ •Ti (Constraint) = 0 (i = 1, . . n)
• kj)) + λ
• 2 2
•Ti (• (A j + B j / Tj •Ti (•Tj —Tasm ) = 0 (i = 1, . . n)
kiBi
λ= (i = 1, . . n)
2Ti(ki+2)

Eliminating λ by expressing it in terms of T1 (arbitrarily selected):

 kiBi 
Ti = k B  1/(ki+2) ·T1(k1+2)/(ki+2) (1)
 1 1

Substituting for each of the Ti in the assembly tolerance sum:

 kiBi 
T2ASM = T12 + • k B  2/(ki+2)·T1 1+2)/(ki+2)
2(k
(2)
 1 1
The only unknown in Eq (2) is T1. One only needs to iterate the value of T1 until both sides of
Eq (2) are equal to obtain the minimum cost tolerances. A similar derivation based on a worst
case assembly tolerance sum yields:
 kiBi  1/(ki+1) (k1+1)/(ki+1)
TASM = T1 + • k B  ·T1 (3)
 1 1
A graphical interpretation of this method is shown in Fig. 11. Note that as the method of
Lagrange Multipliers assumes, the constant cost curve is tangent to the tolerance limit curve for
the minimum cost tolerance value.

Statistical
Minimum Cost Statistical Limit
1.0 T asm = T12 + T22

$14
T2
T asm COST
CURVES $15
Worst Case
0.5 Minimum Cost $16
direction of
decreasing cost $17

Worst Case Limit $18

T asm = T1 + T2

0
0.5 1.0
T1
T
asm

Figure 11. Graphical interpretation of minimum cost tolerance allocation.

Numerical results for the 1-D shaft and housing example problem are shown in Table 4:
Table 4. Minimum Cost Tolerance Allocation
Tolerance Cost Data Allocated Tolerances
Dimension Setup Coefficient Exponent Original Worst Stat.
A B k Tolerance Case ±3σ
A * * .0015* .0015* .0015*
B $1.00 0.15997 -0.43899 .008 .00254 .0081
C * * .0025* .0025* .0025*
D 1.00 0.07202 -0.46823 .002 .001736 .00637
E 1.00 0.12576 -0.46537 .006 .002498 .00792
F 1.00 0.07202 -0.46823 .002 .001736 .00637
G * * .0025* .0025* .0025*
Assembly Variation .0245(WC) .0150(WC) .0150(RSS)
.0111(RSS)
Assembly Cost $9.34 $11.07 $8.06
Acceptance Fraction 1.000 .9973
"True Cost" $11.07 $8.08
*Fixed tolerances

The Setup Cost is coefficient A in the cost function. Setup cost does not affect the optimization.
For this example, the setup costs were all chosen as equal, so they would not mask the affect of
the tolerance allocation. In this case, they merely added $4.00 to the assembly cost for each case.

Parts A, C and G are vendor-supplied. Since their tolerances are fixed, their cost cannot be
changed by re-allocation, so no cost data is included in the table.

Note that in this example the assembly cost increased when worst case allocation was performed.
This is due to the fact that the original tolerances, when summed by worst case, gives an
assembly variation of 0.0245 in. This exceeded the specified assembly tolerance limit of 0.015
in. Thus, the component tolerances had to be tightened, driving the cost up. When summed
statistically, however, the assembly variation was only .00111 in., which was less than the spec
limit, allowing the allocation algorithm to increase the component tolerances, driving the cost
down. A graphical comparison is shown in Fig. 12. It is clear from the graph that tolerances for
B and E were reduced the most in the Worst Case model, while D and F were increased more in
the Statistical model.

Min Cost Allocation Results

Original $9.34 B
D
Min Cost: $11.07 E
F

Min Cost: $8.06

0.000 0.002 0.004 0.006 0.008 0.010

Tolerance

Figure 12. Comparison of minimum cost allocation results

The advantages of the Lagrange multiplier method are:

1) It eliminates the need for multiple-parameter iterative solutions
2) It can handle either worst case or statistical assembly models
3) It allows alternative cost-tolerance models

The limitations are:

1) Tolerance limits cannot be imposed on the processes. Most processes are only
capable of a specified range of tolerance. The designer must check the resulting
component tolerances to make sure they are within the range of the process.
2) It cannot readily treat the problem of simultaneously optimizing an assembly with
interdependent design specifications. That is, when an assembly has more than one
design specification, with common component dimensions contributing to each spec,
some iteration is required to find a set of shared tolerances which satisfies each of the
engineering requirements.

Problems which exhibit these characteristics may be optimized using nonlinear programming
techniques or by optimizing tolerances for one assembly spec at a time, then choosing the lowest
set of component tolerance values required to satisfy all assembly specs simultaneously.

