35A00210 Lecture 11

Operations Management Statistical process control

Lecture 11
Statistical process control Quality control
Quality control
Process capability
Acceptance sampling

How is control part of managing

Quality control
quality?
Goal is to ensure that processes function as
Managing quality planned
- basic questions are; ”is process controlled at which phase?”, ”how
often?” and ”in which way?”

Designing quality Quality control Repairing before Repairing during Quality built into
and after production process the process
Product design Production
Process design Inspection/packaging Acceptance Process Continuous
Sourcing and purchasing Delivery sampling control improvement
Field organization
”Traditional” ”Modern”

OM2013 - 11 3 OM2013 - 11 4
 Many methods in quality control How to use different tools?

1. Identifying of problem
- customer complaints or control charts as starting point
2. Data gathering
- check lists, graphs, histograms as tools
3. Data analysis
- e.g. pareto-analysis as a good place to begin

Correlation diagrams 4. Finding the causes of problems

- e.g. cause-effect analysis
5. Developing and implementing solution
6. Continuous control and improvement of
processes
OM2013 - 11 5 OM2013 - 11 6

Quality control tools 3 standard deviations as a base for

- process control charts - analysis
Statistical process control (SPC)
- valuation is based on samples • Most of the values
Process is under control when there’s only random around the mean
variation (there’s always some variation) • 99,74 % of the values
- for non-random variation there is usually an assignable cause that should be within ±3 standard
eliminated deviation from the
- SPC doesn’t reveal the reason, that management and employees job! mean
- random variation can be reduced only by redesigning the process
Variation is random when sample values are inside the
so called control limits
- limits usually +/- three standard deviation which means 99.74% trust
aver.
- if values are outside the limits, the process is probably not under control
–3 –2 –1 +1 +2 +3
One of the most difficult decisions for management is to
68.26%
decide if the process needs to be changed 95.44%
- task is to reduce both random and non-random variation 99.74%
OM2013 - 11 8 OM2013 - 11 9
 Random and assignable variation Using process control charts
aver. Sign that there might be something
wrong with the process
(highly unlikely that the sample mean would be
outside control limits if the process is in control)

Find out the

x+3 reason!
Upper control limit

Every process has some amount of random variation

Hypothetical 99.74%
process mean
aver. aver. aver. x t

x-3
Lower control limit
Random
One sample variation
mean =
Distribution shifted Distribution increased Distribution skewed process under
control
Three main types of assignable variation

OM2013 - 11 10 OM2013 - 11 11

Different characteristics are

Using process control charts
analyzed with different charts
Process is not under control when...
Characteristics Variables Attributes
- one sample result is outside the control limits
Dimensions X
- two following sample results are near the same control limit Shape X
- 5 following sample results are at the same side of average Durability X
- 5 following sample results are forming a trend either up or down Surface appearance X
- level of sample result is changing a lot very rapidly Packaging X
Design X
- other non-random behavior of sample results
Using three standard deviation is recommended
but it is possible to use others if needed
- if control limits are too tight (e.g. ± 2 stdev.) normal variation is seen
as ”out of control” -situation (mistake I)
- if control limits are not tight enough (e.g. ± 4 stdev.) ”not in control” -
situation is seen as normal variation (mistake II)
- using three standard deviation control limits balances mistakes I ja II
Teapot quality characteristics
OM2013 - 11 12 OM2013 - 11
 Why do you need both X- and R-
Measuring continuous variables
chart?
X-chart
- is used to analyze the development of averages of continuous variables
Samples’
- the average (“middle line”) of X-chart is calculated from the sample averages
distributions
because the real average of process is usually not known
- processes real standard deviation is usually not known, so x-charts control
limits are calculated with samples’ average range (process variability increases) (process mean moves)
UCL UCL
- range (R); difference on largest and smallest value in the sample

X-chart
– the size of the sample is considered in control charts, relatively
small samples are preferred to limit time delays
- if deviation is known, may be calculated by using the basic formula;
LCL LCL
Variability is not exposed Shift is exposed
UCL UCL
R-chart

