Probabilistic Graphical Model for Sizing Assessment in Ultrasonic Testing in Dissimilar Metal Weld

This work investigates an approach for evaluating the uncertainty associated with flaw sizing in Ultrasonic Testing in nuclear inspections using Bayesian Belief Networks (BBN). BBN graphically represents the probabilistic relationship between the sizing accuracy and some of the variables that cause its deviation. The data for BBN training was collected from sizing information of Round-Robin trials reported in the literature and supplemented with sizing results obtained using the Ultrasonic Testing module of CIVA Non-Destructive Evaluation software. The test specimen used in the Round-Robin exercise and simulations was a dissimilar metal welded pipe. Given the available data, the variables considered causing sizing inaccuracy are the flaw’s actual depth, tilt angle, and position in relation to inner and outer surfaces. The proposed approach quantified the probabilities of flaw depth measurement variations. It gives the results in interactive Directed Acyclic Graphs (DAGs), which enables a user-friendly interface for examining various scenarios of sizing uncertainty.