7.0 True Cost and Optimum Acceptance Fraction

The "True Cost" in Table 5 is defined as the total cost of an assembly divided by the acceptance
fraction or yield. Thus, the "true cost" is adjusted to include a share of the cost of the rejected
assemblies. It does not include, however, any parts which might be saved by re-work or the cost
of rejecting individual component parts.
An interesting exercise is to calculate the optimum acceptance fraction, that is, the rejection rate
which would result in the minimum True Cost. This requires an iterative solution. For the
example problem, the results are shown in Table 5:
Table 5. Minimum True Cost
Cost Model ΣA Zasm Opt. Accept Fract True Cost
A + B/tolk $4.00 2.03 .9576 $7.67
A + B/tolk $8.00 2.25 .9756 $11.82

The interpretation of these results is that loosening up the tolerances will save money on
production costs, but will increase the cost of rejects. By iterating on the acceptance fraction, it
is possible to find the value which minimizes the combined cost of production and rejects. Note,
however, that the setup costs were set very low. If setup costs were doubled, as shown in the
second row of the table, the cost of rejects would be higher, requiring a higher acceptance level.

In the very probable case where individual process cost vs. tolerance curves are not available, an
optimum acceptance fraction for the assembly could be based instead on more-available cost per
reject data. This optimum acceptance fraction could then be used in conjunction with allocation
by proportional scaling or weight factors to provide a meaningful cost-related alternative to
allocation by least cost optimization.

8.0 2-D and 3-D Tolerance Allocation

Tolerance allocation may be applied to 2-D and 3-D assemblies as readily as 1-D. The only
difference is that each component tolerance must be multiplied by its tolerance sensitivity,
derived from the geometry by small variations. The proportionality factors, weight factors, cost
factors are still obtained as described above, with sensitivities inserted appropriately.

8.1 One-way Clutch Assembly

The application of tolerance allocation to a 2-D assembly will be demonstrated on the one-way
clutch assembly shown in Fig. 13. The clutch consists of four different parts: a hub, a ring, four
rollers, and four springs. Only a quarter section is shown because of symmetry. During
operation, the springs push the rollers into the wedge-shaped space between the ring and the hub.
If the hub is turned counter-clockwise, the rollers bind, causing the ring to turn with the hub.
When the hub is turned clockwise, the rollers slip, so torque is not transmitted to the ring. A
common application for the clutch is a lawn mower starter [Fortini 1967].
 φ
Ring
Spring

c
b c Roller

e
φ
Hub
a 2
2 Vector Loop

Figure 13 Clutch Assembly with Vector Loop

The contact angle φ, between the roller and the ring, is critical to the performance of the clutch.
The angle φ and variable b, the location of contact between the roller and the hub, are both
dependent assembly variables. The magnitude of φ and b will vary from one assembly to the next
due to the variations of the component dimensions a, c, and e. Dimension a is the width of the
hub; c and e/2 are the radii of the roller and ring, respectively. A complex assembly function
determines how much each dimension contributes to the variation of angle φ. The nominal
contact angle, when all of the independent variables are at their mean values, is 7.0 degrees. For
proper performance, the angle must not vary more than ±1.0 degree from nominal. These are the
engineering design limits.

The objective of variation analysis for the clutch assembly is to determine the variation of the
contact angle relative to the design limits. Table 6 below shows the nominal value and tolerance
for the three independent dimensions which contribute to tolerance stackup in the assembly. Each
of the independent variables is assumed to be statistically independent (not correlated with each
other) and a normally distributed random variable. The tolerances are assumed to be ±3σ.
Table 6: Independent Dimensions for the Clutch Assembly
Dimension Nominal Tolerance
Hub width - a 2.1768 in 0.004 in
Roller radius - c 0.450 in 0.0004 in
Ring diameter - e 4.000 in 0.0008 in

Vector Loop Model and Assembly Function for the Clutch

The vector loop method [Chase 1995], uses the assembly drawing as the starting point. Vectors
are drawn from part-to-part in the assembly, passing through the points of contact. The vectors
represent the independent and dependent dimensions which contribute to tolerance stackup in the
assembly. Fig. 13, above, shows the resulting vector loop for a quarter section of the clutch
assembly.