R-chart
- is used to analyze the internal variability of samples
- the real standard deviation is usually not known and therefore range R (min-
max) is used instead
LCL LCL
- the results are almost the same Variability is exposed Shift is not exposed
OM2013 - 11 14 OM2013 - 11 15

X- and R-chart example X- and R-chart example

Finnish Washer Ltd. produces metal sheets, that are used as components for 1. Calculate sample mean and range as well as average mean and range
different kinds of machines. The critical measure is metal sheet hole’s diameter,
so that the component can be used for its designed purpose. Quality control has
taken 10 samples during 10 days, each consisting of 5 sheets (measurements
below). Find out with the help of SPC if the process is under control. Create
control charts of the machine operation. Give short argument for your answer.

OM2013 - 11 16 OM2013 - 11 17
 X- and R-chart example X- and R-chart example

2. Calculate control limits for X- and R-charts 3. Plot sample values and control limits
Values for X- and R-charts

NOTICE!
n = sample size
4. Interpret results and make conclusions/recommendations
When one has to use samples’ When one has to use samples’
average range R as standard deviation average range R as standard deviation • Mean not in control; one over control limit, rising trend etc.
every control limit has its own formula one needs to use tabular values when
that has to be memorized calculating control limits • Variability in control

OM2013 - 11 18 OM2013 - 11 19

Analyzing process charts Measuring attributes

Graph Description Possible causes

p-chart (quantity of faulty ones within the sample)
- all the variables are not continuous. P-chart is used when
observations can be divided into two groups
Normal Random variation
- functioning vs. not functioning, good vs. bad, acceptable vs. not
acceptable
Assignable causes e.g. - quantity is usually given as a percentage
Unstable tools, materials, people,
overreacting etc.

E.g. wearing of machine, c-chart (quantity of failures per unit)

Trend employee getting tired, - used when only observations can be counted (”not observations”
better working methods can’t be counted)
- phone calls, complaints, breakdowns etc. per time period
Different working shifts,
Cycle - scratches or dents per product etc.
variation of electricity,
periodical changes - cannot be given as a percentage

OM2013 - 11 20 OM2013 - 11 22
 p-chart example p-chart example
Some activists have complained to the city council that all the city residents should have an
equal right to feel safe. They think that number of policemen and investments preventing crime 1. Calculate sample probability p
(e.g. lights at streets) should be made on relative basis so that so called problem areas should
have more attention than safe areas. To investigate the complaint, the city council acquired 2. Calculate mean probability p
data of crimes committed during the last 30 days (the city had been divided into twenty 5000
resident areas). In each area the sample size was 1000 persons. What kind of advice would
you give to city council for using the information to allocate the recourses? Use 95% level
(z=1.96) so that your recommendation would be correct with high probability.
(Base your analysis on quality control methods)
3. Calculate sample’s standard deviation

NOTICE! n = sample size

4. Calculate control limits

Typically z=3 but in this case z=1,96 because the

OM2013 - 11 23 OM2013 - 11 question asked for only “95% confidence level”. 24

p-chart example How to use different tools?

5. Plot sample values and control limits 1. Identifying of problem

- customer complaints or control charts as starting point
2. Data gathering
- check lists, graphs, histograms as tools
3. Data analysis
- e.g. pareto-analysis as a good place to begin
4. Finding the causes of problems
- e.g. cause-effect analysis

6. Interpret results and make conclusions/recommendations 5. Developing and implementing solution

More investments to areas 6, 16 and 20 6. Continuous control and improvement of
Reduce investments from areas 2, 10, 17 and 19 processes
OM2013 - 11 25 OM2013 - 11 26
 Quality control tools How to use different tools?
- correlation diagrams -
Effective graphical presentation style when
1. Identifying of problem
there is a clear cause-effect relationship
- customer complaints or control charts as starting point
2. Data gathering
12
Number of problems

