The Design of a

Fuzzy - Shewhart Control Chart

Jorg Hoppner & Hans Wol

Abteilung Stochastik
Universitat Ulm
Ulm, Germany
Marz 1995
1 Introduction
In this paper we suggest a design for a control chart in the case of vague data.
Such data may arise in form of linguistic data or by imprecise measurements,
which all can modelled by fuzzy - sets. We consider fuzzy - sets as real valued
interpretations of vague data. We leave the traditional approach that we are al-
ways able to assign exactly one real number (ore p-dimensional vectors) to the
outcome of our experiment. Indeed, if we have to meet uncertainty and vagueness
in describing our observations, we should replace the classical dichotomous judge-
ments (we observe "x", x 2 IR) by introducing fuzzy - sets, which themselves can
be described by membership functions. For quality purposes, such an approach
is described in more detail e.g. in Laviolette et al. 14].
Our starting point is based on the pioneering work of L.A.Zadeh 26, 27] on fuzzy -
sets and his extension principle as well as on the concept of a fuzzy random va-
riable introduced by Kwakernaak 11]. The generalization of traditional control
charts to the "fuzzy case" will be studied in this paper for the simple Shewhart
chart. The statistics involved in Shewhart chart techniques are the mean and the
variance of a random sample. Therefore, we need a generalization of such stati-
stics for fuzzy random samples. The corresponding statistics can be found e.g.
in the excellent book of R.Kruse and K.D.Meyer 10], which gives a thorough
discussion of statistical methods for fuzzy - data.
Even for Shewhart charts, we will see that in case of fuzzy - data the stopping rule
will not be as simple as in the traditional case. Simultaneous test problems arise
implying e.g. diculties for an exact evaluation of the operating characteristic
function or the average run length.

2 Fuzzy - sets and some properties

In this section, we want to describe the concept of fuzzy - sets based on the work
of L.A.Zadeh 26, 27]. We give the most important tools of fuzzy - sets needed
for the construction of the fuzzy - Shewhart control chart.

Denition 2.1 :
(i) A fuzzy - set of the real line IR is characterized by its membership function
 , where :
 : IR ;! 0 1] :
We denote by E ( IR ) the class of all fuzzy - sets of IR .

2
 (ii) A fuzzy - set  2 E ( IR ) of the real line is called normal , if
9 t 2 IR :  ( t ) = 1 .
We denote by F ( IR ) the class of all normal fuzzy - sets of IR .

Example 2.2 :
(i) The characteristic function yields a simple fuzzy - set.
Let A  IR be a non-empty subset of IR . Then it holds :
8 1 , if t 2 A
<
IA ( t ) := : 0 , if t 2 IRnA

is a normal fuzzy - set.

(ii)  ( t ) := max f 1 ; ( t ; 2 )2  0 g is a normal fuzzy - set.

Important tools for dealing with fuzzy - sets are their -level-sets and their strong
-cuts, which were discussed by C.V.Negoita and D.A.Ralescu 19].

Denition 2.3 :
Let  2 F ( IR) ,  2 0 1] .
Then dene :
(i)  := f t 2 IR j  ( t )   g as -level-set.
(ii)  := f t 2 IR j  ( t ) >  g as strong -cut.

M.Miakoshi and M.Shimbo 15] discussed the concept of set representations of

fuzzy - sets. This leads to a generalization of the -level-set and the strong -cut
dened by C.V.Negoita and D.A.Ralescu.

Denition 2.4 :
Let  2 F ( IR).
f A j  2 (0 1) g is called a set representation of  if and only if

3
 (i) 0     < 1 =) A  A  IR :
(ii) For all t 2 IR :  ( t ) = sup f  IA ( t ) j  2 (0 1) g  where IA ( t )
denotes the characteristic function of A .

The strong -cut respectively the -level-set are the "smallest", respectively the
"biggest", set representation in the sense of M.Miakoshi and M.Shimbo. We re-
mark that f  j  2 (0 1) g and f  j  2 (0 1) g are both set representations
of the fuzzy - set .
Necessary and sucient conditions that f A j  2 (0 1) g is a set representation
of a fuzzy - set  are given in the following theorem (M.Miyakoshi and M.Shimbo
15]), which is very useful for proofs.

Theorem 2.5 :
Let  2 F ( IR ) and f A j  2 (0 1) g be a system of subsets of IR . Then it
holds :
f A j  2 (0 1) g is a set representation of  if and only if
  A   holds for all  2 (0 1) .

Now, we give an example of a fuzzy - set and the way how to get the set repre-
sentation for a given fuzzy - set.

Example 2.6 :
Let  ( t ) := max f 1 ; ( t ; 2 )2  0 g be a fuzzy - set. Then, we have :
 = f t 2 IR j  ( t ) >  g
p p
= f t 2 IR j 2 ; 1 ;  < t < 2 + 1 ;  g
p p
=)  = ( 2 ; 1 ;   2 < + 1 ;  ) :
in the same way we get :
p p
 =  2 ; 1 ;   2 + 1 ;  ] :

4
The corresponding graphical representation is given below :

 6
1:00 .............
.....................
. . ........... ................ ...........
.....
.... ......................................................................................
.....
0:75 .............. ........ .............
...

..................... ................ ...................

.. . . . . . . . . . . . .
0:50 ....... ............ ............ ........... ......
.......... ............ ............ .......... ...........
0:25 ............ ................ ................ .............. .............
................ ................ ................ .............. .................
0
. . . . . . . . . . . . . . . . . . ..
.. . . . . . . . . . . . . . . . . . . ... - t
1 2 3 4
The extension priciple of L.A.Zadeh oers a powerful tool for generalizing classi-
cal mathematical concepts to the fuzzy case.

Denition 2.7 :(Extension principle of Zadeh, see R.R.Yager 25])

Let n 2 IN , ( 1 : : :  n ) 2  F ( IR ) ]n ,  : IRn ;! IR be a map.
Then, we call the fuzzy number   1  : : :  n ] with
8  9
<  ( t1  : : : tn ) 2 IRn  =
 1  : : :  n ] (t) := sup :min f 1 (t1)  : : :  n (tn) g 
 ( t1  : : : tn ) = t 
for all t 2 IR the image of ( 1  : : :  n ) under .
By using the extension principle of Zadeh, we are able to extend the usual ma-
thematical operations between exact data to operations between fuzzy - sets. The
following example shows how the extension principle works.
Example 2.8 :
8
>
< t ; 1 , if 1  t  2
 (t) = > 3 ; t , if 2 < t  3
:0 , otherwise.