The vectors pass through the points of contact between the three parts in the assembly. Since the
roller is tangent to the ring, both the roller radius c and the ring radius e are colinear. Once the
vector loop is defined, the implicit equations for the assembly can easily be extracted. Eq. 4
shows the set of scalar equations for the clutch assembly derived from the vector loop. hx and hy
are the sum of vector components in the x and y directions. A third equation, hθ, is the sum of
relative angles between consecutive vectors, but it vanishes identically.
 hx = 0 = b + c sin(φ) - e sin(φ) (4)
hy = 0 = a + c + c cos(φ) - e cos(φ)

Equations 16.4 may be solved for φ explicitely:

(
a+c
φ = cos-1 e – c ) (5)

The sensitivity matrix [S] can be calculated from equation 16.5 by differentiation or by finite
difference:
 ∂f ∂f ∂f 
[S] =  ∂a ∂c ∂e  = −2.6469 −10.5483 2.6272 (6)
 ∂b ∂b ∂b  −103.43 −440.69 104.21
 ∂a ∂c ∂e 
The top row of [S] are the tolerance sensitivities for δφ. Assembly variations accumulate or
stackup statistically by root-sum-squares:

δφ = Σ((Sij δxj)2) (7)

= (S11 δa)2+(S12 δc)2+(S13 δe)2
= (-2.6469•0.004)2+(-10.5483•0.0004)2+(2.6272•0.0008)2
= 0.01159 radians = 0.664 degrees
where δφ is the predicted 3σ variation, δxj is the set of 3σ component variations.
By worst case:
δφ = Σ|Sij| δxj (8)
= |S11| δa+|S12| δc+|S13| δe
= 2.6469•0.004+10.5483•0.0004+2.6272•0.0008
= 0.01691 radians = 0.9688 degrees
where δφ is the predicted extreme variation.

8.2 Allocation by Scaling, Weight Factors

Once you have RSS and worst case expressions for the predicted variation δφ, you may begin
applying various allocation algorithms to search for a better set of design tolerances. As we try
various combinations, we must be careful not to exceed the tolerance range of the selected
processes. Table 7 shows the selected processes for dimensions a, c and e and the max and min
tolerances obtainable by each, as extracted from Fig. 6 for the corresponding nominal size.

Table 7. Process Tolerance Limits for the Clutch Assembly

Part Dim Proc Nom(in) Sens Min Tol Max Tol
Hub a Mill 2.1768 -2.6469 0.0025 0.006
Roller c Lap 0.9000 -10.548 0.00025 0.00045
Ring e Grind 4.0000 2.62721 0.0005 0.0012

Proportional Scaling by Worst Case:

Since the rollers are vendor-supplied, only tolerances on dimensions a and e may be altered. The
proportionality factor P is applied to δa and δe, while δφ is set to the maximum tolerance of
±0.017453 radians (±1• ).
δφ = Σ|Sij| δxj (9)
0.017453 = |S11| P δa+|S12| δc+|S13| P δe
0.017453 = 2.6469•P•0.004+10.5483•0.0004+2.6272•P•0.0008
Solving for P:
P = 1.0429
δa = 1.0429•0.004 = 0.00417 in
δe = 1.0429•0.0008 = 0.00083 in

Proportional Scaling by Root-Sum-Squares:

δφ = Σ((Sij δxj)2) (10)

0.017453 = (S11 P δa)2+(S12 δc)2+(S13 P δe)2
0.017453 = (-2.6469•P•0.004)2+(-10.5483•0.0004)2+(2.6272•P•0.0008)2
Solving for P:
P = 1.56893
δa = 1.56893•0.004 = 0.00628 in
δe = 1.56893•0.0008 = 0.00126 in
Both of these new tolerances exceed the process limits for their respective processes, but by less
than 0.001in each. You could round them off to 0.006 and 0.0012. The process limits are not
that precise.

Allocation by Weight Factors:

Grinding the ring is the more costly process of the two. We would like to loosen the tolerance on
dimension e. As a first try, let the weight factors be wa = 10, we = 20. This will change the ratio
of the two tolerances and scale them to match the 1.0 deg limit. The original tolerances had a
ratio of 5:1. The final ratio will be the product of 1:2 and 5:1, or 2.5:1. The sensitivities do not
affect the ratio.
δφ = Σ((Sij δxj)2) (11)
0.017453 = (S11 P•10/30 δa)2+(S12 δc)2+(S13 P•20/30 δe)2
= (-2.6469•P•10/30•0.004)2+ (-10.5483•0.0004)2+(2.6272•P•20/30•0.0008)2
Solving for P:
P = 4.460
δa = 4.460•10/30•0.004 = 0.00595 in
δe = 4.460•20/30•0.0008 = 0.00238 in

Evaluating the results, we see that δa is within the 0.006in limit, but δe is well beyond the
0.0012in process limit. But, since δa is so close to its limit, we cannot change the weight factors
much without causing δa to go out of bounds. After several trials, the best design seemed to be
equal weight factors, which is the same as proportional scaling. We will present a plot later
which will make it clear why it turned out this way.

