AN APPROACH TO THE QUESTION ‘HOW TO ACCOUNT FOR HUMAN ERROR IN MAPOD?’

Speaker:
Spies, Martin; Fraunhofer IZFP; Germany

Authors:
Spies, M.; Fraunhofer Institute for Nondestructive Testing IZFP; Germany
Rieder, H.; Fraunhofer Institute for Nondestructive Testing IZFP; Germany

ID: ECNDT-0571-2018
Download: PDF
Session: NDT Reliability 3
Room: J1
Date: 2018-06-15
Time: 11:30 - 11:50

Abstract: Progress in defect detection by using optimized or newly developed inspection techniques leads to improved reliability in NDT. In ultrasonic testing (UT) a reduction of human factors’ influence can be achieved e.g. by replacing manual UT by mechanized scanning – if thoroughly applied. POD and PFI curves are popular and suitable tools to demonstrate and quantify the achieved improvements. To reduce the experimental efforts necessary for decent POD studies, Model-Assisted POD (MAPOD) has evolved in recent years with efficient simulation techniques at hand to properly account for most of the relevant influencing inspection parameters such as material, component geometry and sensor properties. While in many industries manual UT is still high-valued, the human factor is one of the key issues for NDT reliability. Thus, the question has been raised how MAPOD can take these factors into account.
In this contribution – as a first approach – we address the question how human error can be accounted for. Two main errors with considerable impact on the result of manual UT inspections are (1) missing out on a defect or not declaring an indication as a defect, and (2) assigning an erroneous amplitude to a defect by e.g. misreadings or a false calibration procedure. A straightforward approach to account for this in an â-vs-a analysis is to correspondingly ‘manipulate’ certain data points by setting the amplitude to noise level (Case 1) or assigning a different amplitude (Case 2). In our approach we do this in a statistically random manner. In order to assign different amplitudes to a realistic amount of data points we refer to corresponding studies such as the PISC programme. Here the efficiency of the inspectors in the detection of fatigue cracks and lack of fusion defects was between 65% and 100%. For several inspection situations we compare the resulting POD-curves with the originally determined MAPOD-curves and discuss the implications of the obtained results and the potential benefits of this approach.