•  
  •  
 

Abstract

This paper presents a computationally efficient statistical anomaly detection framework for Permanent Magnet Synchronous Motor (PMSM) fault diagnosis and severity estimation, specifically targeting naval windlass drive systems. The proposed method employs higher-order statistical moments, namely kurtosis and skewness, extracted from phase current signals, combined with a novel composite Fault Severity Index (FSI) to quantify the deviation from healthy operating conditions. A detailed MATLAB/Simulink model incorporating field-oriented control and a modular fault injection subsystem was developed to simulate five fault categories at three severity levels: winding overheating (10%, 25%, 50% resistance increase), phase imbalance (scaling factors of 0.9, 0.8, 0.7), voltage drop (0.9, 0.8, 0.7 multipliers), bearing faults (0.5–1.5 Nm impulsive disturbances at 100–300 Hz), and severe vibrations (broadband multi-frequency torque components). Classification thresholds were statistically determined from 120 healthy baseline windows across multiple operational modes, ensuring data-driven rather than arbitrary threshold selection. Diagnostic validation was performed through quantitative metrics including confusion matrix analysis, yielding an overall detection accuracy of 94.4%, sensitivity of 96.2%, specificity of 92.8%, precision of 93.5%, and F1-score of 0.949. Comparative benchmarking against Root Mean Square (RMS) monitoring, Fast Fourier Transform-based Motor Current Signature Analysis (FFT-MCSA), Total Harmonic Distortion (THD), and a Support Vector Machine (SVM) classifier demonstrated that the proposed framework achieves comparable or superior detection performance at a fraction of the computational cost. The method exhibits rapid detection latency (0.05–0.15 s in simulation), robustness to parameter variations, and interpretability advantages over machine learning alternatives, making it suitable for real-time embedded monitoring in safety-critical maritime applications.

Included in

Engineering Commons

Share

COinS