Physical Model Guided Data Driven Approach for Degradation Prognostics of Rotating Electrical Machine installed in Shore Marine Environment

Abstract Reliable degradation prognostics of electric motor operating in shore marine environments is essential for early fault detection and effective maintenance planning. This paper presents a physical model-guided, data-driven approach for degradation prognostics of an electric motor using thermal Non-Destructive Testing (NDT) data. The proposed hybrid framework integrates a particle filter algorithm with a life acceleration model to capture the nonlinear degradation dynamics influenced by thermal stress. Thermal imaging data acquired periodically from the winding surface are processed to extract degradation-sensitive features, which serve as measurement inputs to the particle filter for real-time state estimation. The life-acceleration model provides the physical degradation trend, enabling physics-informed updates within the state-transition model of the particle-filter-based prediction framework. Experimental analysis on a shore-based motor-generator set operating under marine conditions demonstrates that the proposed approach effectively tracks the degradation evolution and provides accurate prognostic trends. The results confirm that incorporating physical degradation knowledge significantly improves the robustness and interpretability of data-driven prognostic models for rotating electrical machinery.