Optimal Geophone Arrays and 3D Seismic Tomography with Physics-Informed Neural Networks for Subterranean Shipyard Mapping

Inspecting naval facilities (such as piers, wharfs, relieving-platforms, and other shipyard-type structures) for structural and soil voids or support-pile details is a critically important task in supporting the Navy’s broader mission. Effective waterfront facility maintenance management programs play an important role in maintaining fleet readiness, protecting the public and Navy personnel, providing reliable service, protecting the environment, and reducing maintenance and construction costs. Structural conditions above water that could lead to failure, loss of life, or property damage are often observed by engineers, maintenance workers, operations personnel, and sometimes the public. Similar structural conditions underwater cannot be seen and are therefore almost never observed by these same groups until the distress has progressed to the point that damage is evident above water. The primary objective of this project is to develop and deliver a commercially viable seismic imaging system for shipyard non-destructive inspection and testing to the Navy. To do so, we are leveraging state-of-the-art machine learning in combination with physics-based modeling and simulation to develop a high-resolution cost-effective 3D seismic imaging system. Our approach is focused on transforming raw geophone data into a 3D tomographic image of the subsurface below; and is aligned with the Under Secretary of Defense Research and Engineering’s Technology Vision for an Era of Competition[4] – particularly as we are leveraging Advanced Computing and Software to develop this new capability for the Navy.  Through the Phase II research and development effort we will focus on extending both the neural network model and seismic geophone datasets from 2D to 3D, 3D hyperparameter optimization, and synergizing with empirical test design with the objective of producing a field-testable prototype, significantly expanding current capabilities and creating a solution that is readily applicable to several subterranean mapping tasks and stakeholder types. The Phase II work plan includes development, extension, verification, and validation of additional capabilities within the image reconstruction and training dataset, as well as state-of-the-art research in machine learning for seismic tomography for application to real-world subterranean shipyard mapping.