Chapter

12
Multi-Dimensional Tolerance Analysis
(Manual Method)

Dale Van Wyk

Dale Van Wyk

Raytheon Systems Company
McKinney, Texas
Mr. Van Wyk has more than 14 years of experience with mechanical tolerance analysis and mechanical
design at Texas Instruments Defense Group, which became part of Raytheon Systems Company. In
addition to direct design work, he has developed courses for mechanical tolerancing and application
of statistical principles to systems design. He has also participated in development of a U.S. Air Force
training class, teaching techniques to use statistics in creating affordable products. He has written
several papers and delivered numerous presentations about the use of statistical techniques for mechanical tolerancing. Mr. Van Wyk has a BSME from Iowa State University and a MSME from Southern
Methodist University.
12.1

Introduction

The techniques for analyzing tolerance stacks that were introduced in Chapter 9 were demonstrated using
a one-dimensional example. By one-dimensional, we mean that all the vectors representing the component
dimensions can be laid out along a single coordinate axis. In many analyses, the contributing dimensions
are not all along a single coordinate axis. One example is the Geneva mechanism shown in Fig. 12-1. The
tolerances on the C, R, S, and L will all affect the proper function of the mechanism. Analyses like we
showed in Chapters 9 and 11 are insufficient to determine the effects of each of these tolerances. In this
chapter, well demonstrate two methods that can be used to evaluate these kinds of problems.

12-1

12-2

Chapter Twelve

Figure 12-1 Geneva mechanism

showing a few of the relevant dimensions

The following sections describe a systematic procedure for modeling and analyzing manufacturing
variation within 2-D and 3-D assemblies. The key features of this system are:
1. A critical assembly dimension is represented by a vector loop, which is analogous to the loop diagram
in 1-D analysis.
2. An explicit expression is derived for the critical assembly feature in terms of the contributing component dimensions.
3. The resulting expression is used to calculate tolerance sensitivities, either by partial differentiation or
numerical methods.
A key benefit is that, once the expression is derived, this method easily solves for new nominal values
directly as the design changes.
12.2

Determining Sensitivity

Recall the equations for worst case and RSS tolerance analysis equation from Chapter 9 (Eqs. 9.2 and 9.11).
n

t wc = ai t i

(12.1)

i =1

t rss =

(a t )
i =1

i i

(12.2)

The technique well demonstrate for multidimensional tolerance analysis uses these same equations
but well need to develop another way to determine the value of the sensitivity, a i, in Eqs. (12.1) and (12.2)
above. We noted in Chapter 9 that sensitivity is an indicator of the effect of a dimension on the stack. In

Multi-Dimensional Tolerance Analysis (Manual Method)

12-3

one-dimensional stacks, the sensitivity is almost always either +1 or -1 so it is often left out of the onedimensional tolerance equations. For the Geneva mechanism in Fig. 12-1, an increase in the distance L
between the centers of rotation of the crank and the wheel require a change in the diameter, C, of the
bearing, the width of the slot, S, and the length, R, of the crank. However, it wont be a one-to-one
relationship like we usually have with a one-dimensional problem, so we need a different way to find
sensitivity.
To see how were going to determine sensitivity, lets start by looking at Fig. 12-2. If we know the
derivative (slope) of the curve at point A, we can estimate the value of the function at points B and C as
follows:

F ( B ) F ( A ) + x

dy
dx

and

F (C ) F ( A) x

dy
dx

y
F(x)
F(C)
x

F(A)
F(B)

Figure 12-2 Linearized approximation to a curve

Well use the same concept for multidimensional tolerance analysis. We can think of the tolerance as
x, and use the sensitivity to estimate the value of the function at the tolerance extremes. As long as the
tolerance is small compared to the slope of the curve, this provides a very good estimate of the effects of
tolerances on the gap.
With multidimensional tolerance analysis, we usually have several variables that will affect the gap.
Our function is an n-space surface instead of a curve, and the sensitivities are found by taking partial
derivatives with respect to each variable. For example, if we have a function ( y1,y2,yn), the sensitivity
of with respect to y1 is