 6
1:00 ........
.....................
.. ... . ...
0:75 ...... .................
............. ................ ............
0:50 ............. ............ ...........
................ ............ ................
.. . . . . . . . . . . . . ..
.
0:25 .... ........ ........ ....... .....
.......... ............ ............ .......... ..........
0
.
. .. . . . . . . . . . . . . . . . . ...
. . . . . . . . . . . . . . . . ..
.... . . . . . . . . . . . . . . . . . ....
. . . . . . . . . . . . . . . . . .
-
t
1 2 3 4 5

5
Let  (t) := 2 t :
=)   ](t) = sup f (v) j v 2 IR  2v = tg
= sup f (v) j v 2 IR  v = 2t g
= sup f ( 2t ) j t 2 IR g
=  ( 2t ) :

 6
1:00 ....
...............
..
. .................. ........
.. . . . . . . ....
... . . . . . . . ..
0:75 .............. .......... ................
............................ .............. ................ .................
.... . . . . . . . . . . . . . . . .....
........... ........ ....... ........ .............
.

0:50 .
........... .............. ................ .............. ................ .............. ..................
.. . . . . . . . . . . . . . . . . . . . . . . . . ..
............. ....... ........ ....... ........ ....... .............
0:25 .
.. ......... ............ .......... ............ ........... ............ .......... ............ .........
.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ......
.......... ........ ....... ........ ....... ........ ....... ........ ..........
0
.................. ............ .......... ............ ........... ............ .......... ............ ....................
...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...... -
t
1 2 3 4 5 6 7

The addition of fuzzy - sets and the multiplication of a fuzzy - set by a scalar can
be done easier by using only strong -cuts or -level-sets, respectively.

Denition 2.9 :
Let A  B  IR be non-empty sets and 2 IR.
Then, the Minkowski operations are dened by
(i) A + B := f x + y j x 2 A  y 2 B g 
(ii) A := f x j x 2 A g :

For the design of the fuzzy - Shewhart control chart we want to work with a
smaller class of fuzzy - sets than the class of normal fuzzy - sets F ( IR ) . This
leads to the class of convex fuzzy - sets. The motivation for dealing with this
smaller class is e.g. also given by the importance of this class for application
purposes.

6
Denition 2.10 :
(i) The class of all fuzzy - sets with a normal set representation is
denoted by Q( IR ) : It holds  2 Q( IR ) if and only if there is a set
representation f A j  2 (0 1) g of  such that
(a) inf A > ;1 implies inf A 2 A :
(b) sup A < 1 implies sup A 2 A :
(c) If inf A = ;1 or sup A = 1 then for all  2 (0 1) A is
convex.

(ii) The class of all convex fuzzy - sets is denoted by U ( IR ) .

It holds  2 U ( IR ) if and only if
(a)  2 F ( IR) 
(b) there exist a set representation f A j  2 (0 1) g of  with A is
convex and closed for all  2 (0 1) .

(iii) The class of all bounded convex fuzzy - sets is denoted by Uc ( IR) .
It holds  2 Uc ( IR ) if and only if
(a)  2 U ( IR ) 
(b) inf 0 > ; 1 and sup 0 < + 1 .

The class of convex fuzzy - sets has nice properties, which makes the work much
easier (see e.g. R.Kruse and K.D.Meyer 10]).

Theorem 2.11 :
Let n 2 IN  ( 1  : : :  n ) 2 IRn and  2 (0 1) :
(i) If ( 1  : : :  n ) 2  U ( IR ) ]n , then :
(a) P  2 U ( IR ) :
n
i i
i=1
Pn

= P i (i ) :
n
(b) i i
i=1  i=1

7
 (ii) If ( 1  : : :  n ) 2  Uc( IR ) ]n , then :
(a) P  2 U ( IR ) :
n
i i c
i=1
Pn

= P i (i ) :
n
(b) i i
i=1  i=1

3 The fuzzy random variable

The theory of fuzzy random variables is based on two dierent types of uncer-
tainty, namely vagueness and randomness. Fuzzy random variables were discussed
e.g. by H.Kwakernaak 11].
Let (     P ) be a probability space. On this probability space we will de-
ne a random mechanism. For many applications, we have the situation that
our randomness will be aditionally disturbed by errors of measurement, genetic
dierences, etc. These eects are not fuzzy eects. Therefore, we need a second
probability space (     P ) in order to describe these external eects. For
0 0 0

the probability space (     P ) , we assume :

0 0 0

For all 2 0 1] there exist A 2  with P ( A ) = . 0 0 0 0

The random mechanism of the fuzzy random variable depends on the product
space of both probability spaces (compare e.g. E.Henze 5]), separating the ran-
dom mechanism from the above described disturbances. The probability space is
dened by :
( ~  ~  P~ ) = (         P  P ) .
0 0 0

Now, the whole random mechanism under consideration can be be described by

an usual random variable on the probability space ( ~  ~  P~ ) . Hence, an inter-
pretation of a fuzzy random variable may be given as follows :
"A fuzzy random variable (f.r.v.) is a (fuzzy) perception of an unknown, usual
random variable."
We call the possible random variables of this random experiment the originals
of the fuzzy random variable. The vagueness of the observation can be descri-
bed by a fuzzy - set. The fuzzy random variable depends on the rst probability
space (     P ) and is independent of the second one (     P )  on the 0 0 0

other hand we can't observe the disturbance directly. The possible originals of
the f.r.v. are dependend of both probability spaces and therefore of the product
space ( ~  ~  P~ ) .
Now, we are led to the following denition of a f.r.v. (compare e.g. H.Kwakernaak
11]) :

8
Denition 3.1 :
A mapping X :  ;! Q( IR ) is called a fuzzy random variable (f.r.v.), if there
is a system f A ( ! )j ! 2    2 (0 1) g of subsets of IR with the following
properties :
(i) f A ( ! )j  2 (0 1) g is a normal set representation of X! := X (!) for
all ! 2  :
(ii) The mappings A :  ! IR and A :  ! IR are -B1-measurable for
all  2 (0 1)  where A ( ! ) := inf A ( ! ) and A ( ! ) := sup A ( ! ) :
For the following analysis, we have to make a few assumptions :

