You are on page 1of 50

# Quality Management in the Bosch Group | Technical Statistics

5. Statistical Tolerancing

Edition 08.1993

## 1993 Robert Bosch GmbH

Statistical Tolerancing

1. 2. 3. 4. 4.1 4.2 4.3 4.3.1 4.3.2 5. 5.1 5.2 5.3 5.4 5.5 6. 7. 8. 9. 10. Introduction ................................................................................. 4 Terms ........................................................................................... 5 Arithmetic Tolerancing ................................................................ 7 Theoretical Basis for Statistical Tolerancing...............................10 Central Limit Theorem of Statistics ............................................10 Law of Error Propagation............................................................15 Distribution of Characteristics and Tolerance .............................17 Variance of a Rectangular Distribution .......................................18 Variance of a Triangular Distribution .........................................19 Statistical Tolerancing ................................................................20 The Reduction Factor ..................................................................24 Calculation of the Statistical Assembly Tolerance ......................27 Calculation of Individual Tolerances for a Given Assembly Tolerance ..............................................27 The General Case ........................................................................28 Process Capability Indices ..........................................................29 Prerequisites for Statistical Tolerancing......................................31 Statistical Control of the Individual Dimensions .........................33 Generalisation for arbitrary Dimension Chains ...........................35 Bibliography ...............................................................................36 Appendix ....................................................................................38 Index ...........................................................................................48

-3-

1. Introduction
Statistical tolerancing is a method to determine tolerances on the basis of statistical principles. The purpose is to determine a tolerance for the assembly dimension, which is a geometric combination of several individual dimensions. Statistical tolerancing is based on the idea that random positive and negative deviations, present in actual dimensions of individual parts, from their respective nominal dimensions will neutralize each other as a rule, when these individual parts are stacked in an assembly. Thereby the most unfavorable case that can be thought of, for example exclusive combination of maximum possible positive dimensional deviations (arithmetic tolerancing) is very improbable. When the tolerance for an assembly dimension is given, then, under the aforesaid consideration, greater tolerances result for the individual dimensions than those resulting from mere arithmetical calculation. The resultant advantages with respect to the production technique and production costs have been explained in detail in many papers and publications on this subject and have frequently been highlighted already in the titles, as the following examples show (faithful translations of the German titles): Realising advantages in production by applying statistical principles to tolerance calculation  Coupling of tolerances in theory and practice: A possibility to reduce production costs  Economic tolerancing  Determining optimum tolerances on the basis of dimension chain theory and the theoretical properties of the production process  The dates of publication of these and other publications on Statistical Tolerancing (cp. Bibliography) show that the fundamentals of this method were indeed known long back. Its development can be traced back to the twenties of this century (cp. ). Considering this fact as well as the advantages of Statistical Tolerancing, pointed out in the literature, it is surprising that this method, which is also denoted as probability-theoretical method for calculation of dimension chains, is not consistently taken into consideration. New activities in the sphere of Statistical Tolerancing attribute this to the expensive calculation procedures to be applied for general cases, which have become economically possible only of late with the developments in the sphere of computer hardware and software. Within the scope of this booklet a short summary will be provided on the theoretical background of Statistical Tolerancing, the constraints listed out and the problems of practical application presented.

-4-

2. Terms
The assembly of individual parts within a technical system (an assembly) leads often to the situation, that independent dimensions of individual parts form an interrelated chain of linear dimensions, viewed purely from geometric aspects, which in its totality becomes the so-called assembly dimension. It is called, suggestively, a dimension chain, consisting of many individual dimensions. The dimension chain is called linear, if all the individual dimensions can be presented through arrows, parallel to each other, which form a closed chain of lines. The direction of the arrows denotes the additive or subtractive (positive or negative) contributions of the individual dimensions. A dimension is called positive (negative), if, with the change of this dimension and all other dimensions of the chain remaining constant, the assembly dimension changes in the same (opposite) direction.

Fig. 2.1: Tolerance chain graph Dimension chain consisting of individual nominal dimensions Ni Top: only positive contributory direction Bottom: positive and negative contributory direction Apart from this there are plane (two-dimensional) and spatial (threedimensional) dimension chains. Plane (two-dimensional) dimension chains often result from interaction of rotating parts or movement of rocker arms or cams, where angles and radii play a part (cp. for example ). It is clear, that in such nonlinear dimension chains complicated correlations of individual dimensions can occur, which can however be presented in most of the cases through analytical functions (but occasionally can only be determined by recursive calculation).

-5-

A simple example of a (two-element) nonlinear dimension chain is that of two holes, whose position on two axes perpendicular to each other is given by the respective distance from a reference point (intersection point of the axes).

Fig. 2.2: Example of a two-element nonlinear dimension chain The relative distance of the two holes is then defined by

z = x2 + y2 .

(2.1)

Tolerances are determined, according to DIN 7182  (cp. also -) by indicating a maximum dimension G U and a minimum dimension G L :
T = GU G L .

## The midpoint of the tolerance interval is

C= GU + G L . 2

(2.2)

If the nominal dimension of a characteristic does not coincide with this midpoint, then the tolerance is specified by way of a minimum and maximum allowance with reference to the nominal dimension.

Hint (not contained in the German edition): The notations used in this subject field are not standardised and not uniform. Therefore it is sometimes not quite clear what the assembly criterion of interest is. As can be seen from Fig. 2.1 Ns can denote the nominal assembly dimension or an assembly tolerance gap (clearance). It should be clear from the context, however, what is meant.

-6-

3. Arithmetic Tolerancing
Analogous to the linking of individual dimensions to form an assembly dimension, individual tolerances are linked to form an overall assembly tolerance. Assembly tolerances are often determined on the basis of theoretical considerations or on the basis of experience, taking into account their significance for: - function - reliability - lifetime - optical quality (appearance) - compliance with legal and official regulations. As result of a design calculation, tolerances for individual dimensions are then determined. To the contrary, if individual tolerances are given, then the assembly dimension can be calculated by inversion of this concept. The actual dimensions of parts must, as per DIN 7182, be within the tolerance zone bounded by the minimum dimension and the maximum dimension. If the most unfavorable case is assumed, that all actual dimensions of the individual parts of an assembly have the maximum permissible deviation from the nominal dimension, i.e., are lying at the limit of the tolerance zone, all individual tolerances Ti must be added arithmetically to obtain the assembly tolerance, independent of the respective contributory direction of the individual dimensions (cp. Fig. 2.1).

Ta = T1 + T2 + ... + Tk =

T .
i i =1

(3.1)

This instance is called as Arithmetic Tolerancing or Worst Case Tolerancing. Other names for this procedure are: Maximum-Minimum Method or Method of Absolute Interchangeability . Remark 1: Equation 3.1 is valid only when all the tolerance zones of the individual dimensions are symmetric to the respective individual nominal dimensions, i.e., nominal dimension and midpoint of the tolerance interval are the same. Otherwise, the lower and upper allowances are to be given consideration in accordance with the contributory direction of the dimensions (cp. Fig. 2.1). The tolerance zone of the assembly dimension is then, in general, also asymmetric to the nominal assembly dimension. Remark 2: The term absolute interchangeability of a part denotes its property that, without any limitation to the aforementioned design criteria (function, reliability, ...) of the assembly, the part must be able to be replaced by any other part from the same production batch.

-7-

Example for arithmetic tolerancing: Given are five individual parts with dimensions and tolerances. Required are the assembly dimension and its tolerance after assembly (cp. ).

Fig. 3.1: Tolerance stack example with nominal dimensions and tolerances

Fig. 3.2: Tolerance chain graph Schematic representation of the nominal dimensions considering the contributory direction

-8-

Calculation of the arithmetical assembly tolerance with the help of the calculation sheet. It is to be noted that in columns 5 and 6 prefixes and indices of the allowance change when the nominal dimension is negative (according to the contributory direction). For the nominal dimension -23.8 the AL = -0.02 becomes AU' = +0.02. 1 2 3 4 5 6 7 8

Arithmetic addition of the tolerances Dimension N Allowances according to drawing AL A B C D E +44.8 -23.8 -3.5 -8.7 -8.7 +44.8 -44.7 +0.1 s -0.07 AsL +0.05 AsU 0.12 Ta +0.09 Cs -0.02 -0.02 -0.01 +0.01 +0.02 +0.02 -0.01 -0.02 -0.02 AU +0.02 Allowances accord. to contrib. direct. AU' AL' -0.02 +0.02 +0.02 +0.01 Ti C

-9-

## 4. Theoretical Basis for Statistical Tolerancing

The calculation procedures of Statistical Tolerancing are based essentially upon three basic principles which will be presented in the following sections: the central limit theorem of statistics the law of error propagation the distribution of characteristic values within a tolerance zone (distribution of characteristics and tolerance)

## 4.1 Central Limit Theorem of Statistics

When a quality control chart from a process, which is controlled by SPC, is observed it becomes immediately evident, that parts of a mass production are never totally identical and the individual values of variables always deviate, more or less, from the desired target value (nominal value). This becomes easily understandable, when it is explained with an example, say a lathe, as to what lot of errors and causes for errors can have an impact on the process result (cp.  and ): - combination of measuring and drive system - deviation in scale - bearing of scale - deformation due to insufficient rigidity of machine - distortion of workpiece - elastic deformation of workpiece and cutting tool - wear of bearing - inherent vibrations (resonances) of machine - variations in speed and feed rate. These are only a few examples of influence factors which can be assigned to the machine. Apart from this, other external influences are also responsible for the process result (characteristic values of manufactured parts), which can be summarised, together with the aforementioned, into 5 groups: Man, Machine, Material, Method, Environment. Allocation of an influencing factor to one of these groups is, however, not precise in every case. If we once set aside processes with systematic changes of the process average, such as trend and between-batch variation, a process analysis often shows a distribution of characteristic values, which can be described at least approximately by the Gaussian normal distribution. This is a consequence of the central limit theorem of statistics. Expressed somewhat in a free language, it means: due to random interaction of many influence quantities (addition of random variables), the target quantity obtained is approximately normally distributed.

