Wave-Informed Machine Learning for Material Thickness Imaging with Ultrasonic Waves

We present developments on our algorithm for wave-informed learning, currently applied to defect detection. Reliable, efficient, and interpretable defect detection is crucial to ensure material quality in a variety of applications. Last year, we introduced a wave-informed regression framework which integrates wave theory and a linear learning architecture to separate spatially and temporally heterogeneous modes from ultrasonic waves induced into the specimen of interest. We can then identify which modes differ from the baseline wavefield. These variations highlight and characterize heterogeneities in the material, such as thickness variations. In this presentation, we revisit the mathematical and algorithmic framework with additional details and further interpretation of the underlying physics. We show novel results which demonstrate that our framework can independently reconstruct thickness variations in an aluminum plate with high accuracy, thus imaging, detecting, and characterizing defects.