Machine Learning-Based Motor Health Prediction for Enhanced Lifespan Management

The aim of the research is to develop a model for predicting the condition of the motor (normal or faulty) by using the vibrational data recorded from the motor. The main question of the research is to find out, “Are these fault conditions predictable based on the vibrational data or Is vibrational data enough to predict different fault conditions of the motor using machine learning and deep learning?” The objective of the research are to find the core differences of the machine learning and deep leaning strategies in managing the fault prediction of the motors and compares different performance parameters i.e. accuracy of finding the condition of the motor. The study progresses with highlighting the key differences between machine learning and deep learning techniques and the set of principles required to make the predictions accurately. The research initiates with focusing on vibrational analysis of the motors and their key importance in detecting the faults of the motors. Most of the motor’s faults such as imbalance, poor lubrication, bearing defect etc. are related to vibrations. Therefore, by extracting the meaningful pattern of the vibrational data and then training of the model on such pattern could leads in predicting the fault of the motor. In this study, the vibrational dataset have been collected in controlled environment of ABB laboratories for varying speed-load conditions i.e. Normal or Faulty . The proposed methodology encompass both deep and machine learning algorithms in predicting the faults. Machine learning algorithms such as Support Vector Machine and the K-Nearest Means have been applied to the feature set fabricated from time, frequency, and time-frequency domain. In case of deep learning, Siamese Architecture with Feed Forward Neural Network have been applied to map between the feature vector and the desired output automatically. The results of the employed algorithms have demonstrated that they can assist in the prediction of motor faults. The prediction of the motor condition directly correlated with the quality and the quantity of vibrational data of the motor. Furthermore, effective data processing, such as filtering, normalization, and transformation, could improve the accuracy of the developed models.