A Transfer Learning Approach for Effective Motor Fault Identification of Industrial Machines used in Tile Manufacturing

• Published in: National Academy science letters Vol. 45; no. 5; pp. 405 - 409
• Main Authors: Kovilpillai, J. Judeson Antony, Jayanthy, S.
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 2022; Springer Nature B.V
• Subjects: Accuracy; Deep learning; History of Science; Humanities and Social Sciences; Imports; Induction motors; Machinery; Manufacturing; Motor skill learning; multidisciplinary; Science; Science (multidisciplinary); Short Communication; Transfer learning; Vibration; Deep learning; Condition monitoring; Fault identification; Transfer learning
• ISSN: 0250-541X; 2250-1754
• DOI: 10.1007/s40009-022-01145-3

As the development and usage of autonomous machines and industrial IoT in manufacturing/production companies is becoming more active, machine fault identification has become a core component in any industrial environment. This research proposes a fault identification deep learning methodology to identify motor failures in industrial machines used in tile manufacturing. The deep learning model imports open-source machinery fault dataset of induction motors used in tile industries which contains parameters such as rpm, vibration, radial, axial and tangential direction values obtained using high-end sensors interfaced with an industrial test rig known as MFS—Machinery Fault Simulator. Furthermore, the model is fine-tuned using transfer learning techniques using pre-trained networks, and the performance of the model is assessed using accuracy metrics like Kappa Statistic: K, Overall Accuracy: OA and Average Accuracy: AA. The accuracy rate of 97.61% proves the effectiveness of the proposed fault identification model, thereby ensuring proficient and smooth operation of industrial machines.