From the preceding examples, we see that the allocation algorithms work the same for 2-D and
3-D assemblies as for 1-D. We simply insert the tolerance sensitivities into the accumulation
formulas and carry them through the calculations as constant factors.

8.3 Allocation by Cost Minimization

The minimum cost allocation applies equally well to 2-D and 3-D assemblies. If sensitivities are
included in the derivation presented in Sec. 6.0, equations 8 through 10 become:

Table 8. Expressions for Minimum Cost Tolerances in 2-D and 3-D Assemblies

Worst Case RSS

 kiBiS1   kiBiS1 
Ti = k B S  1/(ki+1) ·T1(k1+1)/(ki+1) Ti = k B S  1/(ki+2) ·T1(k1+2)/(ki+2)
 1 1 i  1 1 i
TASM = S1T1 T2ASM = S12T12
 kiBiS1  1/(ki+1) (k1+1)/(ki+1)  kiBiS1 
+ •Si2 k B S  2/(ki+2)·T1 1+2)/(ki+2)
2(k
+ •Si k B S  ·T1
 1 1 i  1 1 i
The cost data for computing process cost is shown in Table 9:
Table 9. Process Tolerance Cost Data for the Clutch Assembly
Part Dim Proc Nom(in) Sens B k Min Tol Max Tol
Hub a Mill 2.1768 -2.6469 0.1018696 -0.45008 0.0025 0.006
Roller c Lap 0.9000 -10.548 0.000528 -1.130204 0.00025 0.00045
Ring e Grind 4.0000 2.62721 0.0149227 -0.79093 0.0005 0.0012

Minimum Cost Tolerances by Worst Case:

To perform tolerance allocation using a Worst Case stackup model, let T1 = δa, and Ti = δe, then
S1 = S11, k1 = ka, and B1 = Ba, etc.

TASM = |S11| δa + |S12| δc + |S13| δe (12)

 e e 11 1/(ke+1) (ka+1)/(ke+1)
k B S
= |S11| δa + |S12| δc + |S13| k B S  ·δa
 a a 13
0.017453= 2.6469 δa + 10.5483•0.0004

+ 2.6272 0.45008•0.1018696•2.6272
0.79093•0.0149227•2.6469 1/(0.20907) (0.54992)/(0.20907)
·δa
 
The only unknown is δa, which may be found by iteration. δe may then be found once δa is
known. Solving for δa and δe:
δa = 0.00198 in (13)

δe = 0.45008•0.1018696•2.6272
0.79093•0.0149227•2.6469 1/(0.20907) (0.54992)/(0.20907)
·0.00198
 
= 0.00304 in
The cost corresponding to holding these tolerances would be reduced from C= $5.42 to C=
$3.14.

Comparing these values to the process limits in Table 9, we see that δa is below its lower process
limit (0.0025< δa <0.006), while δe is much larger than the upper process limit (0.0005< δe
<0.0012). If we decrease δe to the upper process limit, δa can be increased until TASM equals the
spec limit. The resulting values and cost are then:
δa = 0.0038 in δe = 0.0012 in C = $4.30
The relationship between the resulting three pairs of tolerances is very clear when they are
plotted as shown in Fig. 14. Tol e and Tol a are plotted as points in 2-D tolerance space. The
feasible region is bounded by a box formed by the upper and lower process limits, which is cut
off by the Worst Case limit curve. The original tolerances of (0.004, 0.0008) lie within the
feasible region, nearly touching the WC Limit. Extending a line through the original tolerances
to the WC Limit yields the proportional scaling results found in Section 2.0 (0.00417, 0.00083),
which is not much improvement over the original tolerances. The minimum cost tolerances
(OptWC) were a significant change, but moved outside the feasible region. The feasible point of
lowest cost (Mod WC) resulted at the intersection of the upper limit for Tol e and the WC Limit
(0.0038, 0.0012).

This type of plot really clarifies the relationship between the three results. Unfortunately, it is
limited to a 2-D graph, so it is only applicable to an assembly with two design tolerances.
 0.005
WC Limit
0.004 Original

Opt WC
Opt WC
0.003
Tol e Mod WC
0.002
Mod WC WC Limit
0.001 Original

Feasible Region
0
0 0.002 0.004
Tol a

Figure 14. Tolerance allocation results for a Worst Case model.