Assumption 3.2 :
(i) We restrict the denition of a f.r.v. to the class of convex fuzzy sets, i.e. a
f.r.v. X is a map
X :  ;! U ( IR ) respectively Uc( IR ) :
(ii) The possible originals with respect to the random experiment are unknown,
but we assume that the originals U are normally distributed :
U N ( 0  02 )  0 2 IR  02 > 0 .
Therefore, the set ~ of all possible originals is given by all usual normally distri-
buted random variables on ( ~  ~  P~ ) .
With regard to assumption (i) it has been shown e.g. R.Kruse and K.D.Meyer
10] that we will get the same results using either the set ~ or  . The set  is
given by all normally distributed random variables on (     P ) : Therefore, we
restrict the following considerations to the easier case, where X :  ;! U ( IR )
is a f.r.v., and U :  ;! IR is a possible original from the set  . Then, by
denition of the fuzzy logic the acceptability of the vague statement "U is original
of X " can be calculated by :
X ( U ) = inf f X! ( U ( ! ) ) j ! 2  g .

4 The expectation of a fuzzy random variable

For the design of a control chart within the framework of quality control with
fuzzy data, we do need the expectation and the variance of a f.r.v. R.Kruse and

9
K.D.Meyer 10] developed a concept for calculating these quantities. Again, this
concept is based on the extension principle of Zadeh. The expectation and the
variance of a f.r.v. are fuzzy - sets.

Denition 4.1 :
Let (     P ) be a probability space, X :  ;! U ( IR ) be a f.r.v. and t 2 IR :
(i) E :  ;! U ( IR ) with
8  9
>
<  U 2  >
=
( E X ) (t) := sup > !inf X (!)(U (!))  E jU k j < 1  >
: 2  E U k = t 
is called the expectation of X with respect to 

(ii) Var :  ;! U ( IR ) with

8  9
>
<  U 2  >
=
( Var X ) (t) := sup > !inf X (!)(U (!))  Var jU j < 1  >
: 2  Var U = t 
is called the variance of X with respect to  .

These denitions seem to be too complicated for calculations. But set represen-
tation leads to a feasible method for computing the expectation.

Theorem 4.2 :
Let X :  ;! U ( IR ) be a f.r.v. such that E jY 0 j < 1  E jY 0 j < 1  and
dene for ! 2  and  2 0 1) :
(Y ) ( ! ) := inf (X ( ! ))  (Y ) ( ! ) := sup (X ( ! ))
Then it holds :
f E Y   E Y  ] j  2 (0 1) g is a set representation of E X .

The following remark gives us a possibility for computing the expectation of any
linear combinations of f.r.v.'s.

10
Remark 4.3 :
Let 2 IR  X1  X2  X :  ;! U ( IR ) be f.r.v.'s , then it holds :
E (X1 + X2 ) = E (X1) + E (X2 ) 
E ( X ) = E X :

5 The statistics X and S

In this section, we want to discuss two of the most important statistics of a fuzzy
random sample. Both are basic tools for constructing the fuzzy - Shewhart control
chart.

Denition 5.1 :
Let ( X1  : : :  Xn ) :  ;!  U ( IR ) ]n :
( X1  : : :  Xn ) is called a fuzzy random sample if the fuzzy sample variables
Xi are i.i.d. for i 2 f 1  : : :  n g :

Denition 5.2 :
(i) Let ( X1  : : :  Xm ) be a fuzzy random sample
(a) X := 1 P X 
m
m j =1 j

(b) S 2 := P
m
m 1 j =1 ( Xj ; X )2 :
1
;

(ii) Let ( Xi1  : : :  Xim ) be a fuzzy random sample for i 2 f 1  : : :  n g

(a) X := mn1 P P Xij 
n m
i=1 j =1
(b) S := 1 P ( X ; X )2 
2 m
i m 1 ij i
;
j =1
1 P
n
(c) S := n i=1 S i :

Again, these statistics are fuzzy - sets. Since we want to use these statistics as
estimators for the moments of the f.r.v.'s, we have to discuss these statistics. In
the following remark, we rst discribe the corresponding -level-sets.

11
Remark 5.3 :
Let  2 (0 1) and X  S 2  X  Si2  S dened as in (5.2). Then it holds :
8  9
>
<  9 ( t1  : : :  tm )m 2 (X1)  : : :  (Xm ) >
=
(i) X  = > t 2 IR  such that 1 P tj = t > 
:  m j =1 
8  9
>
<  9 ( t1  : : :  tm ) 2m (X1)  : : :  (Xm ) >
=
(ii) S 2 = > t 2 IR  such that 1 P 
: m 1 j =1 ( tj ; t ) = t >

2
 ;

8  9
>
<  9 ( t11  : : :  tnm ) 2m (X11)  : : :  (Xnm ) >
=
(iii) X  = > t 2 IR  such that 1 Pn P 
:  mn i=1 j =1 tij = t >

8  9
>
<  9 ( ti1  : : :  tim )m2 (Xi1)  : : :  (Xim ) >
=
(iv) ( Si2 ) = > t 2 IR  such that 1 P 
: m 1 j =1 ( tij ; ti ) = t >

2
 ;

8  9 ( t  : : :  t ) 2 (S )  : : :  (S ) 9
<  1 n 1  n  =
(v) S  = : t 2 IR  such that 1 Pn ti = t :
 n i=1

Proof :
Applying the extension principle of Zadeh, at once, we will get these results.

For the following analysis, we do need the inmum and the supremum of these
fuzzy - sets.

Lemma 5.4 :
Let X und S be dened as (5.2). Then it holds :
(i) inf X = 1 P inf(X )  sup X = 1 P
m m
 m j =1 j   m j =1 sup(Xj ) 
1 P 1 P
n n
(ii) inf S  = n i=1 inf(Si )  sup S  = n i=1 sup(Si ) :

12
Proof :
Again, the extension principle of Zadeh and Theorem 2.11 implies :
 m !
inf X = inf 1 P X
 m j =1 j

P
m
= inf m j =1 (Xj )
1

1 P
m
= m j =1 inf(Xj ) 2
The other assertions can be shown analogously.

Next, we dene some entities, which will be needed for determing the control
limits (case of unknown variance).

