Generative Adversarial Network for Prediction of Visual Deteriorations and NDE Maps for Bridge Decks

Non-destructive evaluation (NDE) techniques are excellent at identifying subsurface deteriorations (e.g., cracks, delamination, and corrosion). NDE surveys of bridge decks are conducted periodically to identify causes and quantify the progress of deterioration. Many transportation authorities however rely on traditional visual assessment techniques for the generation of bridge deck condition indices. The primary limitation of visual assessment is that subsurface conditions and other hidden anomalies are not visible. Many transportation authorities do deploy NDE for condition assessment, however, few techniques are usually used. In this work, a generative deep learning approach is introduced for the two applications: (i) predicting future visual deterioration from past NDE data, and (ii) predicting current conditions map for an unknown NDE technique using a current condition map from a known NDE technique. This approach may be attractive to transportation authorities that may wish to use NDE condition maps to infer future visual deteriorations and other NDE condition maps.