Semester of Graduation

Spring 2026

Degree

Master of Science in Mechanical Engineering (MSME)

Department

Mechanical Engineering

Document Type

Thesis

Abstract

Accurate modeling of marine vehicle dynamics remains challenging due to strong nonlinear hydrodynamic effects, environmental disturbances, and sensitivity to configuration changes, particularly for small-scale platforms. Classical physics-based models require extensive parameter identification and often exhibit degraded performance outside narrow operating regimes, while purely data-driven approaches may lack structure or impose high computational cost. This thesis presents a data-driven Koopman operator framework with hybrid observables for modeling the dynamics of unmanned marine vehicles. The proposed approach combines structured monomial observables with a learned neural network embedding to construct a lifted state representation in which the nonlinear vehicle dynamics are approximated by a linear system with control inputs. An alternating training procedure is employed to jointly learn the lifting and reconstruction mappings, as well as to estimate the Koopman operator using regularized least-squares regression. To enable adaptation to changing environmental conditions, an online update mechanism is introduced that incrementally refines the Koopman operator using streaming data without retraining the neural networks. The proposed method is evaluated for two physical platforms: a small-scale unmanned underwater vehicle and a small-scale unmanned surface vehicle, using experimental data collected in real operating environments. Predictive performance is assessed under open-loop rollouts for short-term (10 s) and long-term (60 s) horizons and compared against an established physics-based model. The hybrid Koopman models demonstrate improved predictive accuracy across most degrees of freedom, achieving order-of-magnitude reductions in normalized error in several states while maintaining stable long-horizon behavior. Short online adaptation further enhances short-term prediction accuracy, particularly for environmentally sensitive dynamics. These results demonstrate that hybrid Koopman models provide an effective balance between interpretability, computational efficiency, and predictive fidelity, supporting their suitability for onboard prediction, estimation, and model-based control of marine vehicles.

Date

3-24-2026

Committee Chair

Corina Barbalata

LSU Acknowledgement

1

LSU Accessibility Acknowledgment

1

Download

Share

COinS