Denition 5.5 :
Let ( Xi1  : : :  Xim ) 2  U ( IR ) ]m  i 2 f 1  : : :  n g  and S 2i as in Denition
5.2 (ii) .
Dene :
(i) ( (S 2) ) := 1 P ( inf (X ) ; inf (X ) )2 
m
i inf  m 1 ij  i 
;
j =1

(ii) ( (S 2i )sup ) := 1 Pm
( sup (Xij ) ; sup (Xi) )2 
m ; 1
j =1
1 P
n
(iii) ( (S )inf ) := n i=1 ( (S i )inf ) :

Let us consider the usual random variables inf ( X ) , sup ( X ) , ( (S 2i )inf ) 

and ( (S 2i )sup ) .

Lemma 5.6 :
Let ( X1  : : :  Xm ) :  ;!  U ( IR ) ]m be an i.i.d. sequence of f.r.v. with :
E Xj =   Var Xj = 2 forall j 2 f 1  : : :  m g , where   2 2 Uc( IR ) :
Dene for j 2 f 1  : : :  m g  ! 2    2 0 1) :

13
(Yj ) ( ! ) := inf ( (Xj )! )  (Yj ) ( ! ) := sup ( (Xj )! )
Then it holds :
(i) E ( inf ( X ) ) = E ( Y 1 ) = inf  
(ii) E ( sup ( X ) ) = E ( Y 1 ) = sup  
(iii) Var ( inf ( X ) ) = m1 Var ( Y 1 ) 
(iv) Var ( sup ( X ) ) = m1 Var ( Y 1 ) :

Remark 5.7 :
Let us remark :
(i) Var ( Y 1 )  inf 2 
(ii) Var ( Y 1 )  sup 2 
where in general, equality does not hold (see R.Kruse and K.D.Meyer 10]) .

Remark 5.8 :
In Lemma 5.6, we are allowed to replace X and m by X and nm , respectively.

Lemma 5.9 :
Let ( X11  : : :  Xnm ) :  ;!  U ( IR ) ]nm be a i.i.d. - sequence of f.r.v.'s with
E Xij =   Var Xij = 2 for all i 2 f 1  : : :  n g  j 2 f 1  : : :  m g where
  2 2 Uc ( IR ) :
Dene for i 2 f 1  : : :  n g  j 2 f 1  : : :  m g  ! 2    2 0 1) :
(Yij ) ( ! ) := inf ( (Xij )! )  (Yij ) ( ! ) := sup ( (Xij )! )  and let
q
m := m 2 1 ; ;( (m2; )1 ) :
m
;
2

Then it holds :

14
 (i) E ( (S 2i )inf ) = Var ( Y 11 ) 
(ii) E ( (S 2i )sup ) = Var ( Y 11 ) 
q
(iii) E ( S inf ) = m Var ( Y 11 ) 
q
(iv) E ( S sup ) = m Var ( Y 11 ) :

Proof :
X11   X1m   Xn1   Xnm i.i.d. f.r.v.
=) Y 11   Y 1m   Y n1   Y nm i.i.d.
Y 11   Y 1m   Y n1   Y nm i.i.d.

(i) E ( ( (S 2i )inf ) ) = E ( m1 1 P ( inf (Xij ) ; inf (Xi) )2 )

m
j =1
;

= E ( 1 P ( ( Y ) ; 1 P ( Y ) )2 )
m m
m 1 ij  m j =1 ij 
;
j =1
= Var ( Y 11 ) 2

(ii) analogous.
(iii) E ( S inf ) = E ( n1 P ( ( Si )inf ) )
n
i=1
Pn
= 1 E ( (S ) )
n i=1 i inf 

1 P
n q
= n i=1 m Var ( Y 11 )
q
= m Var ( Y 11 ) 2

(iv) analogous.

Lemma 5.9 implies the following unbiased estimator of the standard deviation of
the usual random variables Y 11   Y nm .

15
Remark 5.10 :
The statistics
(i) S := Sm 

(ii) S inf := S inf

m 

(iii) S sup := S sup

m :

are unbiased estimators of the standard deviation.

Now, all relevant statistics for the design of a fuzzy - Shewhart control chart are
provided. First, we want to give the control limits. The question arises how to
choose the control limits. In the classical situation, assuming normally distributed
random variables Ui  i 2 f 1  : : :  n g.
We obtain the upper and lower control limits UCL and LCL, respectively :
p z  p
UCL : IR2 ;! IR with UCL (E U  Var U ) = E U + 1;m2 Var U p

and
p z  p
LCL : IR2 ;! IR with LCL (E U  Var U ) = E U ; 1;m2 Var U p

where z1 2 denotes the (1 ; 2 ) - quantile of the standard normal distribution.

;

Now these usual variables should be extended to fuzzy variables by applying the
extension principle of Zadeh.

Theorem 5.11 :
Let LCL (:  :) and UCL (:  :) are control limits of the usual Shewhart chart,
LCL : IR2 ;! IR
(t1  t2) ;! t1 ; z1;m1 t2 
p

UCL : IR2 ;! IR
(t1  t2) ;! t1 + z1;m1 t2 
p

1  2 2 (0 1) such that 1 + 2 =  : Let m 2 IN be the sample size and

dene for (1  2) 2  F ( IR ) ]2   2 0 1)  t 2 IR :

16
 8  9
<  9 (t1  t2) 2 (1)  (2) =
LCL  1  2 ] := inf : u 2 IR  
 with LCL (t1  t2)  u
8  9
<  9 (t  t ) 2 (1)  (2) =
UCL  1  2 ] := sup : u 2 IR  1 2 
 with UCL (t1  t2)  u

E 1  2] :=
8  LCL    ]  UCL    ] ] , if LCL 1  2] > ;1 
>
>  1 2  1 2
>
> UCL 1  2] < +1 
>
>
>
> ( ;1  UCL 1  2] ] , if LCL 1  2] = ;1 
>
>
>
> UCL 1  2] < +1 
<
>
>  LCL 1  2]  +1 ) , if LCL 1  2] > ;1 
>
>
> UCL 1  2] = +1 
>
>
>
> IR , if LCL 1  2] = ;1 
>
>
>
: UCL 1  2] = +1 