- 10 -

With this abstraction, which includes consideration of a characteristic of a part as a random variable, the phenomenon of deviation experienced in practice in production, viz., actual dimension differing from the desired nominal dimension, can be described mathematically. The fact defined by the central limit theorem of statistics may be explained by a simple example. The result of a throw with a regular dice is a random number x , which can assume the values 1, 2, 3, 4, 5 and 6. The probability that a 6 is thrown is 1/6. The same applies to all other values. The respective probability function f ( x ) can thus be represented as follows:

Fig. 4.1: Probability function of a regular dice. Pure mathematical calculation results in the value of 3.5 as the average of many throws. If the total score from a throw of two dice is the considered random variable, the representation of the probability function of this random variable shows a triangular form (Fig. 4.2).

Fig. 4.2: Probability function of the random variable: Total score from a throw of two dice

- 11 -

The least possible result is 2, the maximum possible result is 12. The probability of getting a total score of 7 is obtained by dividing the number of all combinations yielding a total score of 7 by the number of all possible combinations (36). There are 6 possibilities to get the total score 7 and they are: (1,6), (2,5), (3,4), (4,3), (5,2), (6,1), and in all 36 possible results from the random process considered here: (1,1), (1,2), ..., (1,6), (2,1), (2,2), ..., (2,6), ..., (6,6). Thus it is: f ( 7 ) =

6 1 = . 36 6

The random variable total score as results from a throw with 4 dice has a probability function, whose representation is already very similar to the curve depicting the density function of the normal distribution (continuous line).

Fig. 4.3: Probability function of the random variable: total score as results from a throw with 4 dice The mean of this random variable is 4 35 = 14 . Thus we obtain, as random . result, values which are often very close to this mean and only occasionally big deviations from the mean, for example, the value 24, which corresponds to the throw result (6, 6, 6, 6). Said otherwise, to arrive at this result each of the four random variables (dice) must assume a value (6) which differs very much from the mean 3.5. The probability for this result is very small:

1 < 1% . 6

- 12 -

This principle is of basic importance for the theory of Statistical Tolerancing. Statistical Tolerancing makes use of the fact, that in a random combination of a sufficiently big number of individual parts (elements) positive and negative dimensional deviations neutralize each other. The result from the example shown above can be applied suitably to the properties of a dimension chain. In order that the assembly dimension of k parts assumes a value which deviates very much from the nominal assembly dimension, actual dimensions of all k individual parts must simultaneously be at their respective tolerance limit. The probability for this situation is however very small for a sufficiently big number of k . Assuming a random selection of parts in assembly, the distribution of the random variable assembly dimension is approximately a normal distribution. Figure 4.4 should make this point clear.

Fig. 4.4: Schematic representation of an assembly consisting of 4 parts whose individual dimensions are subject to a rectangular distribution. In all 66 assemblies are arranged in sequence. The assembly dimension or the clearance which is referred to the top edge (broken line) is approximately normally distributed.

- 13 -

The figure shows schematically the interaction of four individual dimensions, having identical tolerance and following rectangular distribution, for an example of 66 assemblies in all, which are arranged here in sequence. It is seen, that the clearance N s (broken line) drawn at the top edge shows big deviations from the average value only seldom. The range 3 of the normal distribution, approximately resulting for the assembly dimension, is obviously somewhat smaller than the arithmetically calculated tolerance, which is shown by the width of big rectangle. The randomness of combination of parts is only one of the many conditions (see Section 6), which must be satisfied, so that this feature of statistics is made use of with advantage for determining tolerances.

Remark: Based on the dice example, it was shown as to how the probability function of a sum of random variables is determined, which can only assume values of whole numbers (1,...,6). This mathematical operation is also called as convolution. The convolution thus denotes the process of linking the probability functions of random variables. This can be applied also to continuous variables, i.e., such variables which can assume any real value. Linear dimensions of parts can be considered as continuous variables, because they can assume any interim value on a scale. In the case of measuring instruments with analogue display, such as slide caliper rule, micrometer screw or indicating precision gauge, this situation is obvious, in contrast to those with digital display. Strictly speaking, a linear dimension can always assume discrete values only, whose minimum distance is limited by the resolution of the measuring instrument or the measurement procedure.

- 14 -

## 4.2 Law of Error Propagation

The Gaussian law of error propagation describes how the measurement errors of several independent measurands xi affect a target quantity z , which is calculated in accordance with a functional correlation

z = f ( x1 , x 2 ,..., x k ) :
z z z 2 2 s12 + s2 + ... + sk . s x1 x2 xk
2 2 2 2

(4.1)

(4.2)

z is, then, generally only an indirectly measurable quantity. For example, the area of a rectangle is determined by measuring the length of sides and multiplying the results of measurement with each other. The accuracy with which z can be indicated depends on the accuracy of the measurands xi . In general, the average xi and the standard deviation si of a sequence of measurements are given for each respective xi .
In the above expression

z denotes the partial derivatives of the function z xi with respect to the variables xi . They are to be calculated at xi .
The derivation of the formula for sz , includes the expansion of z in a Taylors series and overlooks the terms of higher order. Because of this reason, instead of the equals sign the wave symbol is used in the formula, indicating that the two terms are approximately equal. The application of the law of error propagation is understood easily with an example. We consider two Ohmic resistors, on which a series of measurements were carried out. As a result of this measurement series, the mean and the standard deviation are given in each case, the latter being a measure for the average error:

R 1 = 47 0.8

R 2 = 68 11 . .

- 15 -

## In the event of parallel connection, the total resistance R is calculated as follows:

R=

R1 R 2 R1 + R 2
47 68 = 27.8 . 47 + 68

R=

To calculate the corresponding average error, first it is necessary to determine the partial derivatives of R with respect to R1 and R 2 :

R = R1

(
(

2 R2

R1 + R2

)
)

( 47 + 68 )
47 2

68 2

= 0.35

2 R1 R = R2 R1 + R 2

( 47 + 68 )

= 0167 . .

The average error of the total resistance is obtained by substituting these expressions in the law of error propagation:
2 s R = 0.35 0.82 + 0167 2 11 2 = 011 . . .

sR = 0.33 .
Accordingly the result is:

R = 27.8 0.3 .
The law of error propagation becomes simple to present, when the function f , which defines the link of the independent variables x i is a sum. Then the partial derivatives are all equal to one and we find:
2 2 s z2 = s12 + s 2 + ... + s k .

(4.3)

Accordingly, for the variances of the respective underlying (parent) populations the following is valid:
2 2 = 1 + 2 + ... + 2 . z k 2

(4.4)

- 16 -

This means, the variance of an indirect measurand, which is determined by addition of independent individual measurands, is equal to the sum of the variances of these individual measurands. Hint: The aforementioned equation is identical with a general relationship existing between the variances of several independent random variables. It can be derived, also totally independent of the law of error propagation, on the basis of statistical computation rules (see, for example  p. 155). Example: If the two resistors are connected in series, we find:
2 2 2 s R = s R 1 + s R 2 = ( 0.8 ) + (11 ) . 2 2

s R = 1.36

thus

R = 115 14 . .

