Home Contact Sitemap

AFGROW | DTD Handbook

Handbook for Damage Tolerant Design

  • DTDHandbook
    • About
    • Contact
    • Contributors
    • PDF Versions
    • Related Links
    • Sections
    • Examples

Section 5.1.4. Use of Data and Data Scatter

As indicated by the results presented in the previous section, accurate mean trend FCGR descriptions result in accurate fatigue crack life descriptions.  People have worried in the past about trying to account for the substantial amount of scatter that exists in the crack growth rate data.  The amount of crack growth between crack measurements and the accuracy of this incremental crack growth measurement determines a large part of the scatter.  Another inherent reason for data scatter is due to the differentiation techniques that one uses to reduce the data.

Figure 5.1.12a shows a hypothetical example of the crack growth-life behavior observed in a single laboratory test; Figure 5.1.12b represents the FCGR data derived from this test.  An asterisk in Figures 5.1.12a and b indicates outlying data points.  The mean trend curves faired through the data are directly related to each other; the integral of the curve in Figure 5.1.12b gives the curve in Figure 5.1.12a for the test conditions.  If more tests are run and all the data compiled, the plot is as shown in Figure 5.1.12c; each test might have a few outlying data points, but the compilation has many outlying points.  When all data points, including the outliers, are plotted, the data exhibit a wide scatter-band, noted as the apparent scatter-band, shown in Figure 5.1.12c.  However, as previously seen from Figures 5.1.12a and b, the outlier points did not significantly affect the crack growth curve or the mean trend FCGR curve.  When considered collectively, the outlying data points in Figure 5.1.12c can be misleading since they do not represent the mean trend behavior of any specimen.  If the wide scatter-band were considered for a crack growth prediction, the upper bound would predict a consistent high growth rate for each crack size (whereas it happened only incidentally as shown in Figure 5.1.12a).  As a result, the diagram would reflect a large apparent scatter in crack growth lives (Figure 5.1.12d), whereas the real scatter in crack growth lives is much smaller.

Figure 5.1.12.  Crack Growth Data Scatter for Identical Conditions

As indicated by the above remarks, worrying about the random (within specimen) scatter in fatigue crack growth rates is really not that important from a life estimation standpoint.  What has been found from analyses of multiple specimen data sets is that the width of the scatter-bands associated with specimen to specimen mean trend variations in FCGR is closely related to the variability in crack growth-life behavior.  The scatter-band associated with specimen to specimen variations is identified in Figures 5.1.12c and d as the real scatter-band since it focuses on the variability in crack growth-life behavior.

The coefficient in variation of crack growth lives is sometimes similar in magnitude to the root mean square (percentage) error associated with fatigue crack growth rate modeling.  When conservative estimates in crack growth lives are desired, the upper bound of the real scatter-band (identified in Figure 5.1.12c) determined on the basis of four or more specimens should be used.