Real-time super-resolution mapping of locally anisotropic grain orientations for ultrasonic non-destructive evaluation of crystalline material

• DOI: 10.1007/s00521-021-06670-8
• Publication Date: 2021-11-20T18:02:36Z
• Authors: Jonathan Singh; Katherine Tant; Andrew Curtis; Anthony Mulholland
• Keywords: Engineering design; Engineering Mathematics Research Group; /dk/atira/pure/core/keywords/engineering_mathematics_research_group; name=Engineering Mathematics Research Group; ultrasound; neural network; Image and Video Processing (eess.IV); FOS: Physical sciences; 006; Electrical Engineering and Systems Science - Image and Video Processing; /dk/atira/pure/core/keywords/engineering_mathematics_research_group; Geophysics (physics.geo-ph); Physics - Geophysics; machine learning; QA273; TA174; Physics - Data Analysis, Statistics and Probability; FOS: Electrical engineering, electronic engineering, information engineering; Probabilities. Mathematical statistics; Tomography; Data Analysis, Statistics and Probability (physics.data-an)

Abstract Estimating the spatially varying microstructures of heterogeneous and locally anisotropic media non-destructively is necessary for accurate detection flaws reliable monitoring manufacturing processes. Conventional algorithms used solving this inverse problem come with significant computational cost, particularly in case high-dimensional, nonlinear tomographic problems, are thus not suitable near-real-time applications. In paper, first time, we propose a framework which uses deep neural networks (DNNs) full aperture, pitch-catch pulse-echo transducer configurations, to reconstruct material maps crystallographic orientation. We also present application generative adversarial (GANs) achieve super-resolution ultrasonic images, providing factor-four increase image resolution up 50% structural similarity. The importance including appropriate prior knowledge GAN training data set inversion accuracy demonstrated: known information about material’s structure should be represented data. show that after computationally expensive process, DNNs GANs can less than 1 second (0.9 s on standard desktop computer) provide high-resolution map grain orientations, addressing challenge cost faced by conventional tomography algorithms.