Minimum Cost Tolerances by RSS:

Repeating the minimum cost tolerance allocation using the RSS stackup model:
TASM2 = (S11 δa)2 + (S12 δc)2 + (S13 δe)2 (14)
keBeS11 2/(ke+2) 2(ka+2)/(ke+2)
= (S11 δa)2 + (S12 δc)2 + S132 k B S  ·δa
 a a 13
(0.017453)2 = (2.6469 δa)2 + (10.5483•0.0004)2
+ 2.62722 0.45008•0.1018696•2.6272
0.79093•0.0149227•2.6469 2/(1.20907) 2(1.54992)/(1.20907)
·δa
 
Solving for δa by iteration and δe as before:
δa = 0.00409 in (15)
δe = 0.45008•0.1018696•2.6272
0.79093•0.0149227•2.6469 1/(1.20907) (0.54992)/(1.20907)
·0.00409
 
= 0.00495 in
The cost corresponding to holding these tolerances would be reduced from C= $5.42 to C=
$2.20.
Comparing these values to the process limits in Table 9, we see that δa is now safely within its
process limits (0.0025< δa <0.006), while δe is still much larger than the upper process limit
(0.0005< δe <0.0012). If we again decrease δe to the upper process limit as before, δa can be
increased until it equals the upper process limit. The resulting values and cost are then:
δa = 0.006 in δe = 0.0012 in C = $4.07
The plot in Fig. 15 shows the three pairs of tolerances. The box containing the feasible region is
entirely within the RSS Limit curve. The original tolerances of (0.004, 0.0008) lie near the center
of the feasible region. Extending a line through the original tolerances to the RSS Limit yields
the proportional scaling results found in Sec. 2.0 (0.00628, 0.00126), both of which lie just
outside the feasible region. The minimum cost tolerances (OptRSS) were a significant change,
but moved far outside the feasible region. The feasible point of lowest cost (ModRSS) resulted
at the upper limit corner of the feasible region (0.006, 0.0012).
 0.007
RSS Limit
0.006
Opt RSS Original
0.005
Opt RSS
0.004
Tol e
0.003 Mod RSS

0.002 RSS Limit

Original Mod RSS
0.001
0
Feasible Region
0 0.002 0.004 0.006
Tol a

Figure 15. Tolerance allocation results for the RSS model.

Comparing Figs. 14 and 15, we see that the RSS Limit curve intersects the horizontal and
vertical axes at values greater than 0.006 in, while the WC Limit curve intersects near 0.005 in.
tolerance. The intersections are found by letting Tol a or Tol e go to zero in the equation for
TASM and solving for the remaining tolerance. The RSS and WC Limit curves do not converge
to the same point because the fixed tolerance δc is subtracted from TASM differently for WC than
RSS.

9.0 Tolerance Allocation with Process Selection

Examining Fig. 15 further, the feasible region appears very small. There is not much room for
tolerance design. The optimization preferred to drive Tol e to a much larger value. One way to
enlarge the feasible region is to select an alternate process for dimension e. Instead of grinding,
suppose we consider turning. The process limits change to (0.002< δe <0.008), with Be =
0.118048 ke = -0.45747. Table 10 shows the revised data.

Table 10. Revised Process Tolerance Cost Data for the Clutch Assembly
Part Dim Proc Nom(in) Sens B k Min Tol Max Tol
Hub a Mill 2.1768 -2.6469 0.1018696 -0.45008 0.0025 0.006
Roller c Lap 0.9000 -10.548 0.000528 -1.130204 0.00025 0.00045
Ring e Turn 4.0000 2.62721 0.118048 -0.45747 0.002 0.008

Milling and turning are processes with nearly the same precision. Thus, Be and Ba are nearly
equal as are ke and ka. The resulting RSS allocated tolerances and cost are:
δa = 0.00434 in δe = 0.00474 in C = $2.54
The new optimization results are shown in Fig. 16. The feasible region is clearly much larger
and the minimum cost point (Mod Proc) is on the RSS Limit curve on the region boundary. The
new optimum point has also changed from the previous result (Opt RSS) due the change in Be
and ke for the new process.
 0.007
RSS Limit
Original
0.006
Opt RSS
0.005 Mod Proc Opt RSS

0.004 Mod RSS

Tol e
0.003 Mod Proc
0.002
Mod RSS RSS Limit
Original
0.001
0
Feasible Region
0 0.002 0.004 0.006
Tol a

Figure 16. Tolerance allocation results for the modified RSS model.
The resulting WC allocated tolerances and cost are:
δa = 0.00240 in δe = 0.00262 in C = $3.33
The modified optimization results are shown in Fig. 17. The feasible region is the smallest yet
due to the tight Worst Case Limit. The minimum cost point (Mod Proc) is on the WC Limit
curve on the region boundary.

0.005 Feasible Region

WC Limit
Original
0.004
Opt WC Opt WC
0.003 Mod Proc
Tol e Mod WC
0.002
Mod Proc
Mod WC
0.001 Original WC Limit

0
0 0.002 0.004
Tol a

Figure 17. Tolerance allocation results for the modified WC model.

