06-Mar-15

Accept or Reject

Changing data into information

GO
No Go U-turn

Decision making

Photo by brizzle born and bred - Creative Commons Attribution-NonCommercial License https://www.flickr.com/photos/20654194@N07 Created with Haiku Deck

Product Control And Process Control Philosophy

SPCs Historical Background
The product control view:
Walter Shewhart suggested that every process exhibits measures quality of a product in terms of its acceptability as
some degree of variation and therefore is expected. measured by conformance to engineering specifications.
identified two types of variation (chance cause) and (assignable emphasizes detection and containment of defective material
cause) through inspection/screening, therefore making
proposed first control chart to separate these two types of quality and productivity opposing rather than supportive
variation. forces.
SPC was applied during World War II to ensure
interchangeability of parts for weapons/ equipment. The process control view:
Resurgence of SPC in the 1980s in response to Japanese emphasizes the prevention of defective material from being
manufacturing success. made in the first place by seeking the root cause of the
problem and eliminating it altogether.
makes quality and productivity enhancement possible
simultaneously by continually seeking ways to reduce
variation, thereby eliminating waste and inefficiency in the
process and variation in performance of the product.

Mean ,standard deviation

Mean ,standard deviation
Given: X = 57 and X = 10, and process represented is normal distribution .

Example 1 To calculate the probability of passed, we need to find P (X30). Use z transformation to determine the values
of Z associated with the values of X
The scores on a test given to students in a large z = (30 - 57) / 10 = -2.7
Therefore, we need to find P (Z -2.7). Using Table A.1 in the Appendix,
class are normally distributed with a mean of 57 P (Z -2.7) = 1 - P (Z -2.7)
and a standard deviation of 10. The passing score = 1 - 0.0035 = 0.9965
for the exam is 30. If a student is randomly Therefore, probability of student passing is 99.65 %

selected, what is the probability that he or she To calculate the probability of randomly selected student has score A,
Need to find P (X 75). Use Z transformation to determine the values of Z associated with the values of X.
passed the exam? A score of 75 or greater is z = (75 - 57) / 10 = 1.8
needed to obtain an A on the exam. What Therefore, we need to find P (Z 1.8). Using Table A.1 in the Appendix,
percentage of the students received an A? P (Z 1.8) = 1 - P (Z 1.8)
= 1 0.9641 = 0.0359
Therefore, probability of student scoring an A is 3.59 %

1
 06-Mar-15

The accompanying table represent the weight in gram of moulded instrument display
panels. The samples were collected at half hour intervals . Prepare a tally sheet, of the
Answer
First, create a frequency histogram is to select the number of cells and the cell
individual measurements and then prepare a frequency histogram of the data, clearly boundaries. Since the data are integer values, the cell boundaries can be set at xx.5,
labelling the cell boundaries. Comment on the shape of the distribution. xx+ cell width + 0.5, etc. The range of the data is 20- 7=13 so using 13 cells might
Sample X1 X2 X3 X4 X5 work well. Thus, the integer values are the cell midpoints and the cell boundaries
1 14 15 13 14 13 are set at 6.5, 7.5, 8.5, ... , 20.5. For example, the boundaries for the cell with
2 20 18 14 17 8 midpoint 13 are 12.5 and 13.5.
3 14 17 14 11 14 The next step is to make a tally of the frequency of occurrence of the data appear
4 15 16 11 18 14 within each cell, as follows:
5 9 17 18 13 12 Cell Midpoint Frequency per Cell fi
6 19 15 14 15 16 7 1
7 16 13 14 13 17
8 14 17 9 16 15
8 11 2
9 14 14 12 13 13 9 1111 4
10 15 13 17 14 16 10 0
11 18 18 16 15 11 11 11111 5
12 20 12 13 17 14 12 111111111 9
13 1 8 9 12 7
14 12 14 16 14 20
13 11111111111 11
15 18 17 12 19 18 13.5 - 14.0 - 14.5 111111111111111111 18
16 19 17 16 16 17 15 11111111111 11
17 14 13 15 16 18 16 111111111111 12
18 14 17 12 16 11 17 11111111111 11
19 18 15 16 15 12
20 15 9 12 13 20
18 111111111 9
19 111 3
20 1111 4
Total, fi = 100