## 4.3 Distribution of Characteristics and Tolerance

As per DIN 7182, actual dimensions of parts must be within the respective tolerance ranges. However, there is no stipulation for the distribution of the characteristic values in a tolerance zone. Based on the assumption of the most unfavourable case, that all values of the characteristic lie at the limit of the tolerance range, the need for arithmetic tolerancing was derived in Section 3. Instead of this, in the scope of the calculation procedures of Statistical Tolerancing it is assumed that the values of a characteristic are subject to a certain distribution within the tolerance range, bounded by the limiting values. On the other hand, a distribution is characterised, in general, by the parameters mean and standard deviation. Therefore, it is necessary to consider in detail the relationship between the tolerance of a characteristic and the standard deviation of specific distributions. We confine ourselves, in the following sections, to discussions on the normal, triangular and rectangular distribution. Apart from them, the literature also deals with a trapezoidal distribution having different slopes of flanks. (cp. , p. 153).

- 17 -

## 4.3.1 Variance of a Rectangular Distribution

As indicated already in Section 4, we consider in the following the actual dimension of a parameter of a part merely as random variable X . The probability, that the variable X assumes a value x in the range between x and x + d x , is equal to the product f ( x ) d x . f ( x ) is the probability density function. The probability density function for an individual dimension, whose values are distributed uniformly over the entire tolerance range bounded by G L and G U ( T = G U G L ), has the value f ( x ) =

G L + GU 2

(4.5)

## Fig. 4.5: Probability density function f ( x ) of a rectangular distribution

The graphical representation of this function explains the term rectangular distribution. The area of the rectangle corresponds to the probability, that X assumes any value x between G L and G U . It has the value one. The variance

+

( x )

f ( x ) dx .

(4.6)

## As in the case considered here, f ( x ) is equal to zero outside the interval [ G L ,

G U ], it is enough to carry out the integration over this interval, for calculating

=
2

GU

GL

G L + GU x 2

T2 1 . dx = T 12

(4.7)

- 18 -

## 4.3.2 Variance of a Triangular Distribution

The density function of the triangular distribution in the interval [ G L , ] is given by:
f ( x) = 4 x GL . T2

(4.8)

The area of the triangle corresponds to the probability, that X assumes any value x between G L and G U . Its value is equal to one. As the base of this triangle corresponds to the width of the rectangle (cp. rectangular distribution), f ( x ) must have the value 2/T at the position
= G L + GU . 2

(4.9)

Fig. 4.6: Probability density function of a triangular distribution Because of the symmetry of the density function, it is enough to determine the integral in the limits G L and and to multiply by two, for calculating the variance of the triangular distribution.

= 2

GL

(x )
2

4 x GL T2

dx

(4.10)

T2 The result is = . 24

(4.11)

- 19 -

5. Statistical Tolerancing
In the subsections of Section 4, a few theoretical relationships are explained, which form the basis of Statistical Tolerancing. Consequently, fulfillment of the conditions associated with them is an important aspect, which is decisive for the practical applicability of Statistical Tolerancing. However, for the time being, let us set aside this problem. First of all, we consider in the following a dimension chain normally distributed individual dimensions. If the individual interpreted formally as a random variable, then the variance of dimension 2 which is once again a random variable as sum of s
2

## consisting of dimension is the assembly these random

variables, can be calculated as the sum of the variances i of all individual dimensions, according to the law of error propagation, as per section 4.2:

s2 =

i =1

i2 .

(5.1)

In order that this equation can be converted into a relation between the tolerances of the individual dimensions Ti and the tolerance of the assembly dimension Ts we need to know the ratios

Ts s

and

Ti i

For the triangular and rectangular distribution of individual dimensions, such relations exist already in the form of equations 4.7 and 4.11. They were derived under the assumption that the distribution occupies the tolerance range fully, just as it is shown by the figures 4.5 and 4.6. In case of the normal distribution we use the approach Ti = 6 i , i.e., the individual tolerance corresponds to the 3 -range of the distribution of the individual dimension. In practice, this information could be obtained, e.g., from the process capability index C p of a process controlled with SPC:

Cp =

T . 6

(5.2)

Our approach Ti = 6 i corresponds, as such, to the assumption C p = 1. According to the central limit theorem of statistics random combination of sufficiently many individual dimensions produces an assembly dimension, which is approximately normally distributed, independent of the distributions of the individual dimensions.

- 20 -

Ts s

## . This is determined based

on the acceptable fraction of assemblies, whose dimensions may lie outside the specified tolerance zone N s

Ts 2

. (5.3)

## For reasons of simplicity we use Ts = 6 s .

The permissible proportion nonconforming is p 0.27 % in this case. Thus, if Ti = 6 i and Ts = 6 s are used in the equation s2 = (Equation 5.1), we find:

i =1

2 i

2 N

2 TN

36

Ti 2 36

i =1

TN =

T
i =1

## T12 + T22 + ... + Tk2 .

(5.4)

Remark: To identify the distributions that have been taken as a basis for the individual dimensions, the tolerance of the assembly dimension T is provided with an index: N for normal distribution, D for triangular distribution and R for rectangular distribution. The tolerance of the assembly dimension of a dimension chain, consisting of normally distributed individual dimensions is, thus, calculated from the sum of squares of the individual tolerances. Therefore, this is also called as quadratic tolerancing. If the individual dimensions are distributed according to a triangular or a rectangular distribution, then, by using the variances of these distributions in equation 5.1, calculated as per 4.3.1 and 4.3.2, we find once again with the assumption Ts = 6 s :

2 D

T2 = D = 36

Ti 2 24

i =1

TD =

6 24

## T12 + T22 + ... + Tk2

(Triangular distribution) (5.5)

and

2 R

TR2 = = 36

Ti 2 12

i =1

TR =

6 12

## T12 + T22 + ... + Tk2 .

(5.6)

(Rectangular distribution)

- 21 -

If the tolerances of the assembly dimensions T N , TD and TR are compared with each other, then the ratios 1 : 1.2 : 1.7 become apparent. (5.7) Example 5.1: For the individual dimensions given in the following table, the statistical tolerance for the assembly dimension should be calculated and compared with that one calculated arithmetically. Dimension Individual nominal No. dimension Ni 1 2 3 4 3 8 7 14 Allowance Ai 0.01 0.01 0.01 0.01 Tolerance Ti 0.02 0.02 0.02 0.02

Normal:

TN =

i =1

Ti 2 =

## 4 0.02 2 = 2 0.02 = 0.04

Triangular:

TD =

6 24

i =1

Ti 2 =

6 2 0.02 = 0.05 24

Rectangular:

TR =

6 12

i =1

Ti 2 =

6 2 0.02 = 0.07 12

Arithmetical: Ta = 4 0.02 = 0.08 As a result of Statistical Tolerancing, in this example a smaller value is obtained for the assembly tolerance compared to the arithmetically calculated tolerance Ta . The difference is, obviously, maximum for normally distributed individual dimensions (cp. ratios 5.7). Remark: With the assumption Ts = 8 s , corresponding to p 64 ppm, had the value TR = 0.9 been the result for rectangular distribution of the individual dimensions, which is greater than Ta . Accordingly, it is not sensible to use a greater ratio

Ts s

## , just to obtain, by mere calculation, a smaller fraction of

nonconforming assemblies.

- 22 -

If the actual individual dimensions are within their tolerance ranges, then the fraction nonconforming with respect to the arithmetically determined tolerance, is equal to zero. This result is plausible, when the three distributions under consideration are represented in graphical form.

Fig. 5.1: Graphs of the density functions of Normal, Triangular and Rectangular distribution Obviously, in case of normal distribution, the individual values concentrate mostly around the midpoint of the tolerance zone. The gain from statistical tolerancing, compared to arithmetic tolerancing, is the most with normally distributed individual dimensions. If the assembly dimension is assumed to be given, then by inverting the above calculation steps, it can be shown, that under specific conditions and through application of Statistical Tolerancing the individual tolerances can be selected significantly greater (looser) than those resulting from arithmetic tolerancing. The aforementioned considerations make the special advantages of Statistical Tolerancing clear: the smaller assembly tolerance, resulting from the given tolerances for individual dimensions, means for example higher functional reliability or longer life; the greater tolerances for the individual dimensions resulting from a given assembly tolerance lead in general to higher manufacturing reliability or more economic production. Naturally, the comparison is always with reference to arithmetic tolerancing. Statistical Tolerancing bases itself, however, on a number of restrictive conditions, which have not been listed explicitly in the foregoing sections, but have been assumed as complied with without exception. These prerequisites are listed in detail and discussed in Section 6.

- 23 -

## 5.1 The Reduction Factor

The ratios as per equation 5.7 as well as the results of example 5.1 show clearly the advantages of Statistical Tolerancing, in principle, for the special cases of individual dimensions which are subject to the normal, triangular or rectangular distribution. However, it shall not be overlooked, that with the selection of the approach Ts = 6 s (cp. equation 5.3) a permissible proportion of assemblies is accepted ( p = 0.27 % ), whose dimensions lie outside the tolerance zone (fraction nonconforming). Then, there is no unrestricted interchangeability within the assemblies any more. Further, four components have been assumed for the calculations in the scope of the example. The explicit results are, therefore, valid only under the assumptions k = 4 and p = 0.27 % . The resultant reduction in assembly tolerance Ts , which can be achieved from Statistical Tolerancing, compared to the arithmetically calculated assembly tolerance Ta , can be described in form of the reduction factor r and can be determined, in general, for the given special distributions, as long as all k individual dimensions follow, in each case, the same distribution and all individual tolerances Ti are equal:

Ts = r Ta .

