EPIL-Attention: An Edge-Position-Illumination-Layout Guided Attention Mechanism for Fault Detection in Spinning Equipment Operation

In the automated production lines of the spinning industry, the operational status of equipment significantly impacts product quality, production efficiency, and safety. Existing deep learning detection algorithms, such as YOLO, perform well on natural images but face challenges like object edge truncation and metal reflection interference in spinning equipment fault detection. This paper proposes a novel attention mechanism - EPIL-Attention (Edge-Position-Illumination-Layout Attention) - designed to enhance the model's ability to capture structural features in spinning equipment images. EPIL-Attention comprises five parallel branches: horizontal structure perception, mid-section response, edge modeling, illumination adaptation, and spatial layout guidance. These are fused to output enhanced feature maps. We construct a real-world dataset and integrate EPIL-Attention into YOLOv11. Experiments demonstrate that EPIL-Attention consistently improves detection accuracy while maintaining practical deployment efficiency, thus showing promising detection performance and deployment potential.