a1 =

y1

Nominal Values

12-4

Chapter Twelve

Therefore we evaluate the partial derivative at the nominal values of each of the variables. Remember
that the nominal value for each variable is the center of the tolerance range, or the value of the dimension
when the tolerances are equal bilateral. Once we find the values of all the sensitivities, we can use any of
the tolerance analysis or allocation techniques in Chapters 9 and 11.
12.3
A Technique for Developing
Gap Equations
Developing a gap equation is the key to performing a multidimensional tolerance analysis.
Well show one method to demonstrate the
technique. While were using this method as
an example, any technique that will lead to an
accurate gap equation is acceptable. Once we
develop the gap equation, well calculate the
sensitivities using differential calculus and
complete the problem using any tolerance
analysis or allocation technique desired. A flow
chart listing the steps is shown in Fig. 12-3.
Well solve the problem shown in Fig.
12-4. While this problem is unlikely to occur
during the design process, its use demonstrates
techniques that are helpful when developing
gap equations.

Define requirement of interest

Establish gap coordinate system

Draw vector loop diagram

Establish component coordinate systems

Write vectors in terms of

component coordinate systems

Step 1. Define requirement of interest

The first thing we need to do with any tolerance analysis or allocation is to define the requirement that we are trying to satisfy. In this
case, we want to be able to install the two
blocks into the frame. We conducted a study
of the expected assembly process, and decided
that we need to have a minimum clearance of
.005 in. between the top left corner of Block 2
and the Frame. We will perform a worst case
analysis using the dimensions and tolerances
in Table 12-1. The variable names in the table
correspond to the variables shown in Fig.
12-4.
Step 2. Establish gap coordinate system
Our second step is establishing a coordinate
system at the gap. We know that the shortest
distance that will define the gap is a straight
line, so we want to locate the coordinate sys-

Define relationships between coordinate systems

Convert all vectors into gap coordinate system

Generate gap equation

Calculate sensitivities

Perform tolerance analysis or allocation

Figure 12-3 Multidimensional tolerancing flow chart

Multi-Dimensional Tolerance Analysis (Manual Method)

12-5

Frame
M

Block 2
Block 1

H
D
C

E
F

Figure 12-4 Stacked blocks we will use for an example problem

Table 12-1 Dimensions and tolerances corresponding to the variable names in Fig. 12-4

Variable Name

Mean Dimension (in.)

Tolerance (in.)

A
B
C
D
E
F
G
H
J
K
M

.875
1.625
1.700
.875
2.625
7.875
4.125
1.125
3.625
5.125
1.000

.010
.020
.012
.010
.020
.030
.010
.020
.015
.020
.010

12-6

Chapter Twelve

tem along that line. We set the origin at one side of the gap and one of the axes will point to the other side,
along the shortest direction. Its not important which side of the gap we choose for the origin. Coordinate
system{u1,u2}is shown in Fig. 12-5 and represents a set of unit vectors.

Frame
u2
u1

Block 2
Figure 12-5 Gap coordinate system
{u1,u2}

Step 3. Draw vector loop diagram

Now well have to draw a vector loop diagram similar to the dimension loop diagram constructed in section
9.2.2. Just like we did with the one-dimensional loop diagram, well start at one side of the gap and work our
way around to the other. Anytime we go from one part to another, it must be through a point or surface of
contact. When weve completed our analysis, we want a positive result to represent a clearance and a
negative result to represent an interference. If we start our vector loop at the origin of the gap coordinate
system, well finish at a more positive location on the axis, and well achieve the desired result.
For our example problem, there are several different vector loops we can chose. Two possibilities are
shown in Fig. 12-6. The solution to the problem will be the same regardless of which vector loop we
choose, but some may be more difficult to analyze than others. Its generally best to choose a loop that has
a minimum number of vectors that need the length calculated. In Loop T, vectors T2 and T3 need the length
calculated while Loop S has five vectors with undefined lengths. We can find lengths of the vectors S 5 and
S 6 through simple one-dimension analysis, but S 2, S 4, and S 6 will require more work. So it appears that Loop
T may provide easier calculations.
Finish

T3, S7

Start
Loop S

S1

T1

Loop T
T2

S2
S3

S6

S4
S5

Figure 12-6 Possible vector loops to evaluate the gap of interest

Multi-Dimensional Tolerance Analysis (Manual Method)

12-7

As an alternative, look at the vector loop in Fig. 12-7. It has only three vectors with unknown length,
one of which (x 9) is a linear combination of other dimensions. For vectors x 2 and x 10, we can calculate the
length relatively easily. This is the loop we will use to analyze the problem.

