Bearing Fault Detection Using Ensemble Subspace k-Nearest Neighbors

Parts of rotating machines are susceptible to failure, which may lead to significant financial losses and safety risks. In this work, the Ensemble Subspace K-Nearest Neighbours (KNN) and Ensemble Bagged Tree machine learning techniques are applied to identify bearing defects. Four motor conditions are tested under full load conditions: healthy and three types of bearing defects, namely ball, inner race, and outer race. These bearing faults result from wear and fatigue. Condition monitoring plays a crucial role in the early detection of bearing defects, ensuring timely maintenance and preventing catastrophic failures. Machine learning techniques have emerged as powerful tools in this domain, enabling automated fault detection with high accuracy. By leveraging historical vibration data, machine learning models can learn complex patterns associated with different fault types, improving the reliability of diagnosis. Traditional methods for detecting bearing faults can be complex and time-consuming. However, implementing the Ensemble Subspace KNN and Ensemble Bagged Tree models streamlines the fault detection process, making it faster and more efficient. The models were fine-tuned and optimized by selecting the best hyperparameters, including the learning rate and optimizer. Notably, the Ensemble Subspace KNN model achieved an accuracy of 98.7%, demonstrating its high precision and reliability in diagnosing bearing faults. This approach ensures an effective, robust, and precise fault detection system for rotating machinery