Composite Structure Engineering
Safety Awareness Course
Definition of A- and B-basis Values
and An In-depth Look at CMH-17
Statistical Analysis Techniques
Yeow Ng
1
September 14-16, 2010
A-Basis and B-Basis Definitions
Design values must be chosen to minimize the
probability of structural failure due to material variability.
Compliance is typically shown by selecting design values
that ensure material strength with the following
probability:
Where applied loads are eventually distributed through a single
member within an assembly, the failure of which would result in
loss of structural integrity of the component; 99 percent
probability with 95 percent confidence interval (that is,
A-basis value).
For redundant structure, in which the failure of individual
elements would result in applied loads being safely distributed to
other load carrying members; 90 percent probability with 95
percent confidence interval (that is, B-basis values).
2
September 14-16, 2010
A − Basis value = x − ( K A ) ⋅ s
B − Basis value = x − ( K B ) ⋅ s
3
September 14-16, 2010
The internet browser-based simulation program is available at NCAMP website
http://www.niar.wichita.edu/coe/ncamp_media.asp
Effect of Sample Size on K_factors
6
4
K _ fa c to r
A55
3 A75
B18
B30 B18 Pooling
2
0
0 10 20 30 40 50 60 70 80
Sample Size, n
Effects of CV and Sample Size on Basis Values
Mean=100
100
B-value, CV=2%
A-value, CV=2%
95
B-value, CV=4%
B-value, CV=6%
90
A-value, CV=4%
B-value, CV=8%
85
A-value, CV=6%
80
B a s is V a lu e s
A-value, CV=8%
75
70
65 When sample size is 15 or more, CV has
a greater impact on basis values
60 than the sample size. When CV is
unrealistically low, use Modified CV to
55 avoid overly optimistic basis values.
50
0 10 20 30 40 50 60 70 80
Sample Size, n
Methods to Calculate Basis Values
MIL-HDBK-17F CMH-17 combines the two,
Normal, Weibull, Lognormal distributions giving preference to the AGATE method
Based on Observed Significance Level (OSL) unless certain assumption are not met
Non-Parametric
Batches pooled within environment STATISTICAL FACT: Much larger
sample size needed to
Data NOT pooled across environments
Large sample size required at each environment
estimate std deviation than
to estimate mean
ANOVA (assumes normal distribution)
STAT-17 (a.k.a. Single Point) Excel Spreadsheet Macro (J.Adelmann, Sikorsky Aircraft)
AGATE (Ref. P.Shyprykevich, ASTM STP 1003, 1989.)
Normal Distribution only
Batches pooled within environment
Data pooled across environments
Allowables based on mean of small sample and variability of large pooled sample
Published in DOT/FAA/AR-03/19
ASAP Excel Spreadsheet Macro (K. Suresh Raju, Wichita State University)
RECIPE (Ref. M.G.Vangel, A User’s Guide to Recipe, NIST, 1994)
Normal Distribution only
Regression model
FORTRAN program
September 14-16, 2010 7
Material Allowable Generation with STAT-17
• STAT-17 computes basis
values one environment at CTD RTD
a time (thus a.k.a Single ETW
Point)
• STAT-17 method might NORMAL
Material Property
result in “distributional WEIBULL
ANOVA
changes” which produces
allowables that do not
make engineering sense
• Extremely difficult to
obtain “realistic” load
enhancement factors and Environment
environmental
Mean
compensation factor
B-basis from STAT-17
8
September 14-16, 2010
Material Allowable Generation with ASAP
• For the ASAP procedure,
only normal distribution is CTD RTD
utilized thus eliminating the ETW
possibility of “distributional
changes”** NORMAL NORMAL
Material Property
• Common (pooled) standard NORMAL
deviation results in equal
amount of knock-down for
every environment**
• Basis values and mean
values follow the same
trend Environment
** Must meet requirements of normality Mean
and pooling across environment.
Otherwise, use STAT-17
B-basis from ASAP
9
September 14-16, 2010
Unequal Population Standard Deviations
STAT-17 should be used when population standard
deviations are unequal across environment
September 14-16, 2010
Unequal Population Standard Deviations
ASAP vs. STAT17, Unequal Standard Deviations
For samples that have 90
80
5
4.5
ASAP, Mean of B-basis
Stat17, Mean of B-basis
significantly different 70
4
3.5
ASAP, Std Dev of B-basis
Stat17, Std Dev of B-basis
variances, STAT-17
Value of Standard Deviation
60
Value of Mean
50
should be used. 40
2.5
Equality of variance
30
1.5
20
1
test (Levene’s test) is 10 0.5
a diagnostic test
0 0
CTD RTD ETD ETW
within ASAP.
CTD RTD ETD ETW
Mean 80 80 80 80
Stdev 2 8 8 16
n 30 30 30 30
September 14-16, 2010
Equal Population Standard Deviations
ASAP produces slightly higher
and more stable basis values
September 14-16, 2010
Equal Population Standard Deviations
ASAP vs. STAT17, Equal Standard Deviations
For samples with 90 5
ASAP, Mean of B-basis
4.5
equal variances, ASAP
80 Stat17, Mean of B-basis
4 ASAP, Std Dev of B-basis
70
Stat17, Std Dev of B-basis
will provide basis
3.5
Value of Standard Deviation
60
Value of Mean
50
values that are higher 40
2.5
than STAT-17 and
30
1.5
20
1
those basis values will 10
0
0.5
have less fluctuation CTD RTD ETD ETW
due to random CTD RTD ETD ETW
differences between Mean 80 80 80 80
samples. Stdev 8 8 8 8
n 30 30 30 30
September 14-16, 2010
Conclusions
Standard deviations can be very unstable when sample size is small (<30);
resulting in very erratic basis values, unless pooling method is used
Basis values are not material properties
they are not fixed values because they depend on the number of specimens you
test (i.e. how big your budget is)
If you test more specimens, chances are, you will get a higher basis value
Basis values will vary each time you repeat the program
Don’t be surprised if you get a different number each time (because you should
get a different number each time)
Pooling across environment (ASAP) will typically produce higher and more
stable basis values; but it has more stipulations
Single point (STAT-17) is more flexible because it can handle datasets with
batch-to-batch variability, unequal variances across environment, and
various distributions; but it sometimes may produce basis values that seem
illogical to engineers
Therefore, CMH-17 gives preference to ASAP. STAT-17 is used when ASAP
cannot be used.
September 14-16, 2010