Finish

x10

Start
x1

x2
x3

x4

x7

x6

x5

x9
x8

Figure 12-7 Vector loop we will use to analyze the gap. It presents easier calculations of unknown vector lengths.

Step 4. Establish component coordinate systems

The next step is establishing component coordinate systems. The number needed will depend on the
configuration of the assembly. The idea is to have a coordinate system that will align with every component dimension and vector that will contribute to the stack. One additional coordinate system is needed
and is shown in Fig. 12-8.
Coordinate system {v1,v2} is needed for the vectors on Block 2. The dimensions on the frame align
with {u1,u2} so an additional coordinate system is not needed for them. Dimensions J and H on Block 1 do
not contribute directly to a vector length so they do not need a coordinate system.

v2

Block 1

Block 2

v1

Figure 12-8 Additional coordinate system

needed for the vectors on Block 2

12-8

Chapter Twelve

Step 5. Write vectors in terms of component coordinate systems

The vectors in Fig. 12-7 are listed below in terms of their coordinate systems, angle , and the dimensional
variables in Table 12-1.
x 1 = -Mv2

F C B E M sin
x 2 = K
v 1
cos

x 3 = -Eu1
x 4 = -Au2
x 5 = -Bu1
x 6 = -Du2
x 7 = -Cu1
x 8 = Fu1
x 9 = Gu2
x 10 = Kcos u1
Angle is not known yet, so well have to calculate it. Angle contributes to the value of , and is
also needed. The equations for angles and are shown below..

A
a = arctan

B
.875
= arctan

1.625
= 28.30

C Hsin a

J
A 2 + B 2 sina + Hcosa
cosa

= arctan

C Hsin a

E J
A 2 + B 2 cosa + Hsina
cos
a

1.700 1.125 ( .4741 )

3.625
.875 2 + 1.625 2 ( .4741 )
.8805

+ 1.125 (.8805 )

= arctan

1.700 1.125 (.4741 )

2
2
.875 + 1.625 (.8805 )
2.625 3.625
.8805

+ 1.125 (.4741 )

= 23.62

Step 6. Define relationships between coordinate systems

In order to relate the vectors in Step 5 to the gap, we will have to transform them into the same coordinate system as
the gap. Thus, well have to convert vectorsx1 and x2 into coordinate system {u1,u2}. One method follows.

Multi-Dimensional Tolerance Analysis (Manual Method)

u2

12-9

v2
u1
v1

Figure 12-9 Relationship between

coordinate systems {u1,u2} and {v1,v2}

Fig. 12-9 shows the {u1,u2} and {v1,v2} coordinate systems and the angle
between
them. To build a transformation between the two coordinate systems,
v1
well find the components of v1 and v2 in the directions of the unit vectors u1
v2
and u2. For example, the component of v1 in the u1 direction is cos . The
component of v1 in the u2 direction is -sin . The sign of the sine is negative because the component is
pointing in the opposite direction as the positive u2 axis. The table is completed by performing a similar
analysis with vector v2.
A matrix, Z, can be defined as follows:

u1
cos
sin

u2
-sin
cos

cos sin
Z =

sin cos
Multiplying Z by and {u1,u2}T will give us a transformation matrix that we can use to convert any
vector in the {v1,v2} coordinate system to the {u1,u2} coordinate system.
Let Q = Z{u1,u2}T

cos sin u1
Q=

sin cos u2
cos u1 sin u 2
Q=

sin u1 + cos u 2
Now we can transform any vector in the {v1,v2} coordinate system to the {u1,u2} coordinate system
by multiplying it by Q.
Lets see how this works by transforming the vector 2v1 + v2 to the {u1,u2}coordinate system. We start
by representing the vector as a matrix [2 1].

cos u 1 sin u 2
2v 1 + v 2 = [ 2 1]

sin u 1 + cos u 2
= 2( cos u1 sin u 2 ) + sin u 1 + cos u 2
= ( 2 cos + sin ) u 1 + ( cos 2 sin ) u 2

Step 7. Convert all vectors into gap coordinate system

For our problem, we need all the vectors x i that we found in Step 5 to be represented in the {u1,u2}
coordinate system. The only ones that need converting are x 1 and x 2.

