Result: Use of motor current signature analysis for rotor imbalance detection using ML-based classification

In light of the emergence of Industry 4.0, fault diagnosis strategies are predominantly oriented towards data-driven methods. The work presented demonstrates a case study on investigating rotor fault diagnosis using motor current variation and predicting the severity of fault using Machine Learning algorithms. Techniques based on vibration signals and acoustic emission techniques usually require robust and costly sensors and are prone to industrial noise, which makes it harder to classify fault severity. The goal of the work is to develop a systematic, cost-effective, current sensor-based methodology for accurate detection of severity of mass imbalance and fault severity, particularly for multi- class imbalanced data. Experiments are performed on an in-house developed test rig, and faulty data is collected for different imbalance severity. Various commonly used machine learning algorithms are employed to classify the severity of the faults. Fourier transformation is employed to transform signals into the frequency domain and analyse the contributions of various frequency components to the overall signal. Features are extracted from this frequency domain signal which are fed to various machine learning algorithms for training and testing. K- Nearest Neighbour, Support Vector Machine, Linear discriminant analysis and Quadratic discriminant analysis algorithms have been employed in the case study. The results indicated that Linear discriminant analysis and Support Vector Machines have an equivalent accuracy of 98.3 % in classifying the faults, while KNN and Quadratic discriminant analysis classify defects with accuracy of 96.7 % and 95.0 %, respectively. The study demonstrates the effectiveness of the current signal analysis in accurately detecting and classifying rotor mass imbalance severity and contributing to the advancement of MCSA-based predictive maintenance strategies for rotor-bearing systems.