Sunteți pe pagina 1din 38

Statistical Process Control

An Overview

1
THE GOAL

2
OUTLINE OF PRESENTATION
• History of SPC
• What is SPC?
• Benefits of SPC
• SPC explained

3
History of SPC
• 1920: Shewart of Bell Labs develops SPC as
economical means to control quality in
manufacturing
• 1940: Bell Labs train army staff on SPC during
World War II.
• 1950: Deming applies SPC concept in Japanese
industry
• 1960-70: Quality blossoms in Japan
• 1980: “Made in Japan” A Quality symbol
• 1990: SPC widely accepted in all Industries as
Process improvement Tool.

4
Philosophy of SPC
• Variation is the enemy and it is inevitable
– Inherent, random and common
– Special, assignable cause
• Real Quality comes from preventing defect and
not detecting defect.
• Prevent defects by monitoring, controlling &
reducing variability
• Continuous process improvement to reduce
variability

5
Tools for Continuous Improvement
– Cause & Effect – Process Flow diagram
– Brainstorming – Scatter diagram
– Deming’s 14 point TQM – Design of experiments
– Quality Circle – Trend charts
– Process teams – Pareto Analysis
– Check sheets – Histogram
• – Process capability
– SPC

6
Statistical Process Control
What is it?

• Statistical: With the help of data (numbers)


• Process: The process we intend to study
• Control: To make the process behave the way the
CUSTOMER wants it

7
Different thinking
• Product Quality
– The process of controlling quality following
production. Inspection, Rejection and
Rework

• Process Quality
– The process of controlling during
production. Reduces Inspection, rejection
post production and rework. Defect
Prevention.

8
Benefits of SPC
• Increase profits by
– Reduce scrap
– Reduce rework
– Increase production
• Better understanding of process
• Increase customer satisfaction
• Documents and tracks improvement
• Boosts employee morale

9
What does SPC answer?
• How capable is the process of meeting customer
(internal or external) expectations?
• What is the distribution of process output?
– centering, range or “spread”,
likelihood of an extreme value
• What is causing process variability?
• Can we afford to minimize the variability?
• Has the process remained the same over a period
of time?
• When should we alter the process?

10
Sources of Variation
4M’s and OE

• Machine
• Material
• Method
• Measurement
• Operator
• Environment

11
Variation
• Random, chance, common, unknown cause - the
rhythm of the process



• Special, assignable causes - something has
changed

12
Special Variations
Enemy of Process
• Special causes may be due to
– machine troubles
– faulty measuring device
– operator losing control
– fatigue
– change in raw material

13
SPC in Nutshell
• Describe the distribution of the process
• Estimate the limits within which the process operates
under ‘normal conditions’.
• Determine if the process is stable. Sample the output of the
process and compare to the limits.
– Decide:
• Process appears to be ok. No action required
• There is reason to believe that something has changed
and look for the source of change
• Continuous process improvement

14
Distribution of Process Output

Neither accurate nor precise

Precise but not accurate

Precise & Accurate

15
Distribution of Process Output
Frequency Distribution – a tally of process output falling
into specific categories
Histogram – a pictorial representation of the frequency
distribution

Histogram

10
9
8
7
6
5
4
3
2
1
0
1500

2000

2500

3000

3500

4000

4500

5000

5500

6000

6500

7000

7500

8000

8500
Strength

16
Estimating Process Limits
• Central Tendency – arithmetic mean (average)
• Dispersion – range or standard deviation
• Mean

Range 17
Estimating Process Limits

Process Centering

n
Xi
Mean or Average X = 
n
i=1

An estimate of the population mean 

18
Is Average Good Enough?

Target 8 mm

10.5 8.1
5.5 7.9
9.2 8.15
6.5 7.85

X = 7.93 X = 8.01

We need a way to quantify the ‘spread’


of data 19
Estimating Process parameters
Process Spread


n
Standard Deviation s =  ( xi – X )2
i=1 (n-1)

Range r = X max - X min

Both are estimates of the population standard


deviation 

20
Are Standard Deviation and Mean Magic
Numbers?
• Mean and Standard deviations are really magic
numbers.
• We could predict the amount of product that can
be produced within the specifications or within
the given range
• This can be done using an equation to describe the
relation

21
Distributions
• Exponential
• Weibull
• Gamma
• Normal
• The normal distribution is the one which is
relevant to our context (for the time being!!!)

