A Human Factors Perspective on the Use of Automated Aids in the Evaluation of NDT Data

In comparison to manual NDT methods, mechanized NDT is considered to be more reliable for a number of reasons, one of which being that the role of the inspectors and, therewith, the potential for human error, have been reduced. However, human-automation interaction research suggests that in spite of its numerous benefits, automation can lead to new yet unknown risks. One of those risks is inappropriate reliance on automation, which can result in automation misuse and disuse. The aim of this study was to investigate the potential inappropriate use of automation (specifically the automated aids) in NDT addressing therewith the prevalent belief in the high reliability of automation held by the NDT community. To address this issue, 70 NDT trainees were asked to control the results of an eddy current data evaluation, allegedly provided by an automated aid, i.e. indication detection and sizing software. Seven errors were implemented into the task and it was measured to what extent the participants agreed with the aid. The results revealed signs of both misuse (agreeing with the aid even though it is incorrect) and disuse (disagreeing with the aid even though it is correct) of the aid that can affect the reliability with which inspections are carried out. Whereas misuse could be explained by a lower propensity to take risks and by a decreased verification behavior-possibly due to bias towards automation and complacent behavior-, disuse was assigned to problems in establishing the sizing criterion or to general difficulties in sizing. The implications of these results for the NDT praxis including suggestions for the decrease of automation bias are discussed.