A “Model Assisted Probability of Detection” approach for ultrasonic inspection of railway axles

The reliability of non-destructive testing is usually quantified in terms of the “Probability of Detection” curves relating a characteristic linear dimension of the considered defects to the probability to observe them. Actually, POD curves are also function of other implicit factors, so they can be considered stochastic in nature and, consequently, the most rational and reliable approach to their derivation should also consider the determination of a suitable confidence level. The problem arises because such confidence level is usually requested to be equal to 95% for POD curves, so involving high costs and long times for the needed experimental tests. The present paper is focused on the reliability of ultrasonic inspection applied to hollow railway axles. In particular, a novel method, recently proposed by the authors, for the interpretation of POD data is firstly described and experimentally validated. Eventually, with the aim to lower the overall experimental costs, the possibility to substitute part of the needed experimental tests with proper numerical simulations, so applying the so-called “Model Assisted Probability of Detection” approach, is successfully investigated