(5.8)

Instead of the special option Ts = 6 s corresponding to p = 0.27 % (twosided), we substitute in equations 5.4, 5.5 and 5.6 Ts = 2 u 1 p / 2 s and use the number k of combined individual dimensions.

2 u 1 p / 2 is the random variation range of the approximately normally distributed assembly dimension, in which (1 p ) % all values lie.
From this follows for: a) normally distributed individual dimensions (assumption Ti = 6 i ):

T N = 2 u 1 p / 2 N = 2 u 1 p/ 2
u1 p/ 2 3

i =1

Ti 2
36

TN =

k Ti

r=

TN Ta

u1 p / 2 3

k Ti k Ti

u1 p / 2 3 k

(5.9)

- 24 -

TD =

2 u1 p / 2 24

T
i =1

u1 p/ 2 6

k Ti

r=

TD Ta

u1 p/2 6

k Ti k Ti

u 1 p/2 6k

(5.10)

## c) individual dimensions being subject to a rectangular distribution:

TR =

2 u1 p / 2 12 =

T
i =1

u1 p / 2 3 u1 p /2 3 k

k Ti

r=

TR Ta

u1 p / 2 3

k Ti k Ti

(5.11)

Accordingly, the reduction factor r will be the smaller, the higher the proportion p (acceptable fraction nonconforming) and the greater the number k of individual dimensions. The following table shows a few values for r , which have been calculated under the assumption p = 0.27 % , u 1 p / 2 = 3 .

Number of individual dimensions k Distribution of the individual dimensions Normal distribution Triangular distribution Rectangular distribution 4 0.50 0.61 0.87 6 0.41 0.50 0.71 8 0.35 0.43 0.61

Table 5.1: Reduction factor r for different distributions and numbers of individual dimensions of a dimension chain

- 25 -

It is obvious, that Statistical Tolerancing can yield, depending upon the distribution of the individual dimensions, assembly tolerances which are considerably smaller (tighter) than that resulting from arithmetic tolerancing. Difficulties, concerning applicability of Statistical Tolerancing and making use of the theoretically possible advantages in practice, grow because of the need to know exactly the distributions which are underlying the individual dimensions and the stability of the parameters affecting the location and the variation of these distributions (cp. section 6). As a way out, it is often suggested in the literature, to assume basically a rectangular distribution as model distribution for the individual dimensions. With this there is no need for an expensive monitoring of process (cp. section 7), but instead, one must be satisfied with a comparatively small advantage (higher reduction factor). The following figure shows the dependence of reduction factor r upon k (the number of elements in dimension chain) and nonconforming fraction p for individual dimensions following a rectangular distribution.

Figure 5.2: Reduction factor r depending upon k (number of elements in the dimension chain) and p (fraction nonconforming; proportion of assemblies exceeding the specification limits for the assembly dimension) for individual dimensions which are subject to a rectangular distribution.

- 26 -

## 5.2 Calculation of the Statistical Assembly Tolerance

With the use of the reduction factor r , calculation of the statistical assembly tolerance becomes very simple. Under the assumption, that individual dimensions follow a rectangular distribution,

r=

u1 p / 2 3 k

(5.11)

## 5.3 Calculation of Individual Tolerances for a Given Assembly Tolerance

Also here, let us assume that the individual dimensions are subject to a rectangular distribution. Because of TR = r Ta and Ta = k Ti is TR = r k Ti and Ti =

TR rk

If we apply this relationship to Example 5.1 and use for TR the assembly tolerance 0.08 mm, then under the same conditions as in 5.2 ( p = 0.27 % , k = 4 , individual tolerances are of equal size):

Ti =

TR rk

## 0.08 mm = 0.023 mm . 0.866 4

The resultant individual tolerance Ti is thus somewhat greater than that is obtained with mere arithmetical calculation:

Ti =

Ta k

0.08 mm = 0.02 mm . 4

- 27 -

## 5.4 The General Case

The special cases, which were considered so far, wherein all individual dimensions have the same distribution and the same tolerance, will come across in practice really very seldom. In most of the cases here, it is an assembly consisting of same parts (same drawing dimensions, for example, transformer sheets). In general, the designer has to take into account individual tolerances which are different from each other and the distributions followed by the individual dimensions differ. This general case is covered already in the discussions so far. According to 5.1, 2 u 1 p/ 2 is that random variation range of the approximately normally distributed assembly dimension, wherein (1 p ) % of all values are contained. After having selected p , the assembly tolerance can be calculated with Ts = 2 u 1 p/ 2 s , where s is defined by the law of error propagation
2 (equation 4.3 or 4.4): Ts = 2 u 1 p / 2 1 + 2 + . . . + 2 . 2 k

The i are substituted on the basis of actually available tolerances and distributions of individual dimensions (information from production). If for example

T1 4

T2

24
,

12

T4 6 T5 5

## then the tolerance for the assembly dimension results in

Ts = 2 u 1 p /2

2 T2 T T1 + 3 + 24 12 4

T4 T5 + + . 6 5

- 28 -

## 5.5 Process Capability Indices

The process capability index C p k is a measure for the variation of a characteristic within the tolerance zone, considering the average position of the process.

C pk =

D 3

## D is the minimum distance of the mean of the distribution to one of the

specification limits G U or G L respectively (cp. Fig 5.3).

## Fig. 5.3: Representation to illustrate C p k

Assuming that the process is centred, i.e., the process average coincides with the midpoint of the tolerance interval (thus D =

Cp : Cp = T . 6

T ), then C p k is equal to 2

We will consider in the following the impact of statistics on the C p value of the assembly dimension when assembling several parts whose characteristics follow the same distribution.

- 29 -

The C p value of an assembly dimension is, of course, not a process capability index in the original sense. However, it can be used as a formal variable, for example, to assess the functional reliability of the assemblies. We designate the C p value of the i-th individual characteristic of a dimension chain with C p

( ) , and the C
i

( ):
s

(C )
p

Ti 6i

(C )
p

Ts 6 s

## Now, if for Ts the arithmetically calculated assembly tolerance Ta = k Ti is used and s =

expression for C p

( )
=

## k i , which is based on equation 5.1, is substituted in the

s

, then results:

(C )
p

Ts 6s

k Ti 6 k i

k Cp .
i

( )

Thus, when arithmetically tolerancing a dimension chain having k elements, the C p value of the assembly dimension is greater by a factor of k , than the C p value of the individual dimension. Therefore, it is possible, that the C p value of the assembly dimension exceeds a minimum value for C p , although this is not reached by the C p values of the individual dimensions. Contrarily, the value

(C )
p

Ts

(C )
p

Ts 6 s

6s 6s

= 1.

- 30 -

## 6. Prerequisites for Statistical Tolerancing

As pointed out already a number of times in the text so far, application of Statistical Tolerancing assumes a number of prerequisites as fulfilled, which are once again referred to in this section and subjected to a critical scrutiny as for their possibility of fulfillment and checking in practice. The prerequisites in detail (cp. Supplement to QSP 0407 ): 1. The individual dimensions are uncorrelated random variables. It is conceivable, that two or more parts of an assembly are the same (same drawing dimensions), especially if the assembly is symmetric. If these parts are produced directly in succession then it is very probable, that a current deviation of the process average (process-inherent trend, betweenbatch variation) leads to a deviation of the actual dimensions of these parts in the same direction. If such parts go to the assembly station in an ordered sequence (e.g. transfer line or sequential interim storage in boxes), then there is also matched or sequential grouping of assembly with great probability. In this case, a statistical compensation (neutralization) of dimensional deviations in the sense of the Central Limit Theorem is not at all possible or possible only with constraints. 2. The dimension chain is linear. This prerequisite presumes, of course, the knowledge of the interaction of the individual parts with respect to the assembly dimension. In case of nonlinear dimension chains, the functional correlation of the individual dimensions must be known and be able to be described through analytical functions. The statistical tolerancing of the assembly dimension of a linear dimension chain uses the law of error propagation in its simple form - equation 4.3 or 4.4. In case of nonlinear correlations, their partial derivatives are to be given consideration as per equation 4.2. If the nonlinear correlation is known or describable only approximately, it is to be borne in mind that approximation errors can have strong influence because of the law of error propagation. We underline the fact, that nonlinear dimension chains can lead to complicated applications of the law of error propagation (cp. Section 8). 3. The dimension chain consists at least of four links. As the figures 4.2 and 4.3 show, on the basis of the discrete random variable total score from a throw of k dice, the sum of k random variables leads, from k = 4 and above, to a distribution, which can sufficiently good be approximated by the normal distribution. The combination, exclusively of extreme actual dimensions, is sufficiently improbable only from k = 4 onwards.