K 1  2 ] (t) := sup f IE

1 2 ] (t) j  2 (0 1) g. Then it holds :
K :  F ( IR ) ]2 ;! U ( IR ) is a convex fuzzy - set and f E 1  2] j  2 (0 1) g
is a set representation of K 1  2] :

Proof :
1. Let (1  2) 2  F ( IR) ]2 :
Obviously, we get for 0     < 1 :
LCL 1  2]  LCL 1  2]
 UCL 1  2 ]
 UCL 1  2 ] :

Because the intersectionTof a non-increasing sequence of closed non-empty

sets is not empty, i.e. E 1  2] 6=  
 2 (01)

17
 it holds :
K 1  2] 2 F ( IR ) :
Taking into account that f E 1  2] j  2 (0 1) g is a set representation
of K 1  2] , it follows :
K 1  2] 2 U ( IR ) :
2. Let (1  2) 2  F ( IR ) ]2   2 0 1) :
We want to show that
(I) inf ( K 1  2] ) = LCL 1  2] 
(II) sup( K 1  2] ) = UCL 1  2] :
Let r :=  + 1
2r (1 ; ) :
Then it holds :
f LCLr 1  2 ] gr IN & inf ( K 1  2 ] ) 
2

f UCLr 1  2 ] gr IN % sup( K 1  2 ] ) :

2

(I) Let  := LCL 1  2 ]  LCLr 1  2]  r 2 IN :

Choose R 2 IN and ( t(1R)  t(2R) ) 2 (1)  (2) such that
LCL ( t(1R)  t(2R) )  ;R , if  = ;1 
LCL ( t(1R)  t(2R) )   + R
1
, if  > ;1 :
Because it exists an integer s = s(R) with (t(1R) t(2R)) 2 (1 )s  (2)s 
it holds :
r lim LCLr 1  2 ]  LCLs 1  2 ]
;!1

 LCL ( t(1R)  t(2R) )

=) r lim LCLr 1  2]  
;!1

=)  = r lim LCLr 1  2]

;!1

= inf ( K 1  2] ) 2

(II) analogous.

18
Remark 5.12 :
If we restrict ourselves to the case 1  2 2 Uc( IR )  we are led to E 1  2] =
 LCL 1  2]  UCL 1  2] ].

By these results we are able to derive control limits for the fuzzy - Shewhart
control chart. We assume that the expectation 0 2 Uc ( IR ) and the variance
02 2 Uc( IR ) are known.

Lemma 5.13 :
Let 0  o2 2 Uc ( IR) , X :  ;! U ( IR ) be a f.r.v. with normally distributed
originals and E X = 0  Var X = 02 :
Let m be the sample size,  2 0 1)   2 (0 1)  z1  may denote the (1 ; )-
;

quantile of the standard normal distribution.

Furthermore, choose 1  2 2 (0 1) such that 1 + 2 =  :
Then it holds :
LCL 0  0] = inf ( 0 ) ; z1;m1 sup(0) 
p

UCL 0  0] = sup( 0 ) + z1;m2 sup(0) :

p

Proof :
Choose u 2 IR such that t 2 (0) and t2 2 (0) implies
z1;2 1
LCL (t1  t2) = t1 ; m t2  u :
p

Obviously, it holds :
t1  inf (0)  t2  sup(0) 
and therefore :
inf (0) ; z1;m2 sup(0)  LCL (t1  t2)  u
p

z1;2
=) LCL 0  0]  inf (0) ; m
p sup(0) :
Similarly, we show
z1;2
UCL 0  0]  sup(0) + mp sup(0) 2

19
If there is at least one of the parameters 0 and 0 unknown, as in the classi-
cal situation, then we can get corresponding fuzzy - Shewhart control charts by
replacing these parameters by their estimates, which are discussed in section 4.
The dierent cases are treated in the following Lemma.

Lemma 5.14 :
The control limits for the fuzzy - Shewhart control chart are given by :
(i) 0 and o2 known, 0  o2 2 Uc ( IR ) :
LCL 0  0] = inf ( 0 ) ; z1;m1 sup(0) 
p

UCL 0  0] = sup( 0 ) + z1;m2 sup(0) 

p

(ii) 0 known, o2 unknown, 0  S 2 Uc ( IR ) :

z1;1
LCL 0  S ] = inf ( 0 ) ; p
m sup(S ) 
z1;2
UCL 0  S ] = sup( 0 ) + p
m sup(S ) 

(iii) 0 unknown, o2 known, X  o2 2 Uc ( IR ) :

z1;1
LCL X  0] = inf ( X ) ; p
m sup(0) 
z1;2
UCL X  0] = sup( X ) + p
m sup(0) 

(iv) 0 and o2 unknown, X  S 2 Uc( IR ) :

z1;1
LCL X  S ] = inf ( X ) ; p
m sup(S ) 
z1;2
UCL X  S ] = sup( X ) + p
m sup(S ) :

6 The fuzzy - Shewhart control chart

In the last section, we discussed the general method of computing control limits for
fuzzy - Shewhart control charts. For application purposes, two questions remain
to be solved.
1. The stopping rule of the fuzzy - Shewhart - control chart.

20
 2. The choice of the parameters 1  2 2 (0 1) .
The problem which arises is that in principle we have to analyse innite many
 - levels ( 2 (0 1)) of the corresponding statistic. Following R.Kruse and
K.D.Meyer 10], we suggest the following general procedure.
1. Choose an integer N as the number of inspection -levels.
2. Choose K 2 f 1  : : :  N g arbitrary.
3. Choose the set of -levels f 1  : : :  N g  0 1) arbitrary.
4. Reject the null hypothesis, if at least K out of N -levels are rejected.
This procedure, applied to the fuzzy control chart, leads to the following stopping
rule.

