Ensemble-Based Health Index Construction: A Low-Frequency- Vibration-Data-Driven Approach for Electric Motor Remaining Useful Life Prediction

Accurate prediction of remaining useful life (RUL) for vibration devices is crucial for optimizing maintenance schedules and reducing operational costs in industrial systems. However, current feature extraction methods for vibration devices in RUL prediction often rely on high-frequency vibration signals, which are impractical in real-world industrial settings with cost constraints. Furthermore, conventional HI modeling struggles with incomplete lifecycle data and significant inter-device degradation variations. To overcome these limitations, this paper proposes an ensemble-based health index (HI) approach for RUL prediction of electric motors, addressing challenges posed by low-frequency sensor data, variations between devices, and incomplete lifecycle data. The proposed method introduces: (1) a feature extraction framework integrating time-domain statistics and kernel density estimation (KDE)-based probability density function (PDF) analysis for low-frequency scenarios, and (2) an ensemble-based HI construction method using adaptive device clustering, subgroup-specific multilayer perceptron (MLP) modeling, and similarity-weighted fusion. Vibration data from 15 conveyor motors in an electric meter production line validates the effectiveness of the proposed method. The proposed method enables reliable RUL estimation under industrial constraints, offering a practical solution for predictive maintenance in low-frequency, high-variability environments.