- check lists, graphs, histograms as tools

10
3. Data analysis
8
- e.g. pareto-analysis as a good place to begin
6
4
4. Finding the causes of problems
- e.g. cause-effect analysis
2
0 5. Developing and implementing solution
0 10 20 30 6. Continuous control and improvement of
Training hours
processes
OM2013 - 11 27 OM2013 - 11 28

Continuous improvement as key to Process improvement also visible

success in charts
Activities must be improved all the time
- observing and questioning performance essential
Control Process improvement
Employees have central role in developing
operational quality
- management has many ways to get employees committed
- culture change, customer focus, team work, empowerment, training,
awards, incentives

Plan
process
Quality

Deming people
Act cycle Do
machines
material
Check
Time
OM2013 - 11 29 OM2013 - 11 30
 Lecture 11 Process capability
Statistical process control
Process capability means ability to maintain
certain product / process specifications
(often defined by customer)
- focus is on random variation (common causes), assignable causes
presumed to be fixed
Process capability Measured by capability index

- index takes into account if the distribution average shifts

- 1.33 is considered as a limit of good process (99,9936% good)
Capability is often improved by improving the
process itself
OM2013 - 11 32

Everything has its own tolerance

How well should process function?
limits
Tolerance limits are set design specifications In general 99% sounds good, but in many cases
that define the acceptable range for a part, it’s not
product or service - 3,5 days a year without electricity
- 15 minutes of undrinkable water every day
- 15 minutes without phone and television every day
- 20.000.000 wrong prescriptions a year (USA)
- over 15.000 new born babies dropped by doctors a year (USA)
lower upper - 5.000 failures in medical operations per week (USA)
toleranc toleranc - 3 failed landings at Heathrow every day
e e
level level - 20.000 lost letters (by post) every hour (USA)
LTL aver. UTL According to research, basic processes of
700 h. 1000 h. 1300 h.
human beings are on 99,86% (3,2 sigmas) level
e.g. light bulbs lifespan

OM2013 - 11 33 OM2013 - 11 34
 Process capability
Sigma-levels are not linear
and number of good products

6 sigma level
tolerance limit is six
standard deviation
from the mean
99,999999% good
4 sigma level
99,9937% good

2 sigma level
95,45% good

LTL aver. UTL

OM2013 - 11 35 OM2013 - 11 36

Process capability is increased

Why is 6 sigma so desirable?
by improving the process
Probability of product/process able to meet specifications?
LTL UTL LTL UTL
Cpk=0,57 Cpk=1,00
4,3 = 100 1,7 3,0 = 100 3,0

95.54 % 99.74 %

700 1130 1300 700 1000 1300

Cpk=1,13 Cpk=2,00

LTL UTL LTL UTL

8,6 = 50 3,4 6,0 = 50 6,0

99.97 % 99.999 %

700 1130 1300 700 1000 1300

OM2013 - 11 38 OM2013 - 11 39
 Sigma levels and capability have a
Six sigma as a mgmt. philosophy
relationship
Statistic oriented improvement concept that
z z
looks more and more like TQM
- employees improve processes, focus on customer, decisions based
on facts, quality controlled with statistical methods etc.
LTL m UTL Emphasis originally more on eliminating errors
- ”zero defects”, ”fix in the first time”, higher yield, lower costs...

(±)
Required process capability is 1,5
- based on ”typical” 1,5 std. movement of mean in the process
- number of defective 3,4 for every million produced
Limited number of wide success stories

OM2013 - 11 40 OM2013 - 11 41

Lecture 11 Quality control tools

Statistical process control - acceptance sampling -
”Traditional quality control method”
- batch’s quality is ensured by inspecting a sample of the batch
Batch is accepted if the sample has a low
enough amount of defectives
Acceptance sampling - rejected batches are repaired or obliterated
Method has many disadvantages
- expensive from total life-time cost perspective
- assumes that a certain amount of defectives is acceptable
- sampling method may lead to wrong decisions (limited information)
Used in some companies
- easy process, low direct costs (cheaper than investigating all items),
only method for products that ”perish/break” in inspection
- need to decide sample size and acceptable quality level
- motivates manufacturer to produce good quality
OM2013 - 11 43