- 31 -

The paraphrases sufficiently good and sufficiently improbable are indeed as vague as the formulation permissible fraction nonconforming. Such phrases are always closely linked with the assessment of assemblies from consideration of their design such as functional reliability, reliability (lifetime) or economics of production. 4. The individual tolerances are of the same magnitude Expressed in simple terms, the statistical compensation of negative and positive deviations of actual dimensions from nominal dimensions can function only when the random variation ranges of the actual dimensions and therewith also their tolerances are approximately equal. It is shown in [l] by simulation that with growing difference in size of tolerances in a dimension chain the effect of narrow distributions on the total distribution (distribution of the assembly dimension) gets diminished. With a ratio between sizes of 5:1, the effect of the smaller tolerances is highly suppressed (see , page 63). The prerequisite includes the requirement that the distribution of the individual dimension makes the maximum use of tolerance concerned. If the tolerance is only used up, because a small distribution due to process trend drifts systematically within the tolerance range, then Statistical Tolerancing may not be applied. In this case arithmetical tolerancing is to be resorted to. 5. The distributions of the individual dimensions of the dimension chain are known The condition of knowing the stability of distribution of individual dimensions over time is ultimately a condition relating to the properties of the production processes concerned. Accordingly these must be controlled as for average position and variation and must always be centred on the midpoint. The distribution of a characteristic of a part never follows exactly an ideal distribution as observed in Section 5. Apart from this, in a concrete case an estimation of the parameters of a distribution on the basis of process data (random samples) is always affected with an error dependent upon the sample size. In the following we quote, without comment, from a few literature (see bibliography) published on the theme Statistical Tolerancing, which deal with the problems referred to above. (Hint: Faithful translations of the original quotations. The quoted literature is only available in German.) [l] P. 29: To establish as called for in the beginning a close association between tolerance and type of distribution for Statistical Tolerancing as a precondition information is necessary on the distribution of dimensions within the tolerance range. It is necessary, that these conditions are checked carefully, because deviations in production from the assumptions made for calculation get themselves magnified in the convolution process, and because these deviations influence the total distribution, especially the important peripheral zones.

- 32 -

 P. 34/35: It is valid in general, that special distributions are always to be given consideration in statistical error propagation, because they possibly influence very much the properties of the total distribution. [l] P. 69: The distribution of the counterparts influences, during convolution, very much the propagation of a distribution deviating from the standard. [l] P. 37/38: To assess the stability over time, the change of statistical data of a production process is studied by one or many repetitions of a measurement series, ... While evaluating, it is noticed that only 2 of 17 production processes show total concurrence in x and s. [l] P. 38: The same result occurs if a process is governed by a slow trend (i.e., takes long time to become apparent), which does not show itself up in full within one measurement series. Then, it is very difficult to assess a probable distribution of measurement values. [l] P. 39: So, also the constant drifting of a dimension as a result of shift in cutting edge, the Trend, is inconsistent in parts in terms of steadiness and speed. These influences, which are not purely random, but cannot be excluded, lead to the observed result that the distributions and other statistical data show a wide variability and their reproducibility is low.  P. 113: As has been said already, by giving consideration to the type of distribution, certainly we gain by way of maximum possible tolerance range for individual dimensions and thereby ease production; but with this advantage something is also lost, because there must be constant vigilance to see that prerequisites remain valid; one must monitor the distribution and control the process.  P. 138: It can be seen from this example, that under the assumption, the production is known, significant ease in production can be realized.

## 7. Statistical Monitoring of the Individual Dimensions

The prerequisites for Statistical Tolerancing call for consistency over time of the characterisitcs of the distributions and concurrence of average of distribution with the respective midpoints of tolerance zones. The assumption, that is occasionally found in the literature (see for example  p. 67), viz., these conditions can be guaranteed by Statistical Process Control (SPC) of concerned production process is correct only with its own limitations. We wish to make this point clear in this section. The control charts, maintained within the scope of SPC, enable to recognize a shift in location and/or a change in variation of the distribution under observation at the earliest possible time.

- 33 -

It would all the more be extremely uncomfortable, if a control chart would already respond at small deviations, for example, of sample average from centerpoint C; as such small deviations can result very much from random nature of sampling; and consequently a process which is absolutely stable centrally would be interfered with unnecessarily. According to this, a quality control chart which is functioning in an ideal way should respond with 100% certainty only when the prevailing change in process actually exceeds a given limit and it should not respond with the same certainty otherwise. A real control chart does not possess this property by way of an on/off switch. Real shifts in average position of the characteristic distribution are indicated only with a certain probability. This probability is the greater the more the actual average position t differs from the target value C (or ) (see Fig. 7.1). This action probability as a function of shift in average is described by an action characteristic. The course of this curve depends upon the type of control chart and the sample size n opted for (see  and ). Figure 7.1 shows the action characteristics of an average chart with sample size n as parameter.

Fig. 7.1: Action characteristics of the average chart; sample size n as parameter For the case, where a shift in average of a process under control around 1 should be shown with a probability of 90%, then a sample size of about n = 15 would be necessary. In practice, however, a sample size of n = 5 is often used. The inspection expenses thus would have to be tripled.

- 34 -

## 8. Generalisation for Arbitrary Dimension Chains

It is obvious, that all the cases and examples considered in sections 5.1 to 5.3 are always special cases which can be represented and calculated in the simple form shown only under highly restrictive assumptions. In the general cases, where - any distribution of the individual dimensions, - any combination of these distributions, - any ratio for the magnitude of the individual tolerances, and - a nonlinear dimension chain should be permitted, Statistical Tolerancing would be possible only with the help of computer. An appropriate computer program would have to be able to determine the probability density function of the assembly dimension, which arises from a known functional relationship of individual dimensions following in each case known distributions, by calculation of the so-called convolution integral and to determine again the random variation limits of the assembly dimension, using the density function. If it would be desired to avoid the expensive determination of convolution integrals by computation, which can be represented as closed expressions only for simple analytical functions such as sums and products (of random variables), then it would be necessary to apply the law of error propagation in its general form (Equation 4.2). In this case, however, the analytical or numerical determination of partial derivatives would be necessary. Finally there would be still the possibility, to simulate the combination of distributions of individual dimensions (Monte-Carlo-Methods). Although the aforementioned altematives could be realised in form of computer programs, the practicability of a computer-aided Statistical Tolerancing is, in the general case, not very probable. Despite all the new developments in hardware and software spheres, the production processes, their controllability and sufficiently reliable statistical description of them remain the decisive factors for economic application of Statistical Tolerancing. Further the above concepts can apply always only to calculation of tolerances for assembly dimensions from individual dimensions following given (known) distributions. The inversion of calculation, i.e., calculation of tolerances for individual dimensions from a given tolerance for the asembly dimension as shown for example in 5.3 is not possible in the general case.

- 35 -

9. Bibliography
 F. Bttger: Erzielung von Fertigungsvorteilen durch Anwendung statistischer Gesetze auf die Toleranzberechnung, Dissertation, TH Aachen, 1961 Statistische Matolerierung von Groserienprodukten, Konstruktion 41 (1989) S.313-318 Toleranzfestlegung unter Bercksichtigung statistischer Gesichtspunkte, Qualittskontrolle, 13. Jahrgang (1968) Heft 9, S.111 Toleranzkopplungen in Theorie und Praxis, Feinwerktechnik, Jahrgang 63, Heft 4, 1959, S.133-142 Erstellung eines Rechnerprogramms zur statistischen Tolerierung von Toleranzketten, Diplomarbeit, Uni Stuttgart, 1990

 

H. Rper: E. Schltel:

 

F. Goubeau: K. Blaier:

##       

G. Kirschling: Qualittssicherung und Toleranzen, Springer-Verlag Berlin, 1988 Bosch Schriftenreihe, Heft Nr. 7: SPC DGQ-Schrift Nr. 16-30 Qualittsregelkarten Balakschin: E. Kreyszig: Technologie des Werkzeugmaschinenbaus, VEB Verlag Technik, Berlin, 1953 Statistische Methoden und ihre Anwendungen, 7. Auflage, Gttingen, 1979 DIN 7182, Mae, Abmae, Toleranzen und Passungen, Grundbegriffe, 1986 Bosch Norm N13 B11, Toleranzen und Passungen, Grundbegriffe, 1975 Bosch Norm N13 B94, Toleranzrechnung, 1980 Qualittssicherungsplan 0407, Toleranzehrlichkeit DIN 7186 Statistische Tolerierung, 1974 TGL 19115/01 - TGL 19115/04 (DDR-Standard), Berechnung von Masz- und Toleranzketten, 1983