Including cost functions with tolerance selection makes possible the quantitative comparison of
alternate processes to see if cost reductions could be achieved by a change in process. If cost vs.
tolerance data are available for a full range of processes, process selection can even be
automated. A very systematic and efficient search technique, which automates this task has been
published [Chase, et al., 1989]. It compares several methods for including process selection in
tolerance allocation and gives a detailed description of the one found to be most efficient.
10.0 Summary
The results of WC and RSS cost allocation of tolerances are summarized in the two bar charts,
Figs. 18 and 19. The changes in magnitude of the tolerances is readily apparent. Costs have
been added for comparison.

WC Cost Allocation Results

Original $5.42
a
Opt $3.14
c
Mod $4.30 e

Mod $3.33

0 0.002 0.004
Tolerance

Figure 18. Tolerance allocation results for the WC model.

RSS Cost Allocation Results

Original $5.42

a
Opt RSS $2.20
c

Mod RSS $4.07 e

Mod Proc $2.54

0 0.002 0.004 0.006

Tolerance

Figure 19 Tolerance allocation results for the RSS model.

Summarizing, the original tolerances for both WC and RSS were safely within tolerance
constraints, but the costs were high. Optimization reduced the cost dramatically, however, the
resulting tolerances exceeded the recommended process limits. The modified WC and RSS
tolerances were adjusted to conform to the process limits, resulting in a moderate decrease in
cost, about 20%. Finally, the effect of changing processes was illustrated, which resulted in a
cost reduction near the first optimization, only the allocated tolerances remained in the new
feasible region.
A designer would probably not attempt all of these cases in a real design problem. He would be
wise to rely on the RSS solution, possibly trying WC analysis for a case or two for comparison.
Note that the clutch assembly only had three dimensions contributing to the tolerance stack. If
there had been six or eight, the difference between WC and RSS would have been much more
significant.

It should be noted that tolerances specified at the process limit may not be desireable. If the
process is not well controlled, it may be difficult to hold it at the limit. In such cases, the
designer may want to back off from the limits to allow for process uncertainties.

11.0 References
[Chase 1995] Chase, K. W., J. Gao and S. P. Magleby "General 2-D Tolerance Analysis of
Mechanical Assemblies with Small Kinematic Adjustments", J. of Design and
Manufacturing, v 5 n 4, 1995.
[Chase 1988] Chase, K.W. and W.H. Greenwood, "Design Issues in Mechanical Tolerance
Analysis," Manufacturing Review, ASME, vol 1,no 1, March 1988, pp. 50-59.
[Chase 1989] Chase, K. W., W. H. Greenwood, B. G. Loosli and L. F. Hauglund, "Least Cost
Tolerance Allocation for Mechanical Assemblies with Automated Process Selection."
Manufacturing Review, ASME, vol 2,no 4, December 1989 .
[Chase 1991] Chase, K. W. and A. R. Parkinson "A Survey of Research in the Application of
Tolerance Analysis to the Design of Mechanical Assemblies," Research in Engineering
Design, v 3, pp.23-37, 1991.
[Fortini 1967] Fortini, E.T., Dimensioning for Interchangeable Manufacture, Industrial Press,
1967.

[Greenwood 1987] Greenwood, W.H., and K.W. Chase, "A New Tolerance Analysis Method
for Designers and Manufacturers," Journal of Engineering for Industry, Transactions of
ASME, Vol. 109, May 1987, pp. 112-116.

[Speckhart] Speckhart, F.H., "Calculation of Tolerance Based on a Minimum Cost Approach,"

Journal of Engineering for Industry, Transactions of ASME, Vol. 94, May 1972, pp. 447-
453.

[Spotts 1973] Spotts, M.F., "Allocation of Tolerances to Minimize Cost of Assembly," Journal
of Engineering for Industry, Transactions of the ASME, Vol. 95, August 1973, pp. 762-764.

[Trucks 1987] Trucks, H. E., Designing for Economic Production, 2nd ed., Society of
Manufacturing Engineers, 1987.
 APPENDIX

Cost-Tolerance Functions for Metal Removal Processes

Although it is well known that tightening tolerances increases cost, adjusting the tolerances on
several components in an assembly and observing its effect on cost is an impossible task. Until
you have a mathematical model, you can not effectively optimize the allocation of tolerance in an
assembly. Elegant tools for minimum cost tolerance allocation have been developed over several
decades. However, they require empirical functions describing the relationship between
tolerance and cost.

Cost vs. tolerance data is very scarce. Very few companies or agencies have attempted to gather
such data. Companies who do consider it proprietary, so it is not published. The data is site and
machine-specific and subject to obsolescence due to inflation. In addition, not all processes are
capable of continuously adjustable precision.