Stopping rule for the fuzzy - Shewhart control chart

Stop the process if and only if in at least K out of N -levels it holds :
cl (X )i 6  LCLi 1  2]  UCLi 1  2] ]  where
N 2 IN  K 2 f 1  : : :  N g  f 1  : : :  N g  0 1)  i 2 f 1  : : :  N g 
1  2 2 Uc ( IR) :
Next we want to find optimal values for 1  2 2 (0 1) for the case " 0  0
known". This will be done by studying the operating characteristic function.
The following Lemma shows, how to calculate the operating characteristic func-
tion for every -level.

Lemma 6.1 :
Let  2 0 1)  0 2 Uc( IR ) and 02 2 Uc( IR ) be the given expectation and
variance of the f.r.v. X  and let X1  : : :  Xm be a fuzzy - random sample.
Hence, for every  2 0 1) it holds :
L (  ) := P ( f ! 2  j cl ( X ! )   UEG 0  0]  OEG 0  0] ] g )
p

  
sup (p0)
; sup ()
m + z1 2 ;
Var Y ;

 inf (p 0 ) inf () pm ; z 

1 1
;

Var Y ;

where (Y ) ( ! ) := inf (X ( ! ))  (Y ) ( ! ) := sup (X ( ! )) .

21
Proof :
1 P
m
Let N 2 IN  X1  : : :  Xm be a fuzzy random sample and X = m j =1 Xj :
For abbreviation, dene :
 := inf (X ) and  := sup (X ) :
Choose ( U1(N )  : : :  Um(N ) ) 2 m and ( V1(N )  : : :  Vm(N ) ) 2 m such that
U (N ) (!)   + 1
N  if  > ;1 
U (N ) (!)  ; N  if  = ;1 
V (N ) (!)   ; N
1
 if  < 1 
V (N ) (!)  N  if  = 1 :
Furthermore, for the originals let us assume :
E U1(N ) =  (N )   (N ) 2 E X =  
Var U1(N ) =  2 (N )   2 (N ) 2 Var X = (02) 
E V1(N ) = (N )  (N ) 2 E X =  
Var V1(N ) = #2 (N )  #2 (N ) 2 Var X = (02) :
Under the assumptions of our model 0(N ) 2 (0) and (0N ) 2 (0) exist,
where 0(N ) and (0N ) are the expectations of the random variables Uj(N ) and
Vj(N )  j 2 f 1  : : :  m g  in the in-control-state, respectively. It holds:
 (N ) z 


P ! 2   U ( ! )  0 ; m  0 (N )
p1 1 (N )
;

(  (N ) (N )
)!
! 2   U (!)(N;)  pm  0 ;  (N ) pm ; z1
(N )
= P (N ) 1
0 0 ;

(N )
 !
= 1 ;  0 ;  (N ) pm ; z1 1
0(N ) ;

and

22
  

P ! 2   V (N )(!)  (0N ) + Zp1 m2 #(0N ) ;

(  (N ) (N ) p (N ) (N ) p
)!
= P ! 2   V (!)(N;) m  0 ;(N ) m ; z1 2
#0 #0 ;


(N )
!

=  0 ; (N ) pm ; z1 2 :
(N )
#0 ;

Dene for N 2 IN :
8  (N ) zp1 1  (N ) ^ 9
>
>  U (!)  0(N ) ; >
>
< =
;

m 0
AN := > ! 2   (N ) zp1 2 #(N ) > :
>
:  V (!)  (0N ) + ; >

m 0
Thus, AN 2  holds for N 2 IN and
 (N ) (N ) !
; p
P (AN )   0 (N ) m + z1 2 ;
#0 ;

 (N ) (N ) !
 ;  p
 0 (N ) m ; z1 1 :
0 ;

Dene :
A = Nlim sup AN = T S AK 2 
1 1

!1 N =1 K =N
(see e.g. Henze 5]). We have :
A = f ! 2  j ! 2 AN for an innite number of N 's g :
Because of the denition AN := S AK  it holds A = T AN 
1 1
0 0

K =N N =1
hence, we get for all N 2 IN :

23
AN  AN +1 =) AN # A
0 0 0

=) P (Nlim AN ) = Nlim P (AN ) :

!1
0

!1
0

For all N 2 IN  it holds :

 (N ) (N ) !
P( S1

AK )  P (AN )   0 (N ) ; p
m + z 1 2 ;
K =N #0 ;

 (N ) (N ) !
 ;  p
 0 (N ) m ; z1 1 :
0 ;

Hence,
T S

P (A) = P AK
1 1

N =1 K =N

= Nlim P (AN )
!1
0

 Nlim P (AN )
!1

0 1
sup(  ) ; sup(  ) p
 @ q
0  0  m + z1 2 A ;
Var Y 
;

0 1
p
 @ inf(q0 ) ; inf()
m ; z1 1 A 2
Var Y 
;

Remark 6.2 :
For the case  = 0  Lemma 6.1 implies : L ( 0 )  1 ; ( 1 + 2 ) :

According to our stopping rule we have to carry out such a test simultane-
ously at N dierent levels i  i = 1 : : :  N : Therefore, we dene for every
i  i = 1 : : : N the statistics

24
 8
>
< 1 , if cl(X )i 6 UEGi 0 0]  OEGi 0 0] ]
i  X ] := >
: 0 , otherwise,
which leads to the test function
8
< 1 , if P i  X ]  K
> N
 X ] := > i=1
: 0 , otherwise.
From Lemma 6.1, it follows at once
P ( i  X ] = 1 )  1 ; Li (  ) for all i 2 f 1  : : :  N g
and
E ( i  X ] )  1 ; Li (  ) for all i 2 f 1  : : :  N g :
Hence, for the operating charateristic function we get the following result.

