2024/03/31 by Zarghami, Mohammad Ali, Raissi, Sadigh, Tohidi, Hamid +1
Engineering · Environmental Science · #Machine Fault Diagnosis Techniques #Reliability and Maintenance Optimization #Maritime Transport Emissions and Efficiency
paper · doi:10.57647/j.ijeee.2024.1501.04
Electric motors in port ship unloaders operate under harsh environmental conditions that accelerate degradation and increase energy consumption. This study develops a unified hybrid predictive maintenance framework that integrates an adaptive Extended Kalman Filter (EKF), a Long Short-Term Memory (LSTM) network, and Deep Reinforcement Learning (DRL). The EKF—enhanced with the adaptive Sage–Husa algorithm—serves as a dynamic state estimation and noise-mitigation module, improving the quality of multi-sensor data before learning and decision-making. The LSTM component captures nonlinear temporal degradation patterns, while the DRL agent learns optimal maintenance policies that balance reliability, operational safety, and energy efficiency. The proposed EKF–LSTM–DRL framework was validated on real and simulated datasets from Neuero ship unloader motors at Bandar Imam Khomeini Port (Iran). Results show a 20% reduction in Mean Absolute Error (MAE) and a 15% decrease in unplanned downtime compared with baseline models such as GRU and CNN. The main contributions of this work are: (1) a clear integration of adaptive EKF-based preprocessing with deep learning and reinforcement learning for robust Remaining Useful Life (RUL) prediction under maritime uncertainty; and (2) a dual optimization of predictive accuracy and energy efficiency. A detailed workflow diagram has been added to clarify the complete process and ensure conceptual transparency.