Metal removal processes have the capability to tighten or loosen tolerances by changing feeds,
speeds, and depth of cut or by modifying tooling, fixtures, cutting tools and coolants. The
workpiece may also be modified by switching to a more machinable alloy or modifying
geometry to achieve greater rigidity.

A noteworthy study by the U.S. Army in the 1940s experimentally determined the natural
tolerance range for the most common metal removal processes [4]. They also compared the cost
of the various processes and the relative cost of tightening tolerances. Relative costs were used
to eliminate the effects of inflation. The resulting chart, Table. A-1, appears in references [1 and
2]. Least squares curve fits were performed at BYU and are presented here for the first time.
The Reciprocal Power equation, C = A + B/Tk, presented in Chapter 12, was used as the
empirical function. Fig. A-1 shows a typical plot of the original data and the fitted data. The
curve fit procedure was a standard nonlinear method described in reference [3], which uses
weighted logarithms of the data to convert to a linear regression problem. Results are tabulated
in Table A-2 and plotted in Figs. A-2 and A-3.
 Turn

2.5
Size 4: Data
2
Size 4: Fitted
1.5
Size 5: Data
1
Size 5: Fitted
0.5
Size 6: Data
0
0 0.005 0.01 Size 6: Fitted
Tolerance

Fig. A-1 Plot of cost vs tolerance for fitted and raw data for the turning process
 Table A-1 Relative Cost of Obtaining Various Tolerance Levels

Range of Sizes (in.)

From To Tolerances (in.)
0.000 0.599 0.0002 0.00025 0.0004 0.0005 0.0008 0.0012 0.0020 0.0030 0.0050
0.600 0.999 0.00025 0.0003 0.00045 0.0006 0.0010 0.0015 0.0025 0.0040 0.0060
1.000 1.499 0.0003 0.0004 0.0005 0.0008 0.0012 0.0020 0.0030 0.0050 0.0080
1.500 2.799 0.0004 0.0005 0.0006 0.0010 0.0015 0.0025 0.0040 0.0060 0.0100
2.800 4.499 0.0005 0.0006 0.0008 0.0012 0.0020 0.0030 0.0050 0.0080
4.500 7.799 0.0006 0.0007 0.0010 0.0015 0.0025 0.0040 0.0060 0.0100
7.800 13.599 0.0007 0.0008 0.0012 0.0020 0.0030 0.0050 0.0080 0.0120
13.600 20.999 0.0008 0.0010 0.0015 0.0025 0.0040 0.0060 0.0100 0.0150
21.00 and over follow same tolerancing trends
Process Relative Cost of Tightening Tolerance* rocess
ost
Lap and Hone 200% 180% 100% 300%
Grind, Diamond Turn and Bore 200% 180% 140% 100% 300%
Broach 200% 175% 140% 100% 200%
Ream 175% 140% 100% 175%
Turn, Bore, Slot, Plane, and Shape 200% 170% 140% 100% 100%
Mill 150% 125% 100% 100%
Drill 175% 100% 100%
*Total relative cost for a given process is the percentage product of the tolerance tightening cost and the process cost (200%*300%=600%).
Reproduced from reference [2]
 Table A-2 Cost-Tolerance Functions for Metal Removal Processes
Size Range A B k Min Tol Max Tol
Lap / Hone
0.000-0.599 0.00189378 0.9508781 0.0002 0.0004
0.600-0.999 0.00052816 1.1302036 0.00025 0.00045
1.000-1.499 0.00220173 0.9808618 0.0003 0.0005
1.500-2.799 0.00033129 1.2590875 0.0004 0.0006
2.800-4.499 0.00026156 1.3269297 0.0005 0.0008
4.500-7.799 0.00038119 1.3073528 0.0006 0.001
7.800-13.599 0.00059824 1.2716314 0.0007 0.0012
13.600-20.999 0.00427422 1.0221757 0.0008 0.0015
Grind / Diamond turn
0.000-0.599 0.02484363 0.6465727 0.0002 0.0005
0.600-0.999 0.01525616 0.7221989 0.00025 0.0006
1.000-1.499 0.0205072 0.7039047 0.0003 0.0008
1.500-2.799 0.0133561 0.7827624 0.0004 0.001
2.800-4.499 0.01492268 0.790932 0.0005 0.0012
4.500-7.799 0.02467047 0.7413291 0.0006 0.0015
7.800-13.599 0.05119944 0.6548091 0.0007 0.002
13.600-20.999 0.08317908 0.6017646 0.0008 0.0025
Broach
0.000-0.599 0.0438552 0.548619 0.00025 0.0008
0.600-0.999 0.04670538 0.55230115 0.0003 0.001
1.000-1.499 0.04071362 0.58686634 0.0004 0.0012
1.500-2.799 0.048524 0.579761 0.0005 0.0015
2.800-4.499 0.0637591 0.559608 0.0006 0.002
4.500-7.799 0.0922923 0.521758 0.0007 0.0025
7.800-13.599 0.144046 0.46957 0.0008 0.003
13.600-20.999 0.171785 0.45907 0.001 0.004
Ream
0.000-0.599 0.03245261 0.6000163 0.0005 0.0012
0.600-0.999 0.04682158 0.565492 0.0006 0.0015
1.000-1.499 0.04204992 0.6021191 0.0008 0.002
1.500-2.799 0.04809684 0.6021191 0.001 0.0025
2.800-4.499 0.06929088 0.565492 0.0012 0.003
4.500-7.799 0.09203907 0.5409254 0.0015 0.004
Turn / bore / shape
0.000-0.599 0.07201641 0.46822793 0.0008 0.003
0.600-0.999 0.085969502 0.45747142 0.001 0.004
1.000-1.499 0.101233386 0.44723008 0.0012 0.005
1.500-2.799 0.11800302 0.4389869 0.0015 0.006
2.800-4.499 0.11804756 0.45747142 0.002 0.008
4.500-7.799 0.12576137 0.46536684 0.0025 0.01
7.800-13.599 0.15997103 0.4389869 0.003 0.012
13.600-20.999 0.15300611 0.46822793 0.004 0.015
Mill
0.000-0.599 0.0862308 0.4259173 0.0012 0.003
0.600-0.999 0.10878812 0.4044547 0.0015 0.004
1.000-1.499 0.09544417 0.4431399 0.002 0.005
1.500-2.799 0.10186958 0.4500798 0.0025 0.006
2.800-4.499 0.14399071 0.4044547 0.003 0.008
4.500-7.799 0.12976209 0.4431399 0.004 0.01
7.800-13.599 0.13916564 0.4500798 0.005 0.012
13.600-20.999 0.17114563 0.4259173 0.006 0.015
Drill
0.000-0.599 0.00301435 1.0955124 0.003 0.005
0.600-0.999 0.00085791 1.3801824 0.004 0.006
1.000-1.499 0.00318631 1.1906627 0.005 0.008
1.500-2.799 0.00644133 1.0955124 0.006 0.01
2.800-4.499 0.00223316 1.3801824 0.008 0.012
Lap / Hone Turn / bore / shape

