GEOLOGY, RAW MATERIALS AND RAW MIX

Process Engineering Certification Program

April 7 - 8, 2003

Quality Assurance
 Session 1: Basic Analytical Methods

2
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Quality Assurance

Definition

All those planned and systematic actions necessary to

provide adequate confidence that a product or service
will satisfy given requirement for quality.
(ISO 8402)

3
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Quality Assurance

Quality Assurance

Quality Quality Quality

Planning Control Audit

4
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Quality Planning
Strategy and Tactics

ƒ Professional Organizations

External
ƒ Environmental Legislation
ƒ Standard Committees
ƒ Technical Customer Service
ƒ Market
ƒ Material Resources Control Concept
ƒ Product Design
ƒ Mix Design

Internal
ƒ Exploitation Concept
ƒ Control Concept
ƒ Education/Training

5
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Quality Control

ƒ Control Plan
ƒ Procedures/Methods
ƒ Personnel Organization
ƒ Routine Control (Raw Materials Cement)
ƒ Non-Routine Testing (Raw Materials, Refractories, Cement
Application)
ƒ Data Processing/Communication
ƒ Cost Control

6
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Quality Audit

7
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Significance of Quality Control

8
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Statistical Thinking the Basis of Quality Control

ƒ 1. Hypothesize - Problem definition, Variable ID, Plan changes

ƒ 2. Collect Data - Conduct trials, Do on small scale, Use DOE

ƒ 3. Analyze Data - Apply statistics, Significance?, Predict?

ƒ 4. Take Action - Return to step 1, Verify results, Implement

9
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Shewhart Method - Scientific Method

Does the Question/theory

theory Act Plan phase
need Plan a change or
revision? test
Change the Plan data
process collection

Compare data
to theory Check Do Carry out the
Summarize change or test on
what a small scale
was learned Observe the data
Analyze the data

10
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Cycle of Indecision

Plan Plan

Plan Plan

11
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Do - Do - Do- Do Cycle (Do-Do Happens!)

Do Do

Do Do

12
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 What is a statistic?

ƒ Definition: A statistic is a function of the observations in a sample

ƒ The computation of a statistic does not depend upon any unknown

constants called parameters!

ƒ “If the method used to obtain a statistic is good, then the resulting
numbers will be meaningful, interpretable, and relevant to the
problem or system being studied.” Dennis Keller

ƒ The object of taking data and computing statistics is to provide a

basis for action! Those responsible and accountable for taking
action must understand statistical methods in order to justify the
actions they recommend.
13
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Some Key Points to Remember

ƒ 1. Use meaningful statistics to summarize data in order to explore,

understand, describe, and quantify the underlying distribution or
population.
(Enumerative Study)

ƒ 2. Use meaningful statistics to summarize data in order to explore,

quantify, and predict future performance of a system.
(Analytical Study)

ƒ Why use statistics?

A. Separate “good” from “bad” data
B. Understand systems and processes
C. Improve systems and processes
D. Optimize the performance of systems and processes

14
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 What is an enumerative study?

ƒ Consider the following Raw Meal C3S GammaMetrics data taken at

different 3-minute time intervals:
Interval A B C D
1 130 169 162 150
2 155 168 158 146
3 166 166 154 143

ƒ Are there any difference between the C3S measurements? What

types of possible differences are there?

ƒ If there is a difference, how confident are we that this difference is

meaningful? Let’s take a vote on this? How many say there is a
difference? How many say there is not a difference?

15
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 What is an analytical study?
Decide what the future C3S values will be based on the
following results.
Raw Meal C3S via GammaMetrics

180
160
140
120
100
C3S

80
60
40
20
0
0 50 100 150 200
Time (minutes)

16
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Summarizing Data

ƒ Why summarize data?

