Statistical Process Control

Total Quality Manufacturing, Pg.D.MM 506

Faculty of Graduate Studies

University of Colombo

Manone Perera
February 27th, 2022
Quality improvement tools
✓ In the 1930's and 1940's Shewhart and Deming
developed a set of statistical quality control
(QC) tools
✓ In the 1950's these tools were applied in Japan
✓ In the 1960's they were developed into the
seven QC tools

The seven QC tools

✓ Cause and effect diagram

✓ Run charts/Control charts
✓ Scatter diagram
✓ Flow charts
✓ Pareto charts
✓ Histogram
✓ Tally sheets
New quality tools
A few new quality tools were developed in the
early 1970's. These differ from the seven quality
tools as follows

Seven Quality Tools New Quality Tools

Reactive Proactive
Results based Process based
Numerical Non - numerical
Quantitative Qualitative

Statistical Language based

Affinity diagrams
✓ The affinity diagram is a facilitated brainstorming
technique used by groups of people
✓ Ideas are written on "post it notes" and grouped
into clusters of common themes
✓ Each theme is given a title
✓ The tool is useful for idea generation and
grouping but not for detailed problem solving
Inter relationship diagram

✓ This tool is used to identify logical strings of connections between several

problems or issues
✓ The aim is to reach consensus about root causes and the order in which to
proceed
✓ Activities are connected by arrows linking what leads to what
✓ Those items with most arrows pointing to them are important milestones or
bottle necks
✓ Those items which most arrows leading from them are important root causes
✓ This is a simple, effective technique that allows teams to develop ideas and
approaches to tackle improvements
Tree diagram
✓ The tree diagram arranges goals, problems or customer
requirements into a hierarchy
✓ It shows how a problem or goal is broken down into more
detailed sub problems/goals
✓ It differs from the affinity diagram in that it is built top
down
✓ The technique is used to gain an insight into detailed
actions required to achieve goals
✓ Each sub item has only one parent and, as a rule of
thumb, there should not be more than 7 sub items for
each parent
Statistical Process Control
What is SPC ?

✓ SPC is a set of problem solving tools to

achieve process stability and improve
process capability through the reduction
of variability
✓ It is a strategy for continuous
improvement which will only succeed
through management commitment

Benefits of SPC

✓ Quality improves
✓ Yields improve
Role of SPC
✓ Evidence of what the process is doing
✓ Prediction of what the process is likely to do
✓ Assessment of what the process is capable of
✓ Indication of when to look for trouble and when not to
✓ Clues as to where trouble is likely to occur
✓ Helps to understand the operation of the process
✓ Helps to make improvements to the process
Focus of SPC

✓ SPC focuses attention to the process

✓ SPC focuses on prevention rather than deterioration

SPC and Quality

In order to meet customer requirements the product must be

manufactured in a stable and a consistent process

History of SPC
1920's Shewhart introduced control charts in Bell Telephone
Laboratories, USA
1930‘s SPC is used extensively in the USA
World war II SPC becomes critical in the ammunition industry
Post world war II Unparalleled demand for consumer goods in USA

✓ SPC fell into disuse

✓ Quality plummeted

1950’S SPC introduced into Japan

NO PROCESS IS ENTIRELY FREE FROM VARIATION
✓ Common Causes
✓ Special or assignable Causes

Common Causes (chance/system causes) - The contribution to variation which

are random in nature, the sum and complex interaction of many effects

Some examples
✓ Small fluctuations in operating environment (temperature, humidity,
barometric pressure)
✓ Small equipment vibrations
✓ Small variations in materials used
✓ Passing traffic
✓ Small electrical fluctuations

Special Causes (assignable causes) - Sources of variation for which specific

reasons exist giving rise to excessive variation causing the process to be "out of
statistical control" beyond expected random variation

Some examples
✓ Machine wear, drifting out of calibration
✓ Inferior batch of raw materials
✓ Poorly trained operators
✓ Incorrect work method
Can we do the job properly ?
✓ Is the process stable (is it in control) ?
✓ Are there any special causes ?
✓ Is the variability due to common causes
only ?
✓ What is the extent of the variability ?
✓ What is the natural capability when we
only have common causes ?

Who takes action ?

✓85% of causes are common causes and
require management action
✓15% of causes are special causes and
require action by shop floor operators and
supervisors
 Control Charts
✓ Provides a picture which the operators can interpret
✓ Detect the presence of variability at an early stage
✓ Provide a simple means of controlling the average and
spread of a process

Example
A department produces metal shafts, they need to monitor
the length of metal shafts

✓ Process distribution mean (PDM) = 8.520 cm

✓ Process standard deviation (PSD) = 0.00775 cm
✓ Periodic sampling, sample size (n) = 5
✓ Sampling done every 0.5 hour
✓ Sample 1 data = 8.526 cm, 8.511 cm, 8.528 cm, 8.518 cm,
8.513 cm
✓ Sample 1 mean (SM) = 8.519 cm
 Control Charts
Sample Sample mean (SM)
1 8.519
2 8.517
3 8.522
4 8.520
5 8.522
6 8.515

Plotting of sample means

 Control Charts
✓ Sampling distribution (sampling distribution of means)
✓ Process distribution
✓ Normal distribution or bell curve

SDM = 8.520 cm
SDSD = 0.00347

The central limit theorem

✓ Sampling distribution mean (SDM) = Process distribution mean
(PDM) = 8.520 cm
✓ Sampling distribution standard deviation (SDSD) = Process
distribution standard deviation (PDSD i.e. 0.00775 cm) divided by
the root of sample size (n=5)

