Chapter 2

Histograms
2.1 Data have dispersion
Jn Chapter 1 we sued how oealet data Now let us consider how
to arrange these:
ony dain the factory weclletdt in various forms. For example,
absent
centration of products, to record them in daily reports, graphs and control
charts.
‘There is a purpose to collecting all these data. Try thinking about the
son behind the data you are collecting. Is the purpose clear? For
tcample, let's suppose You av es ten sls fom 8 erin ot ed
which
were chosenat andom,we can make inferences about the measurementof
articles from the entire lot from which these data were taken, or from the
ie i 24).
 kisroonaws:
Data are required to obtain the average dimensions and the degree of
dispersion so that we can determine whether it is alright to receive or ship
the lot, and whether the production process used for manufacturing the
Tot was suitable, or if some action must be taken. Jn other words, we are
about to take action on a lot or process on the basis of data gained from
‘the samples we have taken.
Products from the same production line usually differ slightly in
dimension, hardness or other qualities. If, after measuring ten samples,

wrong or we might even wonder if they had ever been measured at all!
We commute to work every day and even if we take the same route and
the same vehicle we usually find that on some days the trip doesn’t take as
Jong as others. If we tried to make the trip in exactly the same time every
day, it would require 2 good deal of effort. In this way, when we look at
trai amount of tae on etcfos pen. Aca, wo ve
in world of denon. To kan the quity of gen snount of
onan we must use averages and disper
Take, for he feof an ceca apples Even if, onthe
average, the “Hf ofthe appliance song, if there much dsperion some
of the appliances wall weer out rapidly. This implies a loss in reiabity of
the product. That means one criterion for judging the quality of products
is whether, on the average, the life is long and at the same time the
dispersion is small
's assume that we take four samples of a certain part from the
production line daily for one month and take measurements, There are
two ways of looking atthe data for the 100 samples:
th,
For (1), we can construct a frequency table showing the number of
parts for each dimension. Then, if we make a histogram,it will be easy to
find the shape, the ot and the manner of dispersion of the size
measurement, For (2), in order to ee the changes in the data chrono-
logically, control charts or graphs giving the date vertically and the
dimensions horizontally are often used. Let us take case (1) and see how
to construct and use a histogram.
2.2. Howto prepares histogra
‘The data in table 2.1 represent the thickness (jn millimetres) of 100
retal blocks that are parts of optical instruments, When there is a8 much
7
GUIDE To QUALITY CONTROL
ta as the
data just by looking at the figures. Ina situation such as this,
if we arrange
the
2.1), and then drew up a graph,we can perceive the overall tendency.
‘There are many kinds of graphs, but one of the most common is the
‘istogram (Sgure 2.2). Let’s examine the method for making a histogram,
(1) Count the data. N=
(2) As shown in table 2.1, divide the data roughly into ten groups
Record the lest vals i cach group a Xan the smallest
‘the ta the sales Xp onthe whole (comperbletoa
satonl eestor). AX, = 3.68, X3=3.30.
“Table 2.1 Metal block thickness(ia mm)
= [x=
356°346 348 350 24 343 352 340 sae 320 [500 342
348 ge6° 350 352 347 348 346 350 356 3.254|356 |3.35

348 2, 352 34s 348 3.45 3.48

Bao See Sas. gas 346 Ser set saa 309 Paes sat
Bat 345 330346 347 347 341 348 354° 947
(0: The largest
vue in the rom N=100, x, =3.68
se Thesmales valve 7330

Figure 2.2 Metal block thickness

 ‘able2.2 Frequency table
Ges Caer bounaain Mica Freer Freeney
i 33 3
:‘ Bass
Be as ws oe aews baw 2
Hi4 300
can :3
H EL :
woo

Table2.3
‘Womberof Data WW) Number of classes (KI
50 5
50-100 6-10
100 ~ 250 7-2
‘over 250 10-0

