Exploring dataset availability and methods for enhanced automated fault detection in induction motors

Nguessan R. Tey
,
Vladan Velisavljevic
,
Vladimir Dyo
,
Massoud khodadadzadeh
,
Jake Newton
,
Barry Newton

University of Bedfordshire
,
School of Computing Engineering and Creative Industries
,
Royal Holloway University of London
,
Uptime Systems Ltd.

Research Output:

Contribution to journal

Article

Peer-review

Open access

Abstract

Induction motors are widely used in industrial applications due to their reliability, efficiency, and cost-effectiveness. However, unexpected faults still occur, leading to unplanned downtime and increased operational costs. The effectiveness of fault detection systems is fundamentally dependent on the availability of high-quality and representative datasets. Fault-detection systems are most effective when trained on high-quality and representative datasets. This paper presents a comprehensive evaluation of publicly available induction motor datasets and machine-learning methods that can help detect faults. The study provides a structured dataset taxonomy and identifies key limitations reported in the literature. We investigated datasets acquired through experimental studies and analyzed sensor-based measurements, focusing on vibration, current, and acoustic modalities. Key challenges include the lack of real-world data, limited standardization, selective fault representation, and incomplete metadata. Multiple datasets are lab-based, with variations in size, structure, and sensor placement that limit their generalization to machine learning models. The study further highlights the importance of multi-sensor integration, improved fault labeling, and open-access data sharing to enhance dataset quality and support the development of standardized, comprehensive datasets for reliable induction motor fault detection.