Assessing the sensitivity of ultrasonic testing using ROC curves

Abstract

One way to estimate the reliability of a technique is through a sensitivity analysis. When working with a high sensitivity, the lower amplitude signals are picked up, increasing the risk of noisy signals being interpreted as signals from a flaw. On the other hand, when the inspector is working with low sensitivity, the system is only capable of distinguishing signals of high amplitude. ROC (receiver operating characteristic) curve is a technique for evaluating, organizing and selecting classifiers based on their performance. The ROC curves may show four signs of return, among them, rates of hits and false alarms of the classifiers, which serve to evaluate the sensitivity of the technique. They have a system ranging from 0 to 1 where they describe the relationship between probability of detection and the probability of false alarm. The paper investigates the sensitivity of the manual ultrasonic technique in API X70 pipes containing defects of different types and at different ranges of length and height. The results showed that the ROC curve is an excellent tool to assess the sensitivity of ultrasonic testing, being capable to demonstrate the significance of the flaw height on the probabilities of detection and false alarm.