Bearing fault detection by machine learning algorithm using ANN

Abstract

Bearing malfunctions represent the primary cause of motor breakdowns. Decision support systems, including Artificial Neural Networks (ANNs), are extensively employed to identify bearing issues at an early stage. Traditional decision support systems distinguish between feature extraction and classification, treating them as separate steps. Constantly extracting fixed attributes may involve considerable computational effort, limiting real-time application capabilities. Moreover, the chosen attributes for the classification process might not be the best possible selection. In this study, we advocate for the adoption of 1D Convolutional Neural Networks (CNNs) as a solution to streamline and refine the bearing fault detection process. By using the 1D CNN, we integrate the detection system's feature extraction and categorization processes into one cohesive framework. The untouched vibration data (signal) collected from the test rig in Nanyang Technological University (NTU). The suggested system directly accepts the vibration data, eliminating the requirement to execute a distinct feature extraction procedure every time the data is evaluated for categorization. The proposed system's categorization performance with real bearing data will be studied using the various machine learning algorithms and a new algorithm will be proposed at the end of study.