Answer continued
What is Statistics?
Next, this information is used to plot the histogram (students please plot the What is it Not
histogram):
8 12 16 20
Frequency
4
Has Something to Do With Data.
Objectives of Data Collection
8
12
16
Weight in gram Understanding, insights, illumination
From the shape of the histogram, it appears that the data came from a
An Inexact Science Given Industrial Realities
process that might be considered as a
candidate for representation by a normal distribution.

Probabilities in Manufacturing

Examples with objectives

classifying parts as being defective or non-defective
-- reducing number of defectives
studying the number of monthly orders received
better adjusting inventory levels to match orders
measuring gas output when acid concentrations are
changed --better predicting and controlling gas levels

2
 06-Mar-15

Statistical Thinking & Modelling

Engineers Think
Deterministically Engineering Method
Depends on Data
Deterministic Models Do Not Real Data Exhibits
Variability
Explain Variability
Obscures Ability to Make
Deterministic Models Do Not
Sound Decisions
Account for Variability

Engineers Must Learn to

Engineering Education/Practice
Think Statistically
Blames
Understanding of Risk and
Factors That Remain a Mystery
Uncertainty
Limitations in Measurement
Key is Discovering Sources
Process
of Variability

Data Collection
Statistic types
What is the fundamental purpose?
What important questions need answers? Deductive statistics describe a complete data set
What is the characteristic of interest?
How will it be measured? Issues
Inductive statistics deal with a limited amount of
What is known about the measurement process? data
How does engineering model impact data collection?
What data does the model require?
How robust is the model to data error?
How do model parameters support problem solution?
Are there physical constraints that impede ability to
collect data?

3
 06-Mar-15

Types of data` Within vs. between subgroup variation

Variables data - quality characteristics that are Under ideal conditions: only common cause
measurable values. variation occurs within subgroups and special
Measurable and normally continuous; may take on any cause variation occurs among between
value. subgroups
Attribute data - quality characteristics that are
observed to be either present or absent,
conforming or nonconforming.
Countable and normally discrete; integer

Descriptive statistics
Care must be taken so that differences in Measures of Central Tendency
operators, machines, lots of raw materials . Describes the center position of the data
Special
production Causes
lines Should
do not showOccur
up within Mean Median Mode
Between
subgroups. Batches not Within
Do different control charts for different Measures of Dispersion
operators if operators make a difference. Describes the spread of the data
Range Variance Standard deviation

Measures of central tendency: Measures of central tendency:

Mean Median - mode
Arithmetic mean x = N Median = the observation in the middle of sorted
1
N
xi
i 1
data
Mode = the most frequently occurring value
where xi is one observation, means add up
what follows and N is the number of
observations
So, for example, if the data are : 0,2,5,9,12 the mean
is (0+2+5+9+12)/5 = 28/5 = 5.6

4
 06-Mar-15

Median and mode Measures of dispersion: range

The range is calculated by taking the maximum
100 91 85 84 75 72 72 69 65
value and subtracting the minimum value.