12-10

Chapter Twelve

x1 = Mv 2
cos u1 sin u 2
= M [0 1 ]

sin u1 + cos u 2
= M (sin u1 + cos u 2 )
Similarly,

F C B E M sin

x 2 = K
cos

F C B E M sin
= K
cos

F C B E M sin

= K
cos

v 1

cos u 1 sin u 2
[1 0 ]

sin u 1 + cos u 2

( cos u 1 sin u 2 )

Step 8. Generate gap equation

To generate the gap equation now is very easy. We only need to observe that no components in the u1
direction affect the gap. Thus, all we need to do is take the components in the u2 direction and add them
together.

F C B E M sin
( sin ) A D + G
Gap = M cos + K
(12.3)
cos

Now we have to insert the nominal values of each of the dimensions along with the values of the sin
and cos into Eq. (12.3).

7.875 1.700 1.625 2. 625 1. 00(. 4007 )

Gap = 1. 000 (.9162 ) + 5. 125
( .4007 )
. 9162

.875 .875 + 4.125

= . 0719
This is the nominal value of the gap.
Step 9. Calculate sensitivities
Next we need to calculate the sensitivities, which well find by evaluating the partial derivatives at the
nominal value for each of the dimensions. As an example to the approach, well find the sensitivity for
variable E, and provide tabulated results for the other variables.
Since is a function of E, well have to apply the chain rule for partial derivatives. Lets start by
redefining the gap as a function of and E, say Gap = ( ,E). All the other terms will be treated as
constants. Then,

Gap dE
=
+
E
E dE E

Multi-Dimensional Tolerance Analysis (Manual Method)

12-11

Solving for each of the terms,

= tan
E
= . 4373
dE
=1
dE

F C B E M sin K ( cos ) 3
=

( cos ) 2
=

7. 875 1.700 1.625 2 .625 1. 000 (.4007 ) 5. 125 (.9162 ) 3

(.9162 ) 2

= 2. 8796

C H sin

J
A 2 + B 2 sin H cos
cos

=
2
E

E J C H sin A 2 + B 2 cos + H sin

cos

C H sin

A 2 + B 2 sin + H cos
+ J

cos

1. 700 1.125 (.4741 )

3 .625
.875 2 + 1. 625 2 (.4741 ) 1 .125 (.8805 )
.
8805

2.625 3. 625 1.700 1. 125 ( .4741 ) .875 2 + 1.625 2 (.8805 ) + 1. 125 ( .4741 )

.8805

1 .700 1.125 (.4741 )

.875 2 + 1 .625 2 ( .4741 ) + 1.125 (. 8805 )

+ 3.625

.
8805

= .1331

Gap
= .4373 (1) + ( 2.8796
E
= .0540

) ( .1331 )

Table 12-2 contains the sensitivities of the remaining variables. While calculating sensitivities manually is difficult for many gap equations, there are many software tools that can calculate them for us,
simplifying the task considerably.

12-12

Chapter Twelve

Table 12-2 Dimensions, tolerances, and sensitivities for the stacked block assembly

Variable Name

Mean Dimension (in.)

Tolerance (in.)

Sensitivity

A
B
C
D
E
F
G
H
J
K
M

.875
1.625
1.700
.875
2.625
7.875
4.125
1.125
3.625
5.125
1.000

.010
.020
.012
.010
.020
.030
.010
.020
.015
.020
.010

-.5146
.1567
.4180
-1.0000
-.0540
.4372
1.0000
-.9956
-.7530
-.4006
-1.0914

Step 10. Perform tolerance analysis or allocation

Now that we have calculated a nominal gap (.0719 in.) and all the sensitivities, we can use any of the
analysis or allocation methods in Chapters 9 and 11. In Step 1, we decided to perform a worst case
analysis. Using Eq. (12.1),

t wc = ( .5146)(.010 ) + (.1567)(.020) + (.4180)(.012) + ( 1)(.010) +

( .0540)(.020) + (.4372)(.030) + (1)(.010) + ( .9956)(.020 )

