Real-Time Ocean Environment and Aquaculture Structure Load/Motion Predictions through Physics-Informed Machine Learning

Objectives: To develop, test and validate the nonlinear dynamic response modeling and predictive capability for aquaculture facilities in rough sea conditions. The specific objective is to develop a prediction technology to act as a “real-time digital twin”, called AQUA-DT, which will enhance and support the design, installation and operation of intelligent open-ocean aquaculture systems. Based on real-time (sparse) upstream wave measurement (e.g. by a wave-directional buoy), AQUA-DT will provide faster-than-real-time reconstruction and forecast of (1) phase-resolved nonlinear wave (and current) environments, and (2) phase-resolved hydrodynamic loads and motions of aquaculture fish cages and vessels in an aquaculture farming site.

Methodology: The basis of AQUA-DT is physics-informed machine learning developed from state-of-the-art nonlinear wave and hydrodynamic load prediction capabilities. We are uniquely situated to develop this technology using our newly developed AQUA-MIT capability, which predicts the nonlinear wavefields and loads/motions of aquaculture cages in rough seas, and our expertise in using field measurements to perform phase-resolved nonlinear wave/current predictions. AQUA-DT will be trained offline using limited available wave measurements and extensive physics-based simulation data. The validity of AQUA-DT will be assessed in a laboratory environment in which the response of a fishing cage under long-crested irregular waves will be tested.

Rationale: Advancing the nonlinear dynamic response modeling and predictive capability for aquaculture facilities in rough sea is critical to support the design, installation and operation of offshore structures and to define new standards for open-sea aquaculture activities. Most existing work is for near shore or protected water and has limited applicability in more energetic environments. A developed digital twin, AQUA-DT, allows real-time monitoring of physical assets and can act as early warning and prediction tools. It can support early detection of the need for intervention or notice of structure compromise, and provide faster-than-real-time path prediction capabilities for autonomous vehicle operations.