(3) The range (R) of all the data is: R = Xz — Xs = 0.38. This range
can be divided into classesand the number of data belongingto
be investigated. The number of classes
(the number
of histogram bars) can be determined on the basis of table 23.
However, to get the rough number of classes, take K = 10, and
divide it into the range (R).
h Aas _ 038 = 0.038
K 10
(4) This ‘class interval, which will be used as the horizontal
it for the histogram, should be expressed as
Senses ace somese fin Ges oe when of x ees,
10 the units of measurement are 0.01). Here ht could be considered
Sea to 0.04, bat to make con ask we wl pr it
(s) (Cau boundary, which we must determinein order to make 2 bar
graph, is demarcated starting at one end of the range. It is
troublesome when actuals fall on the class boundary. To avoid this,
‘the boundary unit is taken as half the actual measurement unit.
QUIDE TO QUALITY CONTROL
In this case it is 0.005,In other words, the boundaries — the width
of bars — will be 3.275~3.325, 3325~3,375, etc. With check
marks such as /,(/, ///,/I/}, #44, etc, the data which belongto each
a smerated as shown in table 2.2 and a frequency table
is made. The total should correspond to V as outlined in step (1)
above. (Mistakes often occur here, s0 be careful.)
6) After examining the frequency table, you can get an idea of the
overall picture, but if it is indicated on a graph it becomes much
clearer. On graph paper, mark the class boundaries horizontally
and the frequency vertically like the histogram in figure 2.2, In the
blank areas write the background of the dataN, average values,
standard
a deviation, et. If there is: « company or industrial standard
i
fon the metal blocks are 3.28~3.60 mm, so this has been recorded
‘on the graph.
th way, a Mtgcan be mad, Since asta is gap with
bars, itis also called a bar graph. Each bar is referred to as a class. The

value of the class is called representative value or mid-ralue. A good deal

of information can be gained simply by prepering @ frequency table and
a histogram, as will be explained in the following section. Hf you wish to
lear the average value # and the standard deviations, it willbe ¥ = 3.476
and s = 0.065, (For the method of calculation, see practice problem 2)
2.3 How to use a histogram
In this section, we will see how to read and how to use histograms.
(1) Whatis the shape of the distribution?
Let’ try to answer the following questions by looking atthe histogram
in figure 2.2. What is the most common thickness of the metal blocks?
How great is the dispersion? Is the distribution symmetrical? Is it skewed?
1s there only one peak? Is it clifike? Does it look like a cogwheel? Are
there any isolated bars? In other words, what are the characteristic ofthe
product?
According to figure 2.2, most of the metal blocks are in the 3.425~
3.525
thinner is fewer in each direction, There is 2 symmetrical
distribution, and 3.3 mm to 3.7 mm dispersion, There are no isolated
sbnormal data values.
10
Example 1.
One QC Circle of company A took up the problem of the reduction of
scrap in the amount of metal which was trimmed from the product during

(lowering of 3), but there was also a reduction in dispersion (lowering of

4). This effect can be seen in the histogram,

wr tore improvement F= 133 4282

Bluae Ate inprommen F= 819218
Eee.)
Fre 2.3 Comparonhsogm
Example2

make certain that company B guarantees the quality of its parts, not
through inpection bat trough improved proces

Figure 2.4 Cutelke histogram

GUIDE TO QUALITY CONTROL
Example3
Data collected on the viscosity of a certain product resulted in the
combike histogram in figure 2.5.

Figure 25 Comblke histogram

This histogram looked abnormal, so the measurement methods were
checked. It was discovered that although the instrument had been set to
show only even numbers,it also gave readings of odd numbers, Thus, the
amount of od
number figues. (Aside from mistakes suchas this, be careful with
histogram class intervals and the integer multiples of the measurement
units, ie. multiples of 1, 2, 3... ete, otherwise you may end up with this
type of hist
Example 4.
All of have ‘i
data ales: However, gues for mbes of defective parts absentee,
inte ume ‘ry tt contnaonata it Figs 2.6. shoe th seb

umber of accidents, and number of defects — the distribution of these

dca wil often be found to tum an seyret for. OF cow,
with continuous data expressed in amounts, with data of 100 per
cent yild and 100 percent purty, datribution gues sometines run on
to the left because the limit
is set at 100 per cent.

