Akademik Veri Yönetim Sistemi
Iranian Journal of Science and Technology - Transactions of Electrical Engineering, cilt.49, sa.4, ss.1673-1692, 2025 (SCI-Expanded, Scopus)
Electric motors are ubiquitous in daily life. Detecting faults in electric motors before they completely fail is highly important for both economic and safety reasons. In this study, a method capable of identifying subclasses of motor faults was developed using the Machinery Fault Database (MaFaulDa). Additionally, unlike studies in the literature, an artificial intelligence classifier architecture was designed using the artificial bee colony (ABC) algorithm. Three different classifiers based on convolutional neural networks, long short-term memory, and recurrent neural network architectures were designed via the artificial bee colony optimization algorithm. Among these, recurrent neural network-based designs achieved the best performance for the classification of normal, vertical misalignment, horizontal misalignment, overhang, and imbalance. Furthermore, with the proposed new design, the vertical fault subclasses were grouped as 0.51 mm, 1.27 mm, and 1.90 mm. The horizontal fault subclasses were grouped as 0.5 mm, 1.0 mm, and 2.0 mm. For overhang, the subclasses were grouped as ball, cage, or outer. Separating subclasses at the millimeter level allows for early intervention before the critical stage by precisely determining the fault severity and location. The obtained results were presented alongside performance metrics commonly used in the literature.