RELIABILITY IN NDT - ROC STUDY OF RADIOGRAPHIC WELD INSPECTIONS

The complexity of modern NDT equipment as well as the demand for high reliability in many industries makes it necessary to have a tool to assess different NDT techniques and devices quickly. As for cost benefit considerations, a measure of accuracy has to include not only the number of correctly detected defects but also the corresponding number of false indications. In order to satisfy a variety of security demands this ratio of true and false indications must be known for a wide range of defect dimensions to be included in the defect count. This information will be provided by only one ROC (relative operating characteristic) study of inspection data using the rating confidence method. An expanded ROC version also yields an assessment of the accuracy of the defect classification and of the correctness of the indicated defect importance. All these measures of performance can also be called up for each individual inspector. For demonstrating the power of the method it is applied to ascertain the difference in detection performance of three distinct experimental modalities in the radiographic testing of welded joints. In order to be capable of drawing conclusions reliably, the study is completed by statistical significance tests. The paper shows the usefulness of ROC for performance assessment in NDT and gives an illustration of what can be done with which effort.