- Characterize the distribution of the data in
Enumerative Studies
- Quantify features of the data in Analytical Studies
ƒ Measures of central tendency
- Sample average
- Sample median
ƒ Measures of dispersion
- Sample range
- Sample variance
- Sample standard deviation
- Coefficient of variation

17
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Central Tendency

ƒ Sample Mean or Arithmetic Average

- Data: y1, y2, y3, y4, y5, …, yn
- Σ yj /n
ƒ In Excel® enter the following instructions
- select the cell for the average
- suppose the data are in B2 to B362
- type
=AVERAGE(B2:B362)
- hit the “enter” key
ƒ The sample mean or arithmetic average is the “center of gravity” of
the data.
ƒ Exercise 1: What are the average C3S values for the 4 time
periods? What can you say about these four averages?

18
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Central Tendency (continued)

ƒ Sample Median or “Middle Value” of the data

- Data: y1, y2, y3, y4, y5, …, yn
- Order Statistic: y(1)< y(2)< y(3)< y(4)< y(5)< … <y(n)
- Calculate the median rank:
mr = (n+1)/2
- If n is odd, Median = P50 = Y[(n+1)/2]
- if n is even, Median = P50 = {Y[n/2] + Y[(n/2)+1] }/2
ƒ In Excel® enter the following instructions
- select the cell for the median
- suppose the data are in B2 to B362
- type
=MEDIAN(B2:B362)
- hit the “enter” key
ƒ The sample median is the 50th percentile of the data.
ƒ Exercise 2: What are the median C3S values for the 4 time periods?
What can you say about these four statistics?
19
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Central Tendency (continued)

ƒ Sample Mode or the most frequently occurring value in the data.

ƒ Referring to the C3S data presented earlier, there are no repeat
values. Thus rounding the data to the nearest tens the files can be
summarized as follows:

Interval A B C D
1 130 170 160 150
2 160 170 160 150
3 170 170 150 140
ƒ Note that there is no mode for the A interval of time. But the mode for
the remaining three time intervals are 170, 160, and 150
respectively.
ƒ The mode is not a very useful measure of central tendency so it will
not be considered in this presentation.

20
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Which measure of central tendency to use?
Average Median
Pro Best if Approximates
distribution is
average if
symmetric distribution is
symmetric
Pro Basis for most Not sensitive
decision to outliers
making
Con Influenced by Slightly more
outliers difficult to
calculate
Con Limited use in
decision
making

21
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Dispersion

ƒ Sample Range
- Data: y1, y2, y3, y4, y5, …, yn
- Order Statistic: y(1)< y(2)< y(3)< y(4)< y(5)< … <y(n)
- Calculate Range = R = Max - Min = y(n) - y(1)
ƒ In Excel® enter the following instructions
- select the cell for the range
- suppose the data are in B2 to B362
- type
=Max(B2:B362)-Min(B2:B362)
- hit the “enter” key

ƒ Exercise 3: What are the ranges of the C3S values for the 4 time
periods? What can you say about these four values?

22
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Dispersion (continued)

ƒ Sample Variance
- Data: y1, y2, y3, y4, y5, …, yn
- Mean-Square Deviations from the sample mean
- S2 =Σ [yj - yavg ]2 / (n-1)
ƒ In Excel® enter the following instructions
- select the cell for the Sample Variance
- suppose the data are in B2 to B362
- type
=(STDEV(B2:B362))^2
- hit the “enter” key
ƒ Sample Standard Deviation:
- S ={Σ [yj - yavg ]2 / (n-1)}½
- Root-Mean-Square Deviation from the average
ƒ Exercise 4: What are the standard deviations of the C3S values for
the 4 time periods? What can you say about these results?

23
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Dispersion (continued)
ƒ Process Capability Standard Deviation
- Data in Time Order: y1, y2, y3, y4, y5, …, yn
- Calculate the moving ranges
MRj = | yj - yj-1 | for j =2, …, n
- Spcl = Median (MRj) / 0.954
ƒ Typically the Root-Mean-Square standard deviation is larger than
the Process Capability standard deviation.
ƒ Exercise 5: Calculate the process capability standard deviation for
the C3S values assuming the time intervals are consecutive and the
values within a time period are in time order.

Data: 131, 155, 166, 170, 169, 166, 162, 158, 154, 151,
147, 143
How does this standard deviation compare with the individual
standard deviations from the four time intervals?

