Investigation of integrated deep learning based electrical connector anomaly detection
The automotive industry’s pivotal role underscores the urgent demand for high-precision anomaly detection in electrical connectors, driven by the surge in connector production and the corresponding increase in defective units. Electrical connectors, integral to signal and energy transmission, are wi...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | en |
| Published: |
IEEE
2026
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/47351/1/Investigation_of_Integrated_Deep_Learning_Based_Electrical_Connector_Anomaly_Detection.pdf https://umpir.ump.edu.my/id/eprint/47351/ https://doi.org/10.1109/SCOReD68498.2025.11399132 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The automotive industry’s pivotal role underscores the urgent demand for high-precision anomaly detection in electrical connectors, driven by the surge in connector production and the corresponding increase in defective units. Electrical connectors, integral to signal and energy transmission, are widely deployed across diverse electronic and automotive systems, where defects can lead to safety hazards, equipment malfunctions, and operational downtime. This research proposes an improved deep learning framework, RES-YOLO, designed to augment manual inspection and mitigate human error. The model synergistically integrates ResNet, recognized for robust feature extraction and defect identification, with YOLO, renowned for real-time object localization and classification. The focus is on Fakra connectors, commonly used in automotive signal transmission, where defect detection accuracy is critical. The architecture employs YOLOv8n as the backbone for rapid detection, while ResNet enhances hierarchical feature extraction, improving sensitivity to subtle connector anomalies. Experimental results demonstrate that YOLOv8n achieves balanced performance across all classes (Precision: 0.770, Recall: 0.779, F1-score: 0.774), albeit with limitations in color-based defect recognition. When excluding the "color" category, the YOLOv8n-ResNet101 hybrid exhibits superior performance in defect detection (Precision: 0.849, Recall: 0.960, F1-score: 0.901) and pin anomaly classification (Precision: 0.979, Recall: 1.000, F1-score: 0.989). These findings validate the effectiveness of the proposed RES-YOLO model in enhancing detection precision and efficiency for automotive connectors, providing a scalable solution for intelligent quality inspection systems in modern manufacturing. |
|---|