Therefore,
✓ Sampling distribution mean (SDM) = 8.520 cm
✓ Sampling distribution standard deviation (SDSD) = 0.00347 cm
 Control Charts
✓ Sampling distribution mean (SDM) = 8.520 cm
✓ Sampling distribution standard deviation (SDSD)
= 0.00347 cm

SDM = 8.520 cm
SDSD = 0.00347

Exercise 8.522
✓ Compute probability of a particular sample
mean being greater than 8.522 cm
✓ Compute probability of a particular sample
mean being less than 8.515 cm
 Control Charts

Internally calculated limits

Internally calculated limits

 Control Charts
Rules for taking action
✓ Any point beyond the upper/lower control limit
✓ Seven consecutive points on one side of the mean
✓ Seven consecutive points, all increasing or decreasing
 Control Charts
Exercise
Compute upper control limit (UCL) and lower control limit (LCL) if we
did use 5% to position control limits

CL = 8.520

SDM = 8.520 cm
SDSD = 0.00347
UCL = 8.5132 UCL = 8.5268
 Control Charts
Type 1 error - Concluding that a process is out of control when, in
fact it is in control

Type 2 error - Concluding that a process is in control when actually it

has moved out of control
 Control Charts
Process variation can only be measured in
statistical terms. Data can be classified as either
continuous (or variable) or discrete (or attribute)

✓ Continuous (or variable) can be measured on

continuous scale, decimal values are possible
such as length, weight, voltage, temperature,
viscosity etc. (NORMAL DISTRIBUTION)

✓ Discrete (or attribute) often takes the form of

counts such as the number of loan applications,
number of emergency room arrivals that have
to wait more than 30 minutes to receive
medical attention etc. (BINOMIAL or POISSON
DISTRIBUTION) C
20
Control Charts

C
 Control Charts
Exercise
Compute Control Limit (CL), Upper Control Limit (UCL) and
Lower Control Limit (LCL)
Control Charts for Continuous Data
 Control Charts
Table: Control Chart Constants
 Control Charts for Continuous Data
Sampling mean is
shifting upward
but the range is
consistent

These sampling distributions result in the charts below

UCL X - chart
X-chart detects shift in
central
LCL
tendency

UCL R - chart does

R-chart not detect a
LCL change in mean

25
 Control Charts for Continuous Data

Sampling mean
is constant but
dispersion is
increasing

These sampling distributions result in the charts below

UCL X-chart does not

detect an
X-chart
increase in
LCL
dispersion

UCL R-chart detects

R-chart an increase in
LCL dispersion

26
Control Charts for Discrete Data
✓ P - charts are used for proportion of sampled parts that are
defective
✓ C - charts are used to monitor the number of times a
defect occurs in a single unit

27
 P Charts
P charts or proportion charts are used to monitor the
proportion of items in a process that fall into a certain
category

Exercise: suppose n = 50, m = 25, z = 3

Sample, i No. of defective units, di Fraction defectives,

pi = di/n
1 2 0.04
2 0 0
3 6 0.12
4 1 0.02

25 2 0.04
∑ 1.02
 C Charts
C charts are used to monitor the number of times a characteristic
(such as a defect) occurs in a single unit. A unit is whatever is being
inspected such as a single part, a single product, a batch of
products or even a physical space or region

Exercise: suppose n = 1, m = 20, z = 3

Shaft No. Imperfections, c Shaft No. Imperfections, c

1 3 11 6
2 5 12 8
3 4 13 5
4 2 14 4
5 7 15 7
6 5 16 2
7 9 17 4
8 6 18 5
9 3 19 7
10 4 20 6
∑ 102
 Use of Control Charts
✓ Once the limits for the R and X bar charts have
been determined, the limits are held fixed for use
thereafter to track the process

✓ Ordinarily both the R and X bar charts are

employed together and displayed in one place
as both the mean and the standard deviation of
a process are important

✓ Use of the control charts involves periodically

taking a new sample, computing the mean and
range and plotting them on the charts

✓ Whenever points appear non-random or lie on

or outside of limits, the process is stopped to
search for a special cause
 Use of Control Charts
✓ Once a control chart has been constructed for a
process, newly sampled data are used only to
assess the state of the process; the new data are
not used to change the control limits

✓ The control limits are changed only when the

process has been intentionally changed or
thought to have changed inadvertently

✓ If the control limits must be changed, then the

process is initiated anew to determine the new
limits
 Process Capability
To determine whether a process is capable of producing output that meets
customer requirements, when the process is in statistical control

Control limits
Computed as a mathematical function of the estimated mean and range

Specification limits
Set by design engineers, customer requirements or other expectations

Scenario 1: Process Capability Index

Assuming the process mean (PM) is centred between USL and LSL

Note: the process standard deviation (PSD) can be estimated to be R/d2

Exercise 1
If the specification width is 6 cm, compute Cp for PSD of 1 cm, 0.5 cm and 2
cm
 Process Capability
Scenario 2: Process Capability Index
If the process mean (PM) is not centred between USL and LSL

Cpk = min (Cpu, Cpt)

Cpu = USL - PM
3 x PSD

Cpt = PM - LSL
3 x PSD

Note
if PM and PSD is not possible to find, use estimated parameters

Exercise 2
USL and LSL are 7 cm and 2 cm respectively, PM = 5 cm, PSD = 1
cm, compute Cpk

Exercise 3
For previous example, compute process capability if shaft
specification limits are 8.520 + 0.010 cm