Development of a hybrid (physics-informed) model of ship motion dynamics

Improving the operational efficiency of marine vessels and the reliability of ship power plants requires the application of advanced methods for intelligent data analysis and modeling of complex dynamic processes. One of the most promising approaches is scientific machine learning, particularly physics-informed neural networks (PINNs), which combine physical laws governing technical systems with the capabilities of deep learning. The aim of this study is to investigate the potential of physics-informed neural networks for modeling hydrodynamic processes and for intelligent diagnostics of the technical condition of ship power plants. The paper analyzes modern approaches to computational hydrodynamics and machine learning methods used to describe ship motion dynamics and operational processes of marine energy systems. Special attention is given to the principles of constructing PINN models in which differential equations describing the physics of the investigated processes are incorporated directly into the neural network loss function. This approach improves prediction accuracy and model robustness when only limited experimental or operational data are available. It is shown that the use of physics-informed neural networks allows more accurate reproduction of nonlinear dynamic relationships between ship motion parameters, hydrodynamic characteristics, and the performance indicators of propulsion and power systems. Based on the analysis of recent scientific studies and published results, the advantages of the PINN approach compared with traditional computational fluid dynamics methods and purely data-driven machine learning algorithms are identified. It is demonstrated that the integration of physical models with neural network algorithms improves the reliability of predicting the technical condition of ship equipment and provides a foundation for developing intelligent monitoring and diagnostic systems for marine power plants. The obtained results confirm the перспективность of applying physics-informed neural networks to problems of ship hydrodynamics analysis, prediction of operational parameters, and improvement of technical diagnostic systems in marine engineering.