Deep Learning Based Inversion of Locally Anisotropic Weld Properties from Ultrasonic Array Data

The ability to reliably detect and characterise defects embedded in austenitic steel welds depends on prior knowledge of microstructural descriptors, such as the orientations weld’s locally anisotropic grain structure. These are usually unknown but it has been shown recently that they can be estimated from ultrasonic scattered wave data. However, conventional algorithms used for solving this inverse problem incur a significant computational cost. In paper, we propose framework which uses deep neural networks (DNNs) reconstruct crystallographic welded material travel time data, real-time. Acquiring large amount training data required DNNs experimentally is practically infeasible problem, therefore model based approach investigated instead, where simple efficient analytical method modelling times through given weld geometries implemented. proposed validated by testing trained arising sophisticated finite element simulations propagation microstructures. network predicts within 3° near real-time (0.04 s), presenting step towards realising real-time, accurate characterisation microstructures non-destructive measurements. subsequent improvement defect imaging then demonstrated via use DNN predicted correct delay laws total focusing algorithm based. An up 5.3 dB signal-to-noise ratio achieved.