Beyond the Single Predictor Variable POD Model

A single predictor variable model is typically used to introduce the underlying regression methods behind the hit/miss and รข versus a Probability of Detection (POD) models. The typical example assumes that POD is related to a single characteristic of the indication, such as length. Often, though, it is more than one characteristic of an indication that affects POD. For example, POD for a Fluorescent Penetrant system is affected by indication length as well as depth and width. Hence, if possible, a model with multiple predictor variables is more appropriate and will result in a more accurate estimate of the capability of a nondestructive testing (NDT) system. The hit/miss POD model will be used to show that the single variable model can easily be extended to more than one predictor variable. In addition, a hit/miss POD analysis will be performed on simulated data, the results of which will be used to illustrate how the analysis yields an NDT system capability range rather than a single capability value when more than one predictor variable is included in the POD model.

1. Lloyd, C.J., Statistical Analysis of Categorical Data, John Wiley & Sons, Inc., 1999. 2. MIL-HDBK-1823A, Non-Destructive Evaluation System Reliability Assessment, 2009. 3. Neter, Kutner, Nachtsheim, and Wasserman, Applied Linear Statistical Models, fourth edition, The McGraw-Hill Companies, Inc., 1996
Usage Shares
Total Views
10 Page Views
Total Shares
0 Tweets
0 PDF Downloads
0 Facebook Shares
Total Usage