Data Reduction
13
may exist in the lower tail. Then [1.8] suggests plotting the raw data for
percent failures on Weibull or log normal paper. This is an integration
off(x) or failure data and tends to smooth out any variations so that
better estimate of A, or 0, B, ~ and 2, s can be obtained. Then Eqs. (1.37)
and (1.38) can be used for the Weibull fit of test data and Eqs. (1.21)
and (1.22) for the Gaussian fit of the data.
VI. PRIORITY ON PROCESSING RAW DATA
Priority is decided when looking at Sturges Rule Section 1.1 when for
Example 1.1
K= l+3.31og10N
K= 1+3.31og~02
K = 1.993(2.00)
Range (R) =X2 -- Xl = 500 psi
R 5000 psi
Cell width (co) - K - ~ -- 2500
Each cell has 1 failure and examination of probability paper for the Weibull
and Gaussian distribution for percent failures may have one data point at
50% but 100% or 0% does not show on a log scale. As a result there
are two plots with one data point at 50% for each and any line passes
through one data point.
The estimates in Eq. (1.33)-(1.36) must be used. They are as accurate
as can be obtained. The Gaussian curve data is the approximated distribution. Up to 20 test points is allowable. A computer analysis is out
of the question for N = 2.
When 2 < N < 20 the individual data points may be used to obtain a
line on both Weibull and Gaussian distribution plots. Note in Table 1.3
individual points allow around 10 data points for N= 10 while the partitioning into cells allows only 4 data points rounded off. In order to obtain
plotted estimates individual data may be used and fitted to partitioned cells
for greater accuracy.
EXAMPLE 1.2. The rainy season annual rainfall data is published
for the civic center in Los Angeles in July of each year. The data for
1877-1997 (published 4 July 1997 in the L.A. Times) is listed in Appendix
F. This data, 120 values, was entered into a SAS program to generate
Gaussian and the Weibull three parameter distribution.
The data is
arranged and partitioned according to Sturges Rule (Section I) then Fig.
13
may exist in the lower tail. Then [1.8] suggests plotting the raw data for
percent failures on Weibull or log normal paper. This is an integration
off(x) or failure data and tends to smooth out any variations so that
better estimate of A, or 0, B, ~ and 2, s can be obtained. Then Eqs. (1.37)
and (1.38) can be used for the Weibull fit of test data and Eqs. (1.21)
and (1.22) for the Gaussian fit of the data.
VI. PRIORITY ON PROCESSING RAW DATA
Priority is decided when looking at Sturges Rule Section 1.1 when for
Example 1.1
K= l+3.31og10N
K= 1+3.31og~02
K = 1.993(2.00)
Range (R) =X2 -- Xl = 500 psi
R 5000 psi
Cell width (co) - K - ~ -- 2500
Each cell has 1 failure and examination of probability paper for the Weibull
and Gaussian distribution for percent failures may have one data point at
50% but 100% or 0% does not show on a log scale. As a result there
are two plots with one data point at 50% for each and any line passes
through one data point.
The estimates in Eq. (1.33)-(1.36) must be used. They are as accurate
as can be obtained. The Gaussian curve data is the approximated distribution. Up to 20 test points is allowable. A computer analysis is out
of the question for N = 2.
When 2 < N < 20 the individual data points may be used to obtain a
line on both Weibull and Gaussian distribution plots. Note in Table 1.3
individual points allow around 10 data points for N= 10 while the partitioning into cells allows only 4 data points rounded off. In order to obtain
plotted estimates individual data may be used and fitted to partitioned cells
for greater accuracy.
EXAMPLE 1.2. The rainy season annual rainfall data is published
for the civic center in Los Angeles in July of each year. The data for
1877-1997 (published 4 July 1997 in the L.A. Times) is listed in Appendix
F. This data, 120 values, was entered into a SAS program to generate
Gaussian and the Weibull three parameter distribution.
The data is
arranged and partitioned according to Sturges Rule (Section I) then Fig.