8 3
6
2
4
1
2
0 0
0 0.0005 0.001 0.0015 0 0.005 0.01 0.015

Grind / Diamond turn Mill

8 2
6 1.5
4 1
2 0.5
0 0
0 0.0005 0.001 0.0015 0.002 0.0025 0 0.005 0.01 0.015

Broach Drill

6 2
1.5
4
1
2
0.5
0 0
0 0.001 0.002 0.003 0.004 0 0.004 0.008 0.012

Ream

4
3
2
1
0
0 0.001 0.002 0.003 0.004

Figure A-2 Plot of Fitted Cost vs. Tolerance Functions

 B k
Lap / Hone

0.006 1.5
0.004 1
0.002 0.5
0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18

Grind / Diamond turn

0.1 1.5
1
0.05
0.5
0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18

Broach

0.2 1.5
0.15 1
0.1
0.05 0.5
0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18

Ream

0.1 1.5
1
0.05
0.5
0 0
0 1 2 3 4 5 6 0 2 4 6

Turn / bore / shape

0.2 1.5
0.15
1
0.1
0.05 0.5
0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18

Figure A-3 Plot of Coefficients vs. Size for Cost-Tolerance Functions

 B k
Mill

0.2 1.5
0.15 1
0.1
0.05 0.5
0 0
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18

Drill

0.01 1.5
1
0.005
0.5
0 0
0 1 2 3 4 0 1 2 3 4

Figure A-3 (continued) Plot of Coefficients vs. Size for Cost-

Tolerance Functions

Curve fits were performed by mechanical engineering student, David Todd.

References:
1. Hansen, Bertrand L., Quality Control: Theory and Applications, Prentice-Hall, Upper Saddle
River, NJ, 1963.
2. Jamieson, Archibald, Introduction to Quality Control, Reston Publishing, Reston, VA, 1982.
3. Pennington, Ralph H., Introductory Computer Methods and Numerical Analysis, 2nd ed,
MacMillan, Toronto, Canada, 1970.
4. U.S.Army Management Engineering Training Activity, Rock Island Arsenal, IL.
(Original report is out of print)