Theorem 6.3 :
Under the assumptions of Lemma 6.1, it holds for every N 2 IN  K 2 f1 : : :  N g
and f 1  : : :  N g  0 1) :
L (  ) := P ( The process will not be stopped )
= P ( H0 :  = 0 will be accepted if  true parameter)
 1 ; 1 P ( 1 ; Li (  ) )
N
K i=1

Proof :
1 P
N
We have to show P (  X ] = 1 )  K i=1 ( 1 ; Li (  ) ) .
In order to proof this assertion, we use the following well-known inequality (see
e.g. D.Morgenstern 16]) :
For every nonnegative random variable U and every K 2 IN , it holds
P (U  K )  K1 E U : Hence, we get

25
P ( X ] = 1) = P ( P iX ]  K )
n
i=1 !
 KE
1 P
N
X ]
i=1
 
= 1 P E X ]
N
K i=1
1 P
N
 K i=1 ( 1 ; Li (  ) )
=)
L (  ) = P (  X ] = 0 )
= 1 ; P (  X ] = 1 )
P
N
 1 ; 1 ( 1 ; Li (  ) ) 2
K i=1

Remark 6.4 :
In the case of  = 0  it follows
L ( 0 )  1 ; K1 P ( 1 + 2 ) = 1 ; NK ( 1 + 2 ) :
N
i=1

If we want to meet the following demands :

1. 1 =! 2 ( symmetry of the quantiles ),
!
2. L ( 0 )  1 ; ,
we get from 1 ; KN ( 1 + 2 ) =! 1 ;  for 1 and 2 :
1 = 2 = K 
2N  2 (0 1) 2
Example 6.5 :
Let the process variable X :  ! Uc( IR ) be a f.r.v. with normally distributed
possible originals. We assume E X = 0 und Var X = 02 with

26
  6
1:00 ....
.......... 8 t ; 3 , if 3  t  4
........ ............ >
>
0:75 .... .......
........ .............. <
0:50 ............... ................... 0 (t) = > 5 ; t , if 4  t  5
0:25
........ ........ ...
.............. ............ .....
........ .............. ................ ............
>0
: , otherwise
0
....... .......... ............ ..........
....... ....... ........ ........ -
t
1 2 3 4 5

 6
1:00 ....
.......
.......... 8 (t ; 2)2 , if 1  t  2
0:75 .......... <
0:50
................
...................
02 (t) = : 0 , otherwise
...................
0:25 ........................
0
................... ...............
. . -t
1 2 3 4

The control limits are given by

z1; k q
LCL = inf ( 0 ) ; m
2Np sup ( 02 ) 
z1; k q
UCL = sup ( 0 ) + m
2Np sup ( 02 ) :
Choosing e.g. m = 4 and z1 2kN = 3:0  we get L ( 0 )  0:9973 : ;

The graphical representation of the fuzzy - set of the acceptance region is given
below.In the in-control-state, the sample mean will be inside of this fuzzy - set
with a probability of at least 0.9973.
 6
1:00 ............................................................................................................................................
...... ........ ....... ........ ....... ........ ....... ........ .......
.............. ................ .............. ................ .............. ................ .............. ................ .................
.
.............. ............ ........... ............ .......... ............ ........... ............ .................
0:75
.
......... .............. ................ .............. ................ .............. ................ .............. ................ ................ ............
.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
.......... ....... ........ ....... ........ ....... ........ ....... ........ ........ ..........
.

0:50 ....................... .............. ................ .............. ................ .............. ................ .............. ................ ................ ........................
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
.......... ............ ........... ............ .......... ............ ........ ........... .......... ............ .......... .......... ..............
..
................... ................ .............. ................ .............. ................ ............. ............... .............. ................ ............... .............. ..........................
0:25 .... . . . . . . . . . . . . . . . . . . . . . . . ....... ..... . . . . . . . . . ...... . . . .. . . . . . ....
....... ........ ........ ....... ........ ....... ........ ....... ........ ....... ........ ........ ....... ........ .........
............... ................ ................ .............. ................ .............. ................ .............. ................ .............. ................ ................ .............. ................ .........................
0
.
..
.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ..
...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ......
.
-
t
1 2 3 4 5 6 7

Let us consider the decision rule (control chart) with N = 6 and K = 3  where
the six -levels may be chosen as
f1  2  3  4  5  6 g = f0:1  0:3  0:5  0:7  0:8  0:9 g   0  1 ) :

27
As an example, let us assume that we have observed the following fuzzy - random -
sample of size 4 :
 6
1:00 ...
......... 8 2(t ; 3 ) , if 3  t  2
............... >
>
0:75 .. . .
.... ......
........ ............ < 5 2 2

0:50
.. . . .
..... ......... X1 (t) = > 2( 2 ; t) , if 2  t  25
............. ................
....... .........
>
:0 , otherwise
0:25 .......... .............
............... ..................
0 ... . . . . . . .... -
t
1 2 3 4

 6
1:00 ..
........
........... .......
8 t ; 2 , if 2  t  3
>
>
0:75 ....... .....
.......... ......... <
0:50
................... ...............
............ ..........
X2 (t) = > 4 ; t , if 3  t  4
>
. . . . . . .

... ........ ....... ...

0:25
......... ................ .............. ........
........ ............ ........... ........
:0 , otherwise
........ ........ ....... .......
0
.............. ............ .......... ............. -
t
1 2 3 4

 6
1:00 ...
.......... 8 t ; 1 , if 1  t  2
......... ............ >
>
0:75 .... .......
.......... ............ <
0:50
................ ....................
......... ........ ... X3 (t) = > 3 ; t , if 2  t  3
.... .......... ............ ......
......... .............. ................ ..........
>
:0 , otherwise
0:25 ..... ....... ........ ........
0
.......... ........... ............ ...........
......... ....... ........ ......... -
t
1 2 3 4

 6
1:00 ..
......
.............. 8 2(t ; 1 ) , if 1  t  1
.. . >
>
0:75 .........
....... ............ < 3 2 2

0:50 .... ...........

..... ........ X4 (t) = > 2( 2 ; t) , if 1  t  23
......... ..................
.. . . . . .
........ ..............
>
:0 , otherwise
0:25 .............. ....................
0
.. . . . . . ..
.. . . . . . . .... -
t
1 2 3 4
If we model our observations by this special class of fuzzy - sets (triangles, not
necessarily symmetric), the sample mean can easily be computed. The graphical
representation of the sample mean is naturally again a triangle. The center, as
well as the left and right slope are the arithmetrical means of the centers, the left
and the right slopes of the "sample triangles", respectively. In our example, we
get :

28
  6
1:00 ....
......... 8 4 (t ; 5 ) , if 5  t  2
.................. >
>
0:75 ...... ....
........... ......
< 34 11 4 4

0:50 ................. ............

