Fault diagnosis of marine propulsion systems based on FEEMD-MPE

Current approaches to fault diagnosis in marine propulsion systems exhibit notable limitations when processing non-stationary signals commonly encountered in marine environments. Traditional techniques such as the Fourier transform are inadequate for effectively capturing the changing characteristic non-steady signals over time. In addition, widely used fault characteristic extraction approaches, including statistical parameter analysis and envelope spectrum analysis, have restricted capability in representing complex nonlinear fault signals, thereby hindering the accurate detection of incipient faults. To address these limitations and the inherent challenges associated with the non-stationary and nonlinear nature of fault signals in marine propulsion systems, this study proposes a novel approache for diagnosing faults that integrates Fast Ensemble Empirical Mode Decomposition (FEEMD) with Multiscale Permutation Entropy (MPE). In the first stage, fault signals are decomposed using FEEMD. Entropy values are then computed using MPE to construct the feature dataset. The Relief algorithm is subsequently applied to reduce the feature dimension from 30 to 15, providing the optimal input for the model. Based on this refined input, an Extreme Learning Machine (ELM) model optimized with Particle Swarm Optimization (PSO) model is developed for fault diagnosis. Experimental findings show that the presented approach attains a fault diagnosis accuracy of 99.17%., representing a 15.84% improvement over the baseline ELM model. Furthermore, the method delivers superior performance in various comparative evaluations, establishing its reliability for fault detection in marine propulsion systems.