Mode
2 , 4 ,6 ,8 ,10, 12 , 14
Median

Mean = 79.22 Range = 14 - 2 = 12

Measures of dispersion: Measures of dispersion:

variance standard deviation
Calculate the deviation from the mean for every The standard deviation is the square root of the
observation. variance. The variance is in square units so the
Square each deviation standard deviation is in the same units as x.
Add them up and divide by the number of
observations
n
( xi
n
( xi

i 1
n
i 1
n

Standard deviation and curve

Chebyshevs theorem
shape
If is small, there is a high probability for getting a If a probability distribution has the mean and the
value close to the mean. standard deviation , the probability of obtaining a
value which deviates from the mean by at least k
If is large, there is a correspondingly higher standard deviations is at most 1/k2.
probability for getting values further away from the
mean.
1
P ( x k
2
k

5
 06-Mar-15

Other measures of dispersion:

As a result
skewness
Probability of obtaining a value beyond x standard When a distribution lacks symmetry, it is
deviations is at most:: considered skewed.
2 standard deviations <0 left 0 = symmetrical >0 right
1/22 = 1/4 = 0.25
n
f i ( xi x) / n
3 standard deviations 3

a3 i 1
1/32 = 1/9 = 0.11
4 standard deviations
1/42 = 1/16 = 0.0625
s3

Other measures of dispersion: The normal frequency

kurtosis distribution
suggests peak-ness of the data
1 ( x ) / 2
f ( x)
a can be used to compare distributions 2 2

e
n
f i ( xi X) / n
4 2
a4 i 1
s4

The normal curve The normal curve

A normal curve is symmetrical about If x follows a bell-shaped (normal) distribution,
then the probability that x is within
The mean, mode, and median are equal
1 standard deviation of the mean is 68%
The curve is uni-modal and bell-shaped 2 standard deviations of the mean is 95 %
Data values concentrate around the mean 3 standard deviations of the mean is 99.7%
Area under the normal curve equals 1

6
 06-Mar-15

The standardized normal

=0
=1

x scale -3 -2 - + +2 +3

z scale -3 -2 -1 0 +1 +2 +3

7
 06-Mar-15

Critical Region, Critical Value,

Test Statistic and Decision Rule
and Significance Level

Type I Error The p-Value

A Type I error is the decision error when the researcher

incorrectly rejects the null hypothesis A p-value is the lowest level (of significance)
(when the null is true). at which the observed value of test statistic is
The probability of that error is a.. significant.
a. is the probability that the test statistic lies in them The p-value gives researcher an alternative to
critical region when the null hypothesis is true.
When the null is rejected, we say that the test is merely rejecting or not rejecting the null.
statistically significant at a 100 a % significance level. A small p-value clearly refutes Ho

Summary For Hypothesis Testing

State the null hypothesis H0: q = q0

Choose an appropriate alternate hypothesis Ha: q < q0,
q > q0, or q q0,
Chose a significant level of size a
Select the appropriate test statistic and critical region (if the decision is
based on a p-value, the critical region is not necessary) and state the
decision rule in terms of the test statistic
Compute the value of test statistic from the sample data
Reject H0 based on the decision rule (if the test statistic is in the critical
region or if the p-value is less than a): otherwise do not reject H0

8
 06-Mar-15

Tool maintenance Objectives:

Functions:
1. to ensure products or components produced Preventive measure, reduce defectives and
as specified by quality in sizes, dimensional, average time between defectives
thickness, finishes and etc. Reduce cost to min.
2. to ensure cost is affordable to consumers and
producers. Cost will be relatively low if low non- Predictive, analysis, diagnose maintenance on
conformities can be reduced and standards scheduled
quality in maintenance of the machines or
equipments increased. Ensure life-span of equipment
3. Optimised usage of machines as planned.

Types of maintenance: Maintenance cost vs Repairing cost

Examining
Services
Preventive
Corrective
Repairing

9
 06-Mar-15

Preventive maintenance Machining efficiency

Days++++ operator Machining efficiency (ME) =
Weekly ___technician Performance/Target x 100%
Monthly **** supervisor / engineer Machining capability = ME (ppm) x operating
Yearly &&&& engineer hour (min) x machining efficiency
Process Capability of machining
Cp>1.67 Excellent, 1.67<Cp<1.33 Good,
1.33<Cp<1.0 OK, 1.0<Cp<0.67 Less good.
Cp<0.67 Not functioning

10