Statistical methods of an assessment of coherence of experts opinions in interpretation of radiogram

This paper presents statistical analysis of interpretation of 10 welded joints radiographs performed independently by 9 experts - qualified level 2 or 3 certification specialists in RT with length of service from 3 up to 30 years. It was offered to use the Contingent coefficient calculated for each type of flaw (porosity and gas pores including elongated cavities, slag inclusion including elongated slag inclusion, incomplete penetration, lack of fusion, longitudinal crack, root concavity, shrinkage grooves) and for each pair of experts to assess the impact of subjective component (human judgment) on the interpretation results. Statistical experiment was carried out in accordance to using ad hoc technique. It showed only weak agreement between radiographers regarding the determination of the discontinuity type. Slag intrusions have strong agreement (coefficient of four pointed correlation more than 0,3) was observed only in 27,3% cases cold lapping in 8,3% cases during the identification. In general the strong expert's agreement in determination of the flaw type by its projection on the radiogram revealed in 35,5% cases. This result causes to doubt on the appropriateness of using imperfection type as the basic quality assessment criteria in radiographic inspection. The separation of different types of flaws into classes was proposed, where each class includes flaws with similar projections on the radiogram.