Advancements in autonomous ship trajectory tracking: a comparative study of mechanistic and Neural network models with NMPC

Islam, Tanjil (2024) Advancements in autonomous ship trajectory tracking: a comparative study of mechanistic and Neural network models with NMPC. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Updated Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (4MB)

Abstract

The evolution of autonomous ships marks a significant stride in maritime operations, promising applications across a wide range of industries. These innovations enhance shipping and marine operations by improving safety through the reduction of human error and by enhancing the quality of life for mariners by alleviating tedious or difficult workloads. Whether in commercial shipping, passenger transport, or scientific research, autonomous ships are set to revolutionize the way we navigate the seas, making maritime activities more efficient and safer. Central to advancing this domain is the precise trajectory tracking of Autonomous Surface Vessels (ASVs), which is vital for their safe and efficient navigation. It is simultaneously required to operate within specified timeframes while adhering to maritime regulations and safely maneuvering amidst dynamic marine conditions such as waves, currents, and winds, which pose formidable technical challenges to autonomous trajectory tracking. Presently, both model-based and data-driven controllers are pivotal in navigating Autonomous Vessels, emphasizing the critical need for accuracy and reliability in trajectory following. However, achieving precise trajectory tracking under real-world conditions remains intricate due to varying ship dynamics and environmental disturbances, necessitating tailored controller designs. In this study, our principal aim is to develop a controller that comprehensively addresses these challenges while upholding safety constraints. We leverage Nonlinear Model Predictive Control (NMPC) for its suitability in handling nonlinear ship models, accommodating unmodeled dynamics, managing diverse constraints, and ensuring course stability amidst multivariable systems. An Unscented Kalman Filter (UKF) is integrated with NMPC to mitigate wave-induced disturbances and enhance robustness. Our NMPC controller with UKF, implemented with mechanistic and Neural Network (NN) ship models, is evaluated through trajectory tracking simulations and experimental trials using the Magne Viking ship model at the National Research Council (NRC) in Canada. Incorporating an Artificial Neural Network captures intricate ship dynamics, exhibiting promising results in simulations and practical experiments. We compare the performances of mechanistic and NN models to validate their efficacy, proposing further enhancements through deep neural network training with natural data. Integrating NMPC with neural network structures represents a core aspect of this research, aiming to advance autonomous ship trajectory tracking capabilities in real-world scenarios.

Actions (login required)

View Item