Effective Diagnosis Approach for Broken Rotor Bar Fault Using Bayesian-Based Optimization of Machine Learning Hyperparameters

Rotor fault diagnosis plays a critical role in ensuring the safety and reliability of rotating machinery. Recently, there has been increasing interest in leveraging advanced signal processing and machine learning techniques to improve fault detection accuracy and overall performance. This study applies the Discrete Wavelet Transform (DWT) for feature extraction from current signals and explores the effectiveness of hyperparameter optimization, specifically Bayesian Optimization (BO), in conjunction with machine learning algorithms to classify rotor health conditions accurately. The primary objective is to differentiate between rotors with four fractured bars and those in a healthy state. Several classification methods are evaluated, including Support Vector Machine (SVM), k-nearest Neighbors (KNN), Random Forest (RF), Extra Trees (ET), and Decision Trees (DT), with accuracy, precision, recall, and F1 score as key performance metrics. Experimental results demonstrate that the combination of BO and RF achieves the highest accuracy, at 96.92%, with a precision of 96.6825%, recall of 96.68%, and F1 score of 96.84%. Additionally, SVM, KNN, ET, and DT also exhibit strong performance in detecting and classifying broken rotor bar (BRB) faults based on their severity. These findings underscore the potential of combining BO with machine learning models to enhance fault diagnosis in rotating machinery.