Modeling the Health Status of a Ball Bearing for Predictive Maintenance Purposes

Modern industries are constantly aiming to maximize utilization and performance of their machine equipment and minimize costly and unscheduled downtime. In the recent years, this was made possible with the advancement of technology and the coming of the fourth industrial revolution, also known as Industry 4.0, which introduced the Internet of Things and Artificial Intelligence systems in industrial applications. This allowed the employment of predictive maintenance methods able to assess the health of the equipment and diagnose possible future failures. In this work, an experimental assembly was constructed comprised by an electrical motor, a reduction gear, an axle, a coupler, a bearing and a vibration sensor attached to the latter. Three types of experiments were conducted for the purpose of producing a labeled dataset based on vibration measurements. Specifically, measurements were made during the bearing's working in normal operating state, one with lack and one with excess amount of grease. A preliminary statistical analysis was performed while a Multilayer Perceptron model was employed to automatically predict the status of the bearing. The dataset is freely available at the Zenodo data repository.