Real-Time Interface-Opening Detection for Nickel Starting Sheet Pre-Stripping Based on an Improved YOLO11n

Area of Science:

• Materials Science and Engineering
• Computer Vision and Machine Learning
• Industrial Automation

Background:

• Accurate real-time recognition of the interface opening between nickel and titanium is crucial for controlling high-frequency tamping during nickel starting sheet pre-stripping.
• Existing methods face challenges with small target scales, weak textures, reflective backgrounds, and motion blur.

Purpose of the Study:

• To develop a deployment-oriented object detection system, Yolov11_tapcheck, for real-time field detection of nickel starting sheet pre-stripping.
• To enhance detection accuracy and reduce latency for closed-loop control in industrial processes.

Main Methods:

• Proposed Yolov11_tapcheck detector based on YOLO11n, incorporating efficient multi-scale attention (EMA) and a BiFPN-inspired bidirectional feature-fusion path.
• Established a dedicated 'tapcheck' dataset from production-line images.
• Evaluated performance using image-level accuracy, latency metrics, and field control data.

Main Results:

• Yolov11_tapcheck improved mAP50 from 95.5% to 98.1% and F1 score from 93.7% to 97.5% compared to the original YOLO11n.
• Achieved an average model-pipeline latency of 7.1 ms, with frame-to-PLC acknowledgement latency averaging 10.24 ms.
• Field tests showed opening success rates increased from 60-65% to over 90%, and detection success reached 100%.

Conclusions:

• The Yolov11_tapcheck system effectively supports real-time recognition of the transient interface opening.
• The proposed method enables reliable closed-loop control for nickel starting sheet stripping processes.
• The system demonstrates significant improvements in accuracy and success rates in industrial field applications.