24
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Measures of Dispersion (continued)

ƒ Coefficient of Variation (CV)

- CV = S / Yavg
- The standard deviation used here can be the Root-
Mean-Square or the Process Capability standard
deviations
ƒ This statistic is used when a two-way plot of standard deviations
versus averages indicates that the two statistics are positively
linearly correlated.
ƒ Sometimes the CV is expressed as the percent change in a
property rather than the fractional change.
ƒ Exercise 6: What is the percentage coefficient of variation for the
C3S values assuming that the data are in time order? Using the
averages and standard deviations from the four time periods should
the standard deviation be expressed as a coefficient of variation?
25
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Which measure of dispersion to use?
2
R S S Spcl CV
Pro Easy to Un- Used in Better Good if
calc. biased most than S if avg and S
Calc. data are correlated
time
ordered
Pro Good for Best for Best for Not Easy to
1<n<6 sym. sym. dist. sensitive understand
dist. to shifts
Con Sensitive Units Not easy to Wrong if
to outliers meaning understand avg and S
-less are not
correlated
Con Not good Need a
if n>5 calc.
26
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Checking for Outliers in Small Sets of Data

ƒ Statistic for Testing for Extreme Values in Data

Statistic: (y(2)- y(1)) / ( y(n) - y(1) )
or
(y(n)- y(n-1)) / ( y(n) - y(1) )
ƒ Criteria for Testing for Extreme Values in Data
Exclude the value if the statistic above exceeds the
tabulated value with indicated confidence
n 90% 95% 98% 99%
3 .8 8 6 .9 4 1 .9 7 6 .9 8 8
4 .6 7 9 .7 6 5 .8 4 6 .8 8 9
5 .5 5 7 .6 4 2 .7 2 9 .7 8 0
6 .4 8 2 .5 6 0 .6 4 4 .6 9 8
7 .4 3 4 .5 0 7 .5 8 6 .6 3 7

ƒ Exercise 7: Are there any outliers for C3S values within the 4 time
periods? 27
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Least Significant Difference (LSD)

ƒ Statistic for Testing for Extreme Values between Individual

Measurements or Averages

Statistic: | Y1 - Y2 |
or
| Yavg-1 - Yavg-2 |

ƒ Criteria for deciding the values are different

| Y1 - Y2 | > 2S(2)½
or
| Yavg-1 - Yavg-2 | > 2S[1/ n +1/ m]½

ƒ The confidence level in these decisions is 95%

ƒ Exercise 8: Let S be the average of the standard deviations for the
C3S values for the 4 time periods. Are any of the 4 averages
28
outliers?
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Checking for Outliers in Time Ordered Data

ƒ Statistic for Testing for Extreme Values in Time Ordered Data

- Data in Time Order: y1, y2, y3, y4, y5, …, yn
- Calculate the moving ranges
MRj = | yj - yj-1 | for j =2, …, n
- Calculate Average MR = Σ MRj / (n-1)
- Calculate Critical Value = 3.268 Σ MRj / (n-1)
ƒ Any MR that exceeds the critical value represents a shift in the time
ordered data.
ƒ Once outlier MR’s have been identified, a new average MR can be
calculated and the process repeated. When more than 10% outlier
MR’s are identified, excessive variation is a problem
ƒ One or both of the two values in the moving range may be outliers.
ƒ This method can be used with repeat tests when questionable
results are observed. Here the Range of the repeat tests replaces
the Moving Range.
29
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Checking for Outliers in Time Ordered Data (cont.)

ƒ Exercise 9: Considering the 12 C3S values as time ordered

observations as in the earlier exercise, determine if any of the
moving ranges are outliers. Do outliers MR’s affect the estimate of
the process capability standard deviation? What do outlier MR’s
mean physically?

30
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two standard deviations significantly different?

