Multi-Sensor Diagnostic Framework for Bearing Fault Detection in Induction Motors

Abstract

Bearing faults are one of the most failure mechanisms in induction motors, leading to increased vibration, reduced efficiency, and unexpected downtime. Existing diagnostic approaches are often limited by reliance on single-sensor measurements or computationally intensive learning-based models, which reduces robustness under variable load conditions and limits real-time industrial applicability. This paper presents a multi-sensor diagnostic framework that integrates current, vibration, torque, and rotational speed measurements under varying load conditions. The proposed method processes synchronized data acquired from a 7.5 kW induction motor operating across 0%-100% load conditions. The signals are segmented using overlapping windows, while Fast Fourier Transform (FFT), envelope spectrum analysis, and Short-Time Fourier Transform (STFT) are applied to extract characteristic fault frequencies (BPFO, BPFI, BSF, FTF) and associated spectral components. The extracted features are normalized and fused into a unified feature vector, from which a lightweight and interpretable Fault Index (FI) is constructed through adaptive weighting based on statistical sensitivity and domain knowledge. Unlike computationally intensive machine learning approaches, the proposed FI enables real-time condition monitoring with low computational cost while maintaining high diagnostic interpretability. Experimental results demonstrate that the proposed fusion strategy provides clear discrimination between healthy and faulty conditions across multiple bearing defect types and load levels. The optimized FI improves class separability compared to the non-optimized baseline, while maintaining stable performance under all load conditions. The proposed approach offers a scalable and computationally efficient solution for real-time bearing fault diagnosis in industrial environments.