Classification-based approach for fault diagnosis in rotating machinery using vibration signal analysis

Abstract

Vibration analysis has emerged as one of the most effective techniques for condition monitoring, providing valuable insights into the state of rotating machinery. Consequently, the use of vibration signals for diagnosing faults has become a standard tool in various industries. This thesis proposes a machine learning classification approach that leverages time and frequency domain features, thoroughly selected to enable accurate fault diagnosis. The classification is performed using multiple algorithms, namely Support Vector Machines (SVM), Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Decision Trees, and Bagged Trees. These algorithms have been tested extensively through different experiments to validate the effectiveness of the proposed approach. Furthermore, we conducted an investigation to identify the most relevant and impactful features in the classification process using the Minimum Redundancy Maximum Relevance (MRMR) as a feature selection algorithm. This analysis helps in identifying the key features that contribute significantly to the fault classification accuracy. The results revealed that SVM exhibited superior accuracy and outperformed other classifiers in most evaluation metrics. KNN demonstrated robustness in noisy environments, while ANN exhibited the fastest prediction time. Bagged trees consistently classified faults using a reduced number of features