i : sf em
eon Pliner den
 HisTOGRAMS
(2) What it threats with specifications?
What centage of out-ofspecification products? Do products
fully meet the spsetion! Is the average value at the exact centre of
timits?
In figure 2.2, where the thickness of metal blocks is shown, we see that
the value is roughly in the centre of the specification limits, but
the dispersion is greater than the width of the specification limits. So this
ispersion must either be reduced or the specification must be
reexamined.
Example5.
‘A histogram showingthe load characteristics of a microswitch is given
in figure 2.7. There are many defective microswitches, and on the
over half of the defects are due to load characteristics. For this reason,
the data on load characteristics taken during the manufacturing process
valve
is broad. These problems were analyzed through control chars and various
statistical methods; the result was a reduction inthe number of defectives.
This is a good example for showing how a histogram can be used to
petceve the state of the manufacturing process, to help people leam what
the problems are and thus to improve procs capability and reduce
defects. A process capability index is used to determine. whether 1
Aipentontuftcent smal in computson wi he pecicaton nt

Figure 2.7 Histogram

of load characteristics
(3) Is there a need to change the histogram?
When the data are stratified in accordancewiththe materials, machines,
Sits, workers, months, et, the dstbuton is probably ifent for
‘each, In extreme situations the histogram distribution may take
3
Quine To auaLITY CONTROL
of two peaks (bi-modal distribution). In the case of bimodal distribution
or broad dispersion, this distribution often includes two or more

ere is no difference
between these factors, the data can be drawn onto one graph.
Example 6.
A asicontacting company prose see metal panel fra etic
‘machine maker, with sheet metal supplied by the parent company.
However, the pressed products had many wrinkles and cracks, and they
often
‘the sheets, and the results were shown in a histogram (figure 2.8).

Since the dispersion was broad, investigations were made. It was dis-
a. ting in the
stratified histograms in figure 2.9. Its clear that there is difference in

‘will not necessarily show up as two separate peaks in one histogram. Two
sow peaks will only appear when the difference between A and B is
the data then we caaot dete the station If tee some someon
that a difference may develop, then 2 history of the data must be kept
even if it is somewhat cumbersome. This is important in improving and
‘controling the manufacturing process.
“4
 HisTOGRAMS

Figure 2.9 Hardness histograms by companies A and B

In addition to the histograms we have already studied, there are other
relevant charts. A histogram was used to show the distribution of machine
flues asin example & when two exes (uch a 10 and 11) a ced
failu
frequency and show thse in ar graph, ain Hgute 210.
Frequency

1234 $8 7
Number of defectives
Figure 2.10 Distributionof detectives (bar chart}
‘This graph indicates the distribution of the number of defectives and,

hhad to be overcome to reduce the number of microswitch defectives

(example 5). With this chart the analysiswas on the load
).
It is convenient to make @ frequency distribution table through the use
of a check sheet without having to put each datum onto data sheets.
1s
GUIDE TO QUALITY CONTROL,
%

Percentage of detectives

- oS

Figure 2.12. Pareto diagram of miroowiteh

defects

Figure 2.12 shows a process capability check sheet for TV manufacturing

this
remain constantly avare of the time while checking, or to colour-code
your checks to show time differences.
If histograms are read and used as outlined, for example in monthly
reports, problems will become apparent quickly and data will be much
‘more meaningful than just mere rows of Figures. Histograms are often used
in charting the precision of machines or in process capability studies
tty to eliminate defectives and improve yield and product quality, when
we probe the relationship between specification and outcome, when We
sly shnomal dt, when we emi the cuts which ed fo cng
in the manufacturing process by stratifying the materials, equipment, ete
and, finally, when we seek to upgrade the workers’ awarenessof
control through actual on-the-job accomplishments.
16
 Sujmeynucusax 105 sous yoeyo Auiqedes
S01 Z1°Z eng
“ces Rnb Bong OD TF woRROap
‘lesen PPO
histoarans

eben
a SNR oe)
w z ve
Et |__,,
$6
z zr
St ze ae
a 50 a |
a #= se
308
ve 33 1
27 _——
(a |
I= == aa ee
—_—_————|
= = —_a a
F == =E =se aesrs |
r = = a
w 7 inane
annetessons =oy sop se
rs Sawa
| oe | omer | ae
=| tam
seszze | vonesyeeds | onsmseeeos—au | Aen oom