Model-assisted Probability of Detection Analysis of Eddy Current Non-destructive Testing Based on Neural Networks

Model-assisted probability of detection (MAPoD) is a significant method in assessing the detection power of non-destructive evaluation systems. However, for the study of probability of detection, to propagate the uncertainty of random inputs into the model response, a large amount of model evaluation is required, which leads to extremely high computational costs. This makes it impossible to complete the probability of detection study of non-destructive testing systems within the specified time. The purpose of this study is to expedite the solution process of MAPoD in eddy current non-destructive testing system by replacing precise physical-based models partially with fully connected neural network surrogate models.