Abstract The detection of bus wheelset bearing defects is a critical and essential challenge due to its direct impact on the safety and efficiency of public transportation systems, as well as the reduction of maintenance and repair costs. Bearing failures can lead to sudden stoppages, increased repair expenses, and, if not detected in time, can result in hazardous incidents such as wheel lockups or looseness. This study aims to leverage machine learning tools for the specific application of intelligent fault detection in bus wheelset bearings. Vibration data from bus wheelset bearings were collected and analyzed under real road and operational conditions using precise sensors to ensure that the developed models align with real-world scenarios. Initially, the data underwent preprocessing, and key features, including statistical parameters, were extracted. Subsequently, machine learning algorithms, including Support Vector Machines (SVM), Ensemble Learning based on SVM, and K-Nearest Neighbors (KNN), were employed to develop fault detection models. The models were trained on the collected data and then eva‎luated on test datasets. The results demonstrated that SVM models with radial basis function (RBF) kernels and Ensemble Learning based on SVM with RBF kernels outperformed other methods in detecting bearing faults, offering higher detection accuracy. These models were capable of early fault detection and could effectively prevent sudden failures. The application of advanced machine learning techniques in this study contributes significantly to improving safety, reducing maintenance costs, and enhancing the efficiency of public transportation systems. This research underscores the high potential of signal processing and machine learning methods in enhancing the performance and safety of complex mechanical systems under operational and real-world conditions.

رضا تيكني

سعيد ضيائي راد , علي لقماني