ƒ Earlier calculations from Exercise 4 showed that the standard

deviations for the C3S values for the 4 time intervals were as
follows:
Interval A B C D
StDev 18.2 1.8 4.2 3.6
ƒ Are these standard deviations equal? In particular are the standard
deviations different for A and B? How about for B and C?
ƒ In Excel® open the spreadsheet with the C3S values for the 4 time
intervals. Then click on “Tools”, “Data Analysis”, “F test: two sample
for variances”.
ƒ Put the cursor in the first window, and highlight the data for interval
A. Put the cursor in the second window, and highlight the data for
interval B. Accept the default risk of deciding the two standard
deviations are not equal when in fact they are equal. This will give a
95% confidence for the decision you are making.
31
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two standard deviations significantly different?

ƒ For a comparison of time intervals A and B, here is the Excel®

input:

ƒ On the next page is the output.

ƒ Exercise 10: Compare the C3S standard deviations for intervals B
and C after discussing the following output on the next page.
32
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two standard deviations significantly different?

F-Test Two-Sample for Variances

Time Interval 1 to 3 4 to 6
Mean 150.6667 168.3333
Variance 320.3333 4.333333
Observations 3 3
df 2 2
F 73.92308
P(F<=f) one-tail 0.013347
F Critical one-tail 19.00003

33
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two averages significantly different?

ƒ Earlier calculations from Exercise 4 showed that the averages for

the C3S values for the 4 time intervals were as follows:

Interval A B C D
Avg 151 168 158 147
ƒ Are these averages equal? In particular are the averages different
for A and B? How about for B and C?
ƒ In Excel® open the spreadsheet with the C3S values for the 4 time
intervals. Then click on “Tools”, “Data Analysis”, “t-Test: Two
sample assuming unequal variances”. This routine is used because
of the results from the F-test.
ƒ Put the cursor in the first window, and highlight the data for interval
A. Put the cursor in the second window, and highlight the data for
interval B. Accept the default risk of deciding the two averages are
not equal when in fact they are equal. This will give a 95%
confidence for the decision you are making.
34
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two averages significantly different?

ƒ In Excel® the input screen is as follows for comparing time intervals

A and B:

ƒ The output from this program is shown on the next page.

ƒ Exercise 11. Are the averages for time intervals B and C unequal?
How confident are you of your decision? 35
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Are two averages significantly different?

ƒ In Excel® the output is as follows for comparing time intervals A

and B:

t-Test: Two-Sample Assuming Unequal Variances

Variable 1 Variable 2
Mean 150.6667 168.3333
Variance 320.3333 4.333333
Observations 3 3
Hypothesized 0
df 2
t Stat -1.69823 Avg are equal
P(T<=t) one-ta 0.115781
t Critical one-ta 2.919987
P(T<=t) two-ta 0.231561
t Critical two-ta 4.302656

36
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Review of Session 1

ƒ There are several ways to summarize the central tendency of data.

Always start by using the AVERAGE function in Excel®. But if
outliers are possible in small data sets, use the MEDIAN function.
ƒ There are several ways to summarize the variation. The range [use
the functions MAX()-MIN()] is useful if the number of observations is
small, say 2 to 5. Otherwise use the STDEV function in Excel®.
ƒ When data are time ordered, use the Spcl=MEDIAN(MR)/.954 This
is often a better estimate of how the process is capable of
performing and is better for determining when the process shifts.
ƒ Outliers can be detected using:
Range Ratio method for small samples sizes, 3 to 5
Least Significant Difference
MR’s if the data are time ordered

37
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 Review of Session 1
ƒ The F-test can be used to determine with specified confidence
whether or not two standard deviations are different. The usual
confidence level is 5%. This means that when one decides two
standard deviations are not equal, one could still be wrong. But this
type of error will only occur 5% of the time.
ƒ There are several t-tests that are used to determine whether or not
two averages are significantly different with specified confidence.
One t-test is appropriate if the standard deviations are not equal
while the other is appropriate if the standard deviations are equal. If
you are in doubt which test to use, use the case for unequal
standard deviations (or variances). If the variances are actually
equal, the test will be essentially equivalent.

38
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002
 End of Session 1

39
Holcim (US) Inc - CTS Geology, Raw Materials and Raw Mix 12.12.2005
PECP02 - 2002