22
Normal Distribution
-(x- )2
1
F(x) = e 2 2

  2

yc ne uqerf

 - 3   + 3 23
How do we use SD & Mean

By calculating mean and standard deviations and


knowing that they are normally distributed
(central line theorem) we know everything about
the distribution – WE CAN PREDICT

But the process has to be STABLE

24
Statistical Control
• A process is in ‘control’ when a stable
system of chance causes seems to be
operating
• Capability and control are quite distinct.
However they are related in continuous
improvement
• First ensure consistency and stability and
then take care of capability issue
• It is unprofitable for an organization to
search for assignable causes of
variability
• ‘Out of control’ means assignable causes
of variation must be present
25
Statistically Rare Event
• An outcome with a very small probability of
occurrence (odds are against it)
• In manufacturing, an outcome that is ‘statistically
rare’ usually signals the presence of assignable
causes of variation
• How can we figure out what is statistically rare
and what is common?

26
Statistically rare event
Obtaining a sample value that is ‘very far away’ from the
average is a highly unlikely event
In SPC the ‘far away limits are  +/- 3

99.73%

 - 3   + 3 27
The Control Charts
• Developed by Dr. Walter Shewart of Bell
Telephone Labs in the 1920’s
• Emphasis was on developing procedures to
economically control quality
• Mathematical and Statistical theory provides the
foundation

28
What is use of control charts?

UCL


 X

Time LCL
29
Control Charts
• Two broad categories
– Attributes
• Defective parts
• Go/ no-go data
– Variables
• Measurements, temperature, weights etc

30
Control Charts for Variables
• Process Centering
– Mean ‘X’
– Individuals
• Process Spread
– Standard deviation ‘s’
– Range ‘R’
– R charts are quite common due to ease of
calculation

31
How to draw Control Charts?
• Decide the variable (temperature, energy, weight etc)
• Should be easily measurable
• Data Collection – Important step
• Data ‘cleaning’, remove unwanted observations, irrelevant
• Decide sample size and subgroup size
• Preferred sample size is minimum 20 and subgroup size
can be between 2 to 10
• Have an eye on the data collection costs

32
Data Representation
• Record measurements of k samples of size n.
• Example energy readings of all BTS sites (k) for a
quarter (n).

S u b g ro u p s iz e
S a m p le s 1 2 3 n
1 2 3 .2 22 5 .8 02 6 .4 0x x .x x
2 2 4 .1 52 3 .8 02 5 .7 4x x .x x
3 2 6 .1 52 6 .7 02 5 .6 0x x .x x
k x x .x xx x .x xx x .x xx x .x x

33
Calculations for X Chart
• For each k samples, calculate the mean Xi and the
Range Ri for all i= 1,2,…k
• Compute the overall average X and the average R
• k

X=  Xi
i=1

k
k
R=  Ri
i=1

34
Calculations for X Chart
• Calculate the Upper Control Limits (UCL) and
Lower Control Limits (LCL) and the R chart as
follows
X-chart R-chart

UCL X + A2R D4R

LCL X - A2R D3R

Coefficients A2, D4 & D3 for Variables chart are


available.

35
Interpretation of Control Charts
• Plot the X-chart and R-chart.
• If all Xi and R i values fall randomly within
control limits, the process is in statistical
control.
• However there are indications within control
limits. Look for
– Data points hugging the control limits
– Data points showing a cyclic trend
– Data points showing increasing trend within
the limits

36
Interpreting Control Chart

    
   
   
   
In Control Hugging control lines

 
   
 

  
 

Cyclic pattern Increasing Pattern

37
Control Chart Rationale
• Distance of Control Limits from grand average is
a measure of process variability

• X  A2 R is equivalent to Mean  3

•  can be approximated to R /d 2

• Hence alternative to X  A2 R is X  3  /  n

38

S-ar putea să vă placă și