.. . . . . X (t) = > 3 ( 4 ; t) , if 2  t  114
0:25
........... .......
........ .............. ............... >
:0 , otherwise
...... ........... ............
0
...... ....... .........
....... ....... ........... -t
1 2 3 4
For the chosen -levels, we give the corresponding -cuts of the acceptance region
and of the sample mean as well as the i-values :
i i LCLi UCLi inf X i sup X i i(X )
1 0.1 1.082 6.918 1.325 2.675 0
2 0.3 1.469 6.531 1.475 2.525 0
3 0.5 1.823 6.177 1.625 2.375 1
4 0.7 2.134 5.866 1.775 2.225 1
5 0.8 2.270 5.730 1.850 2.150 1
6 0.9 2.393 5.607 1.925 2.075 1

Because of P i ( X ) = 4  3 = K , we get for the test function  X ]:

6

i=1
(X ) = 1
Therefore, in this example the process will be stopped, our decision rule signals
an out-of-control-state.
The graphical representation of the acceptance region and the observed sample
mean may illustrate the situation at hand, and the need for  decision rule as
suggested in this paper.
 6
1:00 ...................................................................................................................
....
...
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.
... ...
... ..
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...
...
.... ....
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....
... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
..
.
... ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...
0:75 ...
...
... ....
... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
...
.... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
....
. . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....
..
.
...
.. . . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....
..
...
.
.
..
. . . ..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....
0:50 ....
...
...
. . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....
. .
..
. . .
. . . . . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . .
.. . . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . .....
. . . . . . . . . .

.... .....
.. . . . . . . . . . . . . . . . ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
...
. .. . . . . . . .. . . . . . .
... . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
.... . . . . . . . .

0:25 ..
. .... . . . . . . . . . . . . . ....... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.....
....
. . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
... . . . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.. . ... . . . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. ..... . . . . . . . . . . . ..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .....
. .....
-
...
... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . .... . . . . . . . . . . . ..... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.....
.
... . . .. . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . ... . . . . . . . . . . . . .... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
... ...
. . . ... . . . . . . . . . . . . .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . .... . . . . . . . . . . . . . .
0 .. . . . . . .
.
t
1 2 3 4 5 6 7

In a forthcoming paper we will study the statistical behaviour (e.g. average run
length) of such fuzzy - Shewhart - charts.

29
References
1] Bellmann, R.E., Giertz, M. :
On the analytic formalism of the theory of fuzzy sets
Inform.Sci. (1973) 5, 149 - 156
2] Bandemer, H., Gottwald, S. :
Einfuehrung in Fuzzy - Methoden
Verlag Harri Deutsch, Thun (1990)
3] Barrett, J.D., Woodall, W.H. :
A probabilistic alternative to fuzzy logic controllers
submitted for publication
4] Gebhardt, J. :
Ueber die Varianz unscharfer Zufallsvariabler und ihre Anwendungen
Diplomarbeit Technische Universitaet Braunschweig (1988)
5] Henze, E. :
Einfuehrung in die Masstheorie
2. Au!age (1985), Bibliographisches Institut Mannheim
6] Hoegel, J., Wol, H. :
A general approach to sequentially designed statistical process control
Oesterreichische Zeitschrift fuer Statistik und Informatik (1992) 22, 243 -259
7] Kruse, R. :
The strong law of large numbers for fuzzy random variables
Inform.Sci. (1982) 28, 233 - 241
8] Kruse, R. :
Statistical estimation with linguistic data
Inform.Sci. (1984) 33, 197 - 207
9] Kruse, R. :
On the variance of random sets
J. Math. Anal. Appl. 121 (1987), 469 - 473
10] Kruse, R., Meyer, K.D.
Statistics with vague data
D.Reidel Publishing Company (1987)

30
11] Kwakernaak, H. :
Fuzzy random variables
Part I : Denitions and theorems
Inform.Sci. (1978) 15, 1 - 15
Part II : Algorithms and examples for the discrete case
Inform.Sci. (1978) 17, 253 - 278
12] Lucas, J., Saccucci, M. :
Exponentially weighted moving average control schemes : properties and
enhancements
Technometrics (1990) 32, 1 - 12
13] Meyer, K.D., Kruse, R.
On the variance of a fuzzy random variable
Preprint Technische Universitaet Braunschweig (1985)
14] Laviolette,M., Seaman, J.W., jr., Barrett, J.D., Woodall, W.H. :
A probabilistic and statistical view of fuzzy methods
Technometrics, to appear
15] Miyakoshi, M., Shimbo, M. :
Set representations of a fuzzy set and its applications to Jensen's inequality
Fuzzy sets and systems, to appear
16] Morgenstern, D. :
Berechnung des maximalen Signikanzniveaus des Tests "Lehne H0 ab, wenn
k unter n Tests zur Ablehnung fuehren".
Metrika (1980) 27, 285 - 286
17] Naether, W. :
Linear statistical inference with random fuzzy data
Preprint Technische Universitaet Bergakademie Freiberg (1994)
18] Naether, W., Albrecht, M. :
Linear regression with random fuzzy observations
Statistics (1990) 21, 521 - 531
19] Negoita, C.V., Ralescu, D.A. :
Applications of fuzzy sets to system analysis
Birkhaeuser Verlag, Basel (1975)
20] Rinne, H., Mittag, H. :
Statistische Methoden der Qualitaetssicherung
Carl Hanser Verlag, Muenchen (1989)

31
21] Tong, Y.L. :
The multivariate normal distribution
Springer Verlag (1990)
22] Uhlmann, W. :
Statistische Qualitaetskontrolle
Teubner, Stuttgart (1982)
23] Viertl, R. :
Einfuehrung in die Stochastik
Springer Verlag Wien New York (1990)
24] Wol, H. :
Statistische Prozesskontrolle
Skript zur Vorlesung, Preprint Univesitaet Ulm (1994)
25] Yager, R.R. :
A characterisation of the extension principle
Fuzzy sets and systems (1986) 4, 235 - 242
26] Zadeh, L.A. :
Fuzzy sets
Information and control (1965) 8, 338 - 353
27] Zadeh, L.A. :
The concept of a linguistic variable and its application to approximate rea-
soning
Part I :
Inform.Sci. (1975) 8, 199 - 249
Part II :
Inform.Sci. (1975) 8, 301 - 357
Part III :
Inform.Sci. (1975) 9, 43 - 80

32