Deep Learning Method for Leakage Location Detection of Pneumatic Systems Based on Infrared Thermal Image Evaluation

Pneumatic systems are an essential fluid transmission method in the industrial field, which can achieve the transmission and control of power or signals. Leakage in pneumatic systems is challenging to detect and is a highly harmful fault. Most existing leak detection methods use flow or pressure sensors to detect specific leakage amounts, while leak localization is still in the traditional manual detection, which significantly restricts localization efficiency. This article establishes a deep