<![CDATA[This study presents a real-time vision-based system for detecting the open and closed positions of pneumatic pistons in industrial production lines without using physical sensors. Conventional magnetic and inductive sensors are often affected by cable damage, environmental contamination, vibration, and temperature variations, which can cause unplanned downtime and increased maintenance costs. To address these limitations, a camera-based monitoring approach is proposed as a reliable and low-maintenance alternative.The main objective of this work is to develop a low-cost, robust, and easily integrable sensorless position-detection system using deep learning–based object detection. A dataset consisting of 250 RGB images was collected from a production-like test platform and annotated into two classes representing open and closed piston states. The dataset was split into training and testing sets with ratios of 80% and 20%, respectively.A YOLOv8 object detection model was fine-tuned using transfer learning and deployed on a Raspberry Pi 4B for real-time operation. To improve reliability, a high confidence threshold and a frame-based stability filter requiring consistent predictions across multiple frames were applied. Detected piston states were converted into digital control signals via GPIO outputs.Experimental results show that the proposed system achieves over 97% detection accuracy with a processing latency of 25–40 ms per frame on embedded hardware. The stability filter effectively reduces false state transitions, ensuring reliable output. The results indicate that the proposed approach provides a practical visual backup solution for sensor failures and a scalable alternative for new production line designs.]]>
