Using machine learning in the detection of faulty ball bearings and the development of the internet of things based condition monitoring and warning system

In these times of the fourth industrial revolution, the Internet of Things (IoT) facilitates human life in the field of industrial maintenance, as in many areas. By using IoT systems, condition monitoring of engine bearings and early warning system, it is possible to operate the system properly without serious damage. In this thesis, a unique condition monitoring and warning system is designed, which is capable of closely and remotely real-time status monitoring with IoT technology, using low-cost and mostly open-source hardware and software. This design consists of three main parts. These are the experimental setup, IoT-based mobile condition monitoring applications, and attributes based detailed data analysis with machine learning (ML) algorithms. Attributes such as mechanical vibration acceleration, bearing relative temperature, sound intensity, current, voltage and rotational speed were investigated in classifying the bearing in the experimental setup as healty or faulty. Among the investigated attributes, it was observed that the relative bearing temperature, sound intensity and mechanical vibration acceleration values contributed to the better classification of the ML models. The designed condition monitoring and warning software enable the data coming from the experimental setup to be monitored in real-time on mobile devices via Bluetooth connection and cloud server via IoT technology. These applications can send SMS, e-mail or audible warning notifications to the maintenance team in problem detection. In the last stage of the study, the data collected from the experimental setup was modeled by twelve ML algorithms considering eight attributes. All models were evaluated in detail with sixteen different metrics based on the 95% confidence interval. Although ML models give different results depending on the features, XGBoost and GBT algorithms have produced remarkable results with fast and high classification success.