( .7530)(.015) + ( .4006)(.020) + (1)(.010 )

= .0967
The minimum gap expected at worst case will be .0719 - .0967 = -.0248 in.
The negative number indicates that we can have an interference at worst case, and we do not satisfy
our assembly requirement of a minimum clearance of .005 in.
12.4

Utilizing Sensitivity Information to Optimize Tolerances

Since we dont meet our assembly requirement, we need to consider some alterations to the design. We
can use the sensitivities to help us make decisions about what we should target for change. For example,
dimensions B and E have small sensitivities, so changing the tolerance on them will have little effect on the
gap. To reduce the magnitude of the worst case tolerance stack, we would target the dimensions with the
largest sensitivity first.
Also, the sensitivities help us decide which dimension we should consider changing to increase the
gap. It takes a large change in a dimension with a small sensitivity to make a significant change in the gap.
For example, making Dimension E .018 in. smaller will make the gap only about .001 in. larger. Conversely,
making Dimension M .001 in. smaller will make the gap slightly more than .001 in. larger. If our goal is to
correct the problem of assembly fit without changing the design any more than necessary, working with
the dimensions with the largest sensitivities will be advantageous.
The simplest solution would be to increase the opening in the frame, Dimension G, from 4.125 in. to
4.160 in. which will provide the clearance we need. However, if we assume the thickness of the top of the
frame cant change, that will cause us to increase the size of the frame. That could be a problem. So instead,
well change one of the internal dimensions on the frame, making Dimension A equal to .815 in. With this

Multi-Dimensional Tolerance Analysis (Manual Method)

12-13

change, the nominal gap will be .1044 in., worst case tolerance stack is .0980 in. and the minimum clearance
is .0064 in.
The worst case tolerance stack increased because many of the sensitivities changed when A was
changed. This is because we evaluate the partial derivatives at the nominal value of the dimensions, so
when the nominal value of A was changed, we changed the calculated result. Another way to think of it is
that we moved to a different point in our design space when we changed Dimension A, so the slope
changed in several different directions.
The final dimensions, tolerances and sensitivities are shown in Table 12-3.
Table 12-3 Final dimensions, tolerances and sensitivities of the stacked block assembly

12.5

Variable Name

Mean Dimension (in.)

A
B
C
D
E
F
G
H
J
K
M

.815
1.625
1.700
.875
2.625
7.875
4.125
1.125
3.625
5.125
1.000

Tolerance
(in.)
.010
.020
.012
.010
.020
.030
.010
.020
.015
.020
.010

Sensitivity
-.5605
.1642
.3846
-1.0000
-.0552
.4488
1.0000
-.9811
-.7450
-.4094
-1.0961

Summary

In this section, weve demonstrated a technique for analyzing tolerances for multi-dimensional problems.
While this is an approximate method, the results are very good as long as tolerances are not too large
compared to the curvature of the n-space surface represented by the gap equation. Its good to remember
that once we have found the gap equation and calculated the sensitivities, we can use any of the analysis
or allocation techniques discussed in Chapters 9 and 11.
An important point to reiterate is that we show one method for developing a gap equation. While this
will give accurate results, it may be more cumbersome at times than deriving the equation directly from the
geometry of the problem. In general, the more complicated problems will be easier to solve using the
technique shown here because it helps break the problem into smaller pieces that are more convenient to
evaluate.
In this section, we evaluated an assembly that is not similar to ones found during the design process,
but the technique works equally well on typical design problems. In fact, one thing very powerful about
this technique is that it is not limited to traditional tolerance stacks. For example, we can use it to evaluate
the effect of tolerances on the magnitude of the maximum stress in a loaded, cantilevered beam. Once we
have developed the stress equation, we can calculate the sensitivities and determine the effect of things
like the length, width and thickness of the beam, location of the load, and material properties such as the
modulus of elasticity and yield strength. It even works well for electrical problems, such as evaluating the
range of current well see in a circuit due to tolerances on the electrical components.