Detection of Stator Faults in Three-Phase Induction Motors Using Stray Flux and Machine Learning
Three-phase induction motors are widely applied in industrial systems due to their durability and efficiency. However, electrical faults such as inter-turn short circuits can compromise performance, leading to unplanned downtime and maintenance costs. Traditional fault detection methods rely on stat...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-03-01
|
| Series: | Energies |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1996-1073/18/6/1516 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Three-phase induction motors are widely applied in industrial systems due to their durability and efficiency. However, electrical faults such as inter-turn short circuits can compromise performance, leading to unplanned downtime and maintenance costs. Traditional fault detection methods rely on stator current or vibration analysis, each with limitations regarding sensitivity to specific failure modes and dependence on motor power ratings. Despite advancements in non-invasive sensing, challenges remain in balancing fault detection accuracy, computational efficiency, and adaptability to real-world conditions. This study proposes a stray flux-based method for detecting inter-turn short circuits using an externally mounted search coil sensor, eliminating the need for intrusive modifications and enabling fault detection independent of motor power. To account for variations in fault manifestation, the method was evaluated with three different relative positions between the search coil and the faulty winding. Feature extraction and selection are performed using a hybrid strategy combining random forest-based ranking and collinearity filtering, optimizing classification accuracy while reducing computational complexity. Two classification tasks were conducted: binary classification to differentiate between healthy and faulty motors, and multiclass classification to assess fault severity. The method achieved 100% accuracy in binary classification and 99.3% in multiclass classification using the full feature set. Feature reduction to eight attributes resulted in 92.4% and 85.4% accuracy, respectively, demonstrating a trade-off between performance and computational efficiency. The results support the feasibility of deploying stray flux-based fault detection in industrial applications, ensuring a balance between classification reliability, real-time processing, and potential implementation in embedded systems with limited computational resources. |
|---|---|
| ISSN: | 1996-1073 |