Development of a Physics-Informed Artificial Intelligence Framework for Real-Time Thermodynamic Performance Prediction and Energy Optimization of Marine Diesel Engines

Abstract

Marine diesel engines operate under highly dynamic thermodynamic conditions, requiring intelligent monitoring and energy optimization systems capable of providing accurate real-time decision support. Conventional Artificial Intelligence (AI) models have demonstrated strong predictive performance; however, they often neglect fundamental thermodynamic principles, resulting in limited physical consistency and reduced reliability under varying operating conditions. This study proposes a Physics-Informed Artificial Intelligence (PI-AI) framework that integrates the first and second laws of thermodynamics into a machine learning architecture to improve real-time thermodynamic performance prediction and optimize energy efficiency in marine diesel engines. The proposed framework utilizes Internet of Things (IoT)-based sensor data, including cylinder pressure, engine speed, fuel flow rate, intake air pressure, intake air temperature, exhaust gas temperature, cooling water temperature, and lubricating oil temperature. These data are processed using a Physics-Informed Neural Network (PINN) integrated with a Digital Twin model to represent the engine's dynamic behavior, while a Deep Reinforcement Learning (DRL) algorithm continuously determines optimal operating strategies for maximizing thermal efficiency and minimizing fuel consumption within safe operational constraints. Model performance is evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The effectiveness of the optimization framework is assessed through improvements in thermal efficiency, reductions in Brake Specific Fuel Consumption (BSFC), and decreases in entropy generation. The proposed framework is expected to provide more accurate, robust, and physically consistent thermodynamic predictions than conventional data-driven AI models while enabling adaptive real-time energy optimization. The integration of Physics-Informed AI, Digital Twin technology, and Deep Reinforcement Learning constitutes the primary novelty of this study, offering a comprehensive intelligent framework for predictive monitoring, energy-efficient engine operation, and emission reduction in next-generation smart maritime transportation systems.