Non-Invasive Fault Detection and Localization of Induction Motors

Abstract

With the rapid rise of Industry 4.0, the growing use of sensors, the rapid development of the Internet of Things (IoT), and the use of artificial intelligence techniques, smart factories may automate their operations to greatly enhance their efficiency and quality of output. However, ultimately, even the most well-maintained equipment have defects. Given that Prognostics and Health Management (PHM) is essential for optimal machine performance, Predictive Maintenance (PdM) is an emerging topic within maintenance methodologies with the objective of predicting failure prior to its occurrence in order to schedule maintenance only when it is necessary. Deep learning is a useful technique for using big data for data-driven fault diagnostic approaches, since data can be generated at an unprecedented rate. Diagnostic of faults of induction motors is playing an important role in industries. Fault detection of induction motors is one of the most noteworthy need of the industries. Several motor components (rotor, bearings, insulation, stator and rotor circuits) deteriorate with time and stress. Degraded electric rotating motor parts can lead to machine accidents and downtimes. Localizing faults, repairing, or replacing a damaged motors takes time and money. In this thesis, a non-invasive acoustic signal-based fault monitoring and localization using machine learning (ML) will be designed and analyzed for the induction motors. A mic will be installed near induction motor to extract the dataset of faulty motors and healthy motors based on acoustic signals. Audial based fault detection of induction motors monitors the system and detects the faults earlier to improve the technical issues, cost reduction and high reliability. We will develop this system by using machine learning taking the computational values of the acoustic signals, splitting the audio, amplitude scaling. This thesis investigates two distinct fault diagnosis approaches related with predictive maintenance: anomaly identification via a fault classifier recurrent neural network and failure mode and effects analysis (RNN). The technology under consideration is an industry-standard AC induction motor. The results demonstrate excellent performance and suggest the method's potential for industrial applications.