Beyond Signal Processing: A Model-Based Luenberger Observer Approach for Accurate Bearing Fault Diagnosis

Traditionally, diagnosis of bearing faults involves analyzing the frequency spectra of monitored signals, like vibration and stator current, using various signal processing techniques. However, signal-based methods for fault diagnosis often produce false alarms due to changes in load and voltage imbalances in the motor's input. Furthermore, these methods have limited performance in detecting faults at early stages and readjusting based on speed, load, and voltage levels. To overcome these challenges, this paper proposes a model-based approach for bearing fault diagnosis utilizing the Luenberger observer. The suggested model-based method compares the real behavior of the system with the estimated behavior of its nominal model, eliminating non-fault-related factors that have similar effects on both the system and its mathematical model. The efficiency of the suggested model-based bearing fault diagnosis method is validated by comparing simulation and experimental results obtained from the proposed model-based method with a recent signal-based method. The proposed method introduces a novel application of the Luenberger observer for fault detection in induction motors, offering a simple and efficient approach to diagnosing bearing faults. It uniquely distinguishes mechanical faults without direct electrical signal correlation and incorporates a systematic noise cancellation technique, enhancing robustness and accuracy under varying loads.