Using surrogate models and mechanics to predict pressure from strain data

Abstract

Many marine structures rely upon hull monitoring systems that utilize strain information to make inferences about structural integrity. One of the most significant factors affecting structural integrity are wave loads, specifically the resultant pressures. While direct measurement is not often available, this study explores how to translate strain information into localized pressure estimates through two different low-order modeling paradigms, physics-based and surrogate modeling. Strain data was collected via strain gauges mounted on a scaled ship model that was tested in towing tank under a variety of regular and irregular wave conditions. Using a set of strain calibration data, two different models are developed, one using principal stress and strain relations and the other using a machine learning approach (Gaussian process model), for predicting pressure time histories for different wave conditions. The two modeling approaches are compared in terms of both a variety of error metrics as well as their ability to handle measurement noise and disturbances commonly encountered during marine structures testing and operations.