- 36 -

 

E. Rusch:

## Wirtschaftliches Tolerieren, QZ 16 (1971), Heft 1, S.19-23

H. Trumpold, Ch. Beck: Optimale Toleranzfestlegung unter Bercksichtigung der Makettentheorie und der statistischen Eigenschaften des Fertigungsprozesses, Fertigungstechnik und Betrieb, 21 (1971) H.4, S.242-246 E. Grler: Bercksichtigung der Lage und Form statistischer Verteilungen bei der Tolerierung von Maketten, Wiss. Z. d. Techn. Hochsch. Karl-Marx-Stadt, 21 (1979) H. 2, S.203-210 Lage und Form statistischer Verteilungen bei der Tolerierung von Maketten, Feingertetechnik, 28. Jahrg., Heft 7, 1979, S.314-316 Rationelle Verfahren bei der statistischen Qualittsanalyse in der mechanischen Fertigung, Fertigungstechnik und Betrieb 26 (1976) 3, S.142-145 Die Hufigkeitsverteilung der Summentoleranzen von zusammengebauten Teilen, Werkstattstechnik 55. Jahrg., 1965, Heft 12, S.590-593





E. Grler:



E. Grler:



W. Hanka:



A. Kettmann: Festlegung von Toleranzen nach statistischen Gesichtspunkten, Konstruktion, 13. (1961) Heft 5, S.202-204 K. Stange: Die Bestimmung von Toleranzgrenzen mit Hilfe statistischer berlegungen, Qualittskontrolle 14. Jahrg. (1969) Heft 5, S.57-63 Statistische Tolerierung als Problem der Fertigung und der Prfung, Werkstattstechnik, 55. (1965) H.12 S.585-589 Optimierung der Funktionssicherheit mechanischer Maketten durch Statistische Tolerierung, QZ 16 (1971) H. 12, S.269-271





Stuhlmann, Schmidt:



R. Mller:

- 37 -

10. Appendix
Revision - Normal Distribution Solutions to Excercises The Probability Plot Standard Normal Distribution Table of the Standard Normal Distribution

- 38 -

## Revision - Normal Distribution (Excercises)

Relays were tested for series production start-up. A functionally decisive characteristic is the so-called response voltage. With 50 relays, the following values of the response voltage U resp were measured in volts. 6.2 6.5 6.2 6.0 6.4 6.5 5.5 6.4 5.7 6.3 6.1 5.7 5.8 6.5 6.7 6.3 6.2 5.6 6.2 5.9 5.9 5.9 6.2 5.6 6.6 6.0 6.5 6.1 6.4 6.3 6.0 6.1 5.8 6.1 6.0 6.3 6.6 5.9 6.3 6.0 6.2 6.1 6.0 6.1 5.8 6.4 6.8 6.1 6.6 6.2

a) Calculate x and s with pocket calculator! b) The customer asks for U resp = 6.8 V . Estimate the proportion of relays which does not satisfy the customers requirement (fraction nonconforming)! c) What is the value of response voltage that will be exceeded by maximum 0.1% of all relays manufactured? d) Determine with the help of x , s and the upper specification limit USL = 6.8 V the C pk value of the production!

- 39 -

Solutions to Excercises
To a) Mean x = 615 . Standard deviation s = 0.2998 0.3

To b)

u=

## Table value ( 2.17) = 0.985 Fraction nonconforming p = 1 0.985 = 15 % .

To c)

p = 1 (u) = 0.1%

or:

(u) = 0.999

Read from the table that value of u , at which ( u ) is as close as possible to the value 0.999. We find u = 3.09 . If the transformation equation u = values are substituted, this produces

## x = x + u s = 615 + 3.09 0.3 7.1 . .

Not more than 0.1% of all relays will have a response voltage which is in excess of 7.1 V.

To d)

C pk =

- 40 -

## The Probability Plot

If a normal distribution is being spoken of, one usually associates this term with the Gaussian bell-shaped curve. This curve is a representation of the probability density function f ( x ) of the normal distribution:

f ( x) =

1 2

1 x 2

The normal distribution gives the probability for each value x that the random variable X will assume a value between and x . One obtains the distribution function ( x ) of the normal distribution by integrating the density function given above:
2

( x) =

1 2

1 v 2

dv

## ( x ) corresponds to the area below the Gaussian curve up to the value x .

The graphical representation of this function follows an s-shaped curve. Strictly speaking, we should always think of this curve, when we talk of a normal distribution. If the y-axis of this representation is distorted in such a manner, shaped curve becomes a straight line, a new coordinate system probility paper. Here the x-axis remains unchanged. Because of lation, the representation of a normal distribution in this new system produces a straight line. that the sarises, the this correcoordinate

This fact is made use of to test a given data set for normality graphically. As long as the number of measurement values is sufficiently big, one can determine the relative frequencies of values within the classes of a grouping and draw a histogram. If the corresponding cumulative relative frequencies are plotted versus the upper class limits on probability paper and a sequence of dots is obtained, which approximately lie on a staight line, then it can be concluded that the values of the data set are approximately normally distributed.

- 41 -

## The Standard Normal Distribution

A normally distributed random quantity X with the mean and standard deviation is converted to an equally normally distributed random quantity U through the transformation U = standard deviation is one.

## X . The average of U is zero, and its

This fact is illustrated by the following figure. Here x = 238 and x = 0.5 . .

## 25 23.8 = 2.4 . 0.5

The special normal distribution N(0,1) is called standard normal distribution. The distribution function ( u ) indicates the probability with which the random quantity U assumes a value between and u .

## ( u) corresponds to the area below the Gaussian curve up to the value u .

The total area under the bell-shaped curve has the value one. The values for ( u ) can be taken from the table. For example, we find (2.4) = 0.9918. The proportion of all X , which are greater than 25, corresponds to the proportion of U, which exceed the value u = 2.4. Accordingly it is 1 - 0.9918 0.8%. The following representations show a few examples from which the meaning of the terms and the table values become apparent.

- 42 -

(-u)=1- (u)

## D(u)= (u)- (-u)

- 43 -

u 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14 0.15 0.16 0.17 0.18 0.19 0.20 0.21 0.22 0.23 0.24 0.25 0.26 0.27 0.28 0.29 0.30 0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38 0.39 0.40 0.41 0.42 0.43 0.44 0.45 0.46 0.47 0.48 0.49 0.50

0.496011 0.492022 0.488033 0.484047 0.480061 0.476078 0.472097 0.468119 0.464144 0.460172 0.456205 0.452242 0.448283 0.444330 0.440382 0.436441 0.432505 0.428576 0.424655 0.420740 0.416834 0.412936 0.409046 0.405165 0.401294 0.397432 0.393580 0.389739 0.385908 0.382089 0.378281 0.374484 0.370700 0.366928 0.363169 0.359424 0.355691 0.351973 0.348268 0.344578 0.340903 0.337243 0.333598 0.329969 0.326355 0.322758 0.319178 0.315614 0.312067 0.308538

(-u)

0.503989 0.507978 0.511967 0.515953 0.519939 0.523922 0.527903 0.531881 0.535856 0.539828 0.543795 0.547758 0.551717 0.555670 0.559618 0.563559 0.567495 0.571424 0.575345 0.579260 0.583166 0.587064 0.590954 0.594835 0.598706 0.602568 0.606420 0.610261 0.614092 0.617911 0.621719 0.625516 0.629300 0.633072 0.636831 0.640576 0.644309 0.648027 0.651732 0.655422 0.659097 0.662757 0.666402 0.670031 0.673645 0.677242 0.680822 0.684386 0.687933 0.691462

(u)

D(u) 0.007979 0.015957 0.023933 0.031907 0.039878 0.047845 0.055806 0.063763 0.071713 0.079656 0.087591 0.095517 0.103434 0.111340 0.119235 0.127119 0.134990 0.142847 0.150691 0.158519 0.166332 0.174129 0.181908 0.189670 0.197413 0.205136 0.212840 0.220522 0.228184 0.235823 0.243439 0.251032 0.258600 0.266143 0.273661 0.281153 0.288617 0.296054 0.303463 0.310843 0.318194 0.325514 0.332804 0.340063 0.347290 0.354484 0.361645 0.368773 0.375866 0.382925 - 44 -

u 0.51 0.52 0.53 0.54 0.55 0.56 0.57 0.58 0.59 0.60 0.61 0.62 0.63 0.64 0.65 0.66 0.67 0.68 0.69 0.70 0.71 0.72 0.73 0.74 0.75 0.76 0.77 0.78 0.79 0.80 0.81 0.82 0.83 0.84 0.85 0.86 0.87 0.88 0.89 0.90 0.91 0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1.00

0.305026 0.301532 0.298056 0.294598 0.291160 0.287740 0.284339 0.280957 0.277595 0.274253 0.270931 0.267629 0.264347 0.261086 0.257846 0.254627 0.251429 0.248252 0.245097 0.241964 0.238852 0.235762 0.232695 0.229650 0.226627 0.223627 0.220650 0.217695 0.214764 0.211855 0.208970 0.206108 0.203269 0.200454 0.197662 0.194894 0.192150 0.189430 0.186733 0.184060 0.181411 0.178786 0.176186 0.173609 0.171056 0.168528 0.166023 0.163543 0.161087 0.158655

(-u)

0.694974 0.698468 0.701944 0.705402 0.708840 0.712260 0.715661 0.719043 0.722405 0.725747 0.729069 0.732371 0.735653 0.738914 0.742154 0.745373 0.748571 0.751748 0.754903 0.758036 0.761148 0.764238 0.767305 0.770350 0.773373 0.776373 0.779350 0.782305 0.785236 0.788145 0.791030 0.793892 0.796731 0.799546 0.802338 0.805106 0.807850 0.810570 0.813267 0.815940 0.818589 0.821214 0.823814 0.826391 0.828944 0.831472 0.833977 0.836457 0.838913 0.841345

(u)

D(u) 0.389949 0.396936 0.403888 0.410803 0.417681 0.424521 0.431322 0.438085 0.444809 0.451494 0.458138 0.464742 0.471306 0.477828 0.484308 0.490746 0.497142 0.503496 0.509806 0.516073 0.522296 0.528475 0.534610 0.540700 0.546745 0.552746 0.558700 0.564609 0.570472 0.576289 0.582060 0.587784 0.593461 0.599092 0.604675 0.610211 0.615700 0.621141 0.626534 0.631880 0.637178 0.642427 0.647629 0.652782 0.657888 0.662945 0.667954 0.672914 0.677826 0.682689

u 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10 1.11 1.12 1.13 1.14 1.15 1.16 1.17 1.18 1.19 1.20 1.21 1.22 1.23 1.24 1.25 1.26 1.27 1.28 1.29 1.30 1.31 1.32 1.33 1.34 1.35 1.36 1.37 1.38 1.39 1.40 1.41 1.42 1.43 1.44 1.45 1.46 1.47 1.48 1.49 1.50

0.156248 0.153864 0.151505 0.149170 0.146859 0.144572 0.142310 0.140071 0.137857 0.135666 0.133500 0.131357 0.129238 0.127143 0.125072 0.123024 0.121001 0.119000 0.117023 0.115070 0.113140 0.111233 0.109349 0.107488 0.105650 0.103835 0.102042 0.100273 0.098525 0.096801 0.095098 0.093418 0.091759 0.090123 0.088508 0.086915 0.085344 0.083793 0.082264 0.080757 0.079270 0.077804 0.076359 0.074934 0.073529 0.072145 0.070781 0.069437 0.068112 0.066807

(-u)

0.843752 0.846136 0.848495 0.850830 0.853141 0.855428 0.857690 0.859929 0.862143 0.864334 0.866500 0.868643 0.870762 0.872857 0.874928 0.876976 0.878999 0.881000 0.882977 0.884930 0.886860 0.888767 0.890651 0.892512 0.894350 0.896165 0.897958 0.899727 0.901475 0.903199 0.904902 0.906582 0.908241 0.909877 0.911492 0.913085 0.914656 0.916207 0.917736 0.919243 0.920730 0.922196 0.923641 0.925066 0.926471 0.927855 0.929219 0.930563 0.931888 0.933193

(u)

D(u) 0.687505 0.692272 0.696990 0.701660 0.706282 0.710855 0.715381 0.719858 0.724287 0.728668 0.733001 0.737286 0.741524 0.745714 0.749856 0.753951 0.757999 0.762000 0.765953 0.769861 0.773721 0.777535 0.781303 0.785024 0.788700 0.792331 0.795915 0.799455 0.802949 0.806399 0.809804 0.813165 0.816482 0.819755 0.822984 0.826170 0.829313 0.832413 0.835471 0.838487 0.841460 0.844392 0.847283 0.850133 0.852941 0.855710 0.858438 0.861127 0.863776 0.866386 - 45 -

u 1.51 1.52 1.53 1.54 1.55 1.56 1.57 1.58 1.59 1.60 1.61 1.62 1.63 1.64 1.65 1.66 1.67 1.68 1.69 1.70 1.71 1.72 1.73 1.74 1.75 1.76 1.77 1.78 1.79 1.80 1.81 1.82 1.83 1.84 1.85 1.86 1.87 1.88 1.89 1.90 1.91 1.92 1.93 1.94 1.95 1.96 1.97 1.98 1.99 2.00

0.065522 0.064256 0.063008 0.061780 0.060571 0.059380 0.058208 0.057053 0.055917 0.054799 0.053699 0.052616 0.051551 0.050503 0.049471 0.048457 0.047460 0.046479 0.045514 0.044565 0.043633 0.042716 0.041815 0.040929 0.040059 0.039204 0.038364 0.037538 0.036727 0.035930 0.035148 0.034379 0.033625 0.032884 0.032157 0.031443 0.030742 0.030054 0.029379 0.028716 0.028067 0.027429 0.026803 0.026190 0.025588 0.024998 0.024419 0.023852 0.023295 0.022750

(-u)

0.934478 0.935744 0.936992 0.938220 0.939429 0.940620 0.941792 0.942947 0.944083 0.945201 0.946301 0.947384 0.948449 0.949497 0.950529 0.951543 0.952540 0.953521 0.954486 0.955435 0.956367 0.957284 0.958185 0.959071 0.959941 0.960796 0.961636 0.962462 0.963273 0.964070 0.964852 0.965621 0.966375 0.967116 0.967843 0.968557 0.969258 0.969946 0.970621 0.971284 0.971933 0.972571 0.973197 0.973810 0.974412 0.975002 0.975581 0.976148 0.976705 0.977250

(u)

D(u) 0.868957 0.871489 0.873983 0.876440 0.878858 0.881240 0.883585 0.885893 0.888165 0.890401 0.892602 0.894768 0.896899 0.898995 0.901057 0.903086 0.905081 0.907043 0.908972 0.910869 0.912734 0.914568 0.916370 0.918141 0.919882 0.921592 0.923273 0.924924 0.926546 0.928139 0.929704 0.931241 0.932750 0.934232 0.935687 0.937115 0.938516 0.939892 0.941242 0.942567 0.943867 0.945142 0.946393 0.947620 0.948824 0.950004 0.951162 0.952297 0.953409 0.954500

u 2.01 2.02 2.03 2.04 2.05 2.06 2.07 2.08 2.09 2.10 2.11 2.12 2.13 2.14 2.15 2.16 2.17 2.18 2.19 2.20 2.21 2.22 2.23 2.24 2.25 2.26 2.27 2.28 2.29 2.30 2.31 2.32 2.33 2.34 2.35 2.36 2.37 2.38 2.39 2.40 2.41 2.42 2.43 2.44 2.45 2.46 2.47 2.48 2.49 2.50

0.022216 0.021692 0.021178 0.020675 0.020182 0.019699 0.019226 0.018763 0.018309 0.017864 0.017429 0.017003 0.016586 0.016177 0.015778 0.015386 0.015003 0.014629 0.014262 0.013903 0.013553 0.013209 0.012874 0.012545 0.012224 0.011911 0.011604 0.011304 0.011011 0.010724 0.010444 0.010170 0.009903 0.009642 0.009387 0.009137 0.008894 0.008656 0.008424 0.008198 0.007976 0.007760 0.007549 0.007344 0.007143 0.006947 0.006756 0.006569 0.006387 0.006210

(-u)

0.977784 0.978308 0.978822 0.979325 0.979818 0.980301 0.980774 0.981237 0.981691 0.982136 0.982571 0.982997 0.983414 0.983823 0.984222 0.984614 0.984997 0.985371 0.985738 0.986097 0.986447 0.986791 0.987126 0.987455 0.987776 0.988089 0.988396 0.988696 0.988989 0.989276 0.989556 0.989830 0.990097 0.990358 0.990613 0.990863 0.991106 0.991344 0.991576 0.991802 0.992024 0.992240 0.992451 0.992656 0.992857 0.993053 0.993244 0.993431 0.993613 0.993790

(u)

D(u) 0.955569 0.956617 0.957644 0.958650 0.959636 0.960602 0.961548 0.962475 0.963382 0.964271 0.965142 0.965994 0.966829 0.967645 0.968445 0.969227 0.969993 0.970743 0.971476 0.972193 0.972895 0.973581 0.974253 0.974909 0.975551 0.976179 0.976792 0.977392 0.977979 0.978552 0.979112 0.979659 0.980194 0.980716 0.981227 0.981725 0.982212 0.982687 0.983152 0.983605 0.984047 0.984479 0.984901 0.985313 0.985714 0.986106 0.986489 0.986862 0.987226 0.987581 - 46 -

u 2.51 2.52 2.53 2.54 2.55 2.56 2.57 2.58 2.59 2.60 2.61 2.62 2.63 2.64 2.65 2.66 2.67 2.68 2.69 2.70 2.71 2.72 2.73 2.74 2.75 2.76 2.77 2.78 2.79 2.80 2.81 2.82 2.83 2.84 2.85 2.86 2.87 2.88 2.89 2.90 2.91 2.92 2.93 2.94 2.95 2.96 2.97 2.98 2.99 3.00

0.006037 0.005868 0.005703 0.005543 0.005386 0.005234 0.005085 0.004940 0.004799 0.004661 0.004527 0.004397 0.004269 0.004145 0.004025 0.003907 0.003793 0.003681 0.003573 0.003467 0.003364 0.003264 0.003167 0.003072 0.002980 0.002890 0.002803 0.002718 0.002635 0.002555 0.002477 0.002401 0.002327 0.002256 0.002186 0.002118 0.002052 0.001988 0.001926 0.001866 0.001807 0.001750 0.001695 0.001641 0.001589 0.001538 0.001489 0.001441 0.001395 0.001350

(-u)

0.993963 0.994132 0.994297 0.994457 0.994614 0.994766 0.994915 0.995060 0.995201 0.995339 0.995473 0.995603 0.995731 0.995855 0.995975 0.996093 0.996207 0.996319 0.996427 0.996533 0.996636 0.996736 0.996833 0.996928 0.997020 0.997110 0.997197 0.997282 0.997365 0.997445 0.997523 0.997599 0.997673 0.997744 0.997814 0.997882 0.997948 0.998012 0.998074 0.998134 0.998193 0.998250 0.998305 0.998359 0.998411 0.998462 0.998511 0.998559 0.998605 0.998650

(-u)

D(u) 0.987927 0.988264 0.988594 0.988915 0.989228 0.989533 0.989830 0.990120 0.990402 0.990678 0.990946 0.991207 0.991461 0.991709 0.991951 0.992186 0.992415 0.992638 0.992855 0.993066 0.993272 0.993472 0.993666 0.993856 0.994040 0.994220 0.994394 0.994564 0.994729 0.994890 0.995046 0.995198 0.995345 0.995489 0.995628 0.995763 0.995895 0.996023 0.996147 0.996268 0.996386 0.996500 0.996610 0.996718 0.996822 0.996923 0.997022 0.997117 0.997210 0.997300

u 3.01 3.02 3.03 3.04 3.05 3.06 3.07 3.08 3.09 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 3.20 3.21 3.22 3.23 3.24 3.25 3.26 3.27 3.28 3.29 3.30 3.31 3.32 3.33 3.34 3.35 3.36 3.37 3.38 3.39 3.40 3.41 3.42 3.43 3.44 3.45 3.46 3.47 3.48 3.49 3.50

0.001306 0.001264 0.001223 0.001183 0.001144 0.001107 0.001070 0.001035 0.001001 0.000968 0.000936 0.000904 0.000874 0.000845 0.000816 0.000789 0.000762 0.000736 0.000711 0.000687 0.000664 0.000641 0.000619 0.000598 0.000577 0.000557 0.000538 0.000519 0.000501 0.000483 0.000467 0.000450 0.000434 0.000419 0.000404 0.000390 0.000376 0.000362 0.000350 0.000337 0.000325 0.000313 0.000302 0.000291 0.000280 0.000270 0.000260 0.000251 0.000242 0.000233

(-u)

0.998694 0.998736 0.998777 0.998817 0.998856 0.998893 0.998930 0.998965 0.998999 0.999032 0.999064 0.999096 0.999126 0.999155 0.999184 0.999211 0.999238 0.999264 0.999289 0.999313 0.999336 0.999359 0.999381 0.999402 0.999423 0.999443 0.999462 0.999481 0.999499 0.999517 0.999533 0.999550 0.999566 0.999581 0.999596 0.999610 0.999624 0.999638 0.999650 0.999663 0.999675 0.999687 0.999698 0.999709 0.999720 0.999730 0.999740 0.999749 0.999758 0.999767

(-u)

D(u) 0.997387 0.997472 0.997554 0.997634 0.997711 0.997786 0.997859 0.997930 0.997998 0.998065 0.998129 0.998191 0.998252 0.998310 0.998367 0.998422 0.998475 0.998527 0.998577 0.998626 0.998673 0.998718 0.998762 0.998805 0.998846 0.998886 0.998924 0.998962 0.998998 0.999033 0.999067 0.999100 0.999131 0.999162 0.999192 0.999220 0.999248 0.999275 0.999301 0.999326 0.999350 0.999374 0.999396 0.999418 0.999439 0.999460 0.999479 0.999498 0.999517 0.999535 - 47 -

u 3.51 3.52 3.53 3.54 3.55 3.56 3.57 3.58 3.59 3.60 3.61 3.62 3.63 3.64 3.65 3.66 3.67 3.68 3.69 3.70 3.71 3.72 3.73 3.74 3.75 3.76 3.77 3.78 3.79 3.80 3.81 3.82 3.83 3.84 3.85 3.86 3.87 3.88 3.89 3.90 3.91 3.92 3.93 3.94 3.95 3.96 3.97 3.98 3.99 4.00

0.000224 0.000216 0.000208 0.000200 0.000193 0.000185 0.000179 0.000172 0.000165 0.000159 0.000153 0.000147 0.000142 0.000136 0.000131 0.000126 0.000121 0.000117 0.000112 0.000108 0.000104 0.000100 0.000096 0.000092 0.000088 0.000085 0.000082 0.000078 0.000075 0.000072 0.000070 0.000067 0.000064 0.000062 0.000059 0.000057 0.000054 0.000052 0.000050 0.000048 0.000046 0.000044 0.000042 0.000041 0.000039 0.000037 0.000036 0.000034 0.000033 0.000032

(-u)

0.999776 0.999784 0.999792 0.999800 0.999807 0.999815 0.999821 0.999828 0.999835 0.999841 0.999847 0.999853 0.999858 0.999864 0.999869 0.999874 0.999879 0.999883 0.999888 0.999892 0.999896 0.999900 0.999904 0.999908 0.999912 0.999915 0.999918 0.999922 0.999925 0.999928 0.999930 0.999933 0.999936 0.999938 0.999941 0.999943 0.999946 0.999948 0.999950 0.999952 0.999954 0.999956 0.999958 0.999959 0.999961 0.999963 0.999964 0.999966 0.999967 0.999968

(-u)

D(u) 0.999552 0.999568 0.999584 0.999600 0.999615 0.999629 0.999643 0.999656 0.999669 0.999682 0.999694 0.999705 0.999717 0.999727 0.999738 0.999748 0.999757 0.999767 0.999776 0.999784 0.999793 0.999801 0.999808 0.999816 0.999823 0.999830 0.999837 0.999843 0.999849 0.999855 0.999861 0.999867 0.999872 0.999877 0.999882 0.999887 0.999891 0.999896 0.999900 0.999904 0.999908 0.999911 0.999915 0.999918 0.999922 0.999925 0.999928 0.999931 0.999934 0.999937

Index
Action characteristic 34 Allowance 9, 22 Assembly dimension 5, 7, 13, 20, 22 tolerance 7, 9, 22, 27 Average chart 34 Bell-shaped curve 41, 42 Central limit theorem 10, 11, 20 Clearance 6, 13 Contributory direction 5, 7, 8, 9 Control chart 10, 33, 34 Convolution 14, 35 Density function 12, 18, 19, 23 Dimension chain 5, 6, 13, 20, 35 maximum 6, 7 minimum 6, 7 nominal 6 Distribution function 41, 42 normal 10, 13, 21, 41 rectangular 18, 21 standard normal 42 trapezoidal 17 triangular 19, 21 Fraction nonconforming 24, 25 Variance 18, 19 Gaussian bell-shaped curve 41 Interchangeability 7, 24 Law of error propagation 15, 16 Monte-Carlo-Methods 35 Probability 11 density function 18, 41 function 11, 12 paper 41 plot 41 Process capability index 20, 29 Proportion nonconforming 21 Random variable 11, 12, 13 Random variation range 24, 28 Reduction factor 24, 25, 26, 27 Specification limit 29 Statistical Process Control 33 Tolerance 22 chain graph 5, 8 stack 8 zone 7, 17 Tolerancing arithmetic 7, 8 statistical 10, 20, 35 worst case 7

- 48 -

Robert Bosch GmbH C/QMM Postfach 30 02 20 D-70442 Stuttgart Germany Phone +49 711 811-4 47 88 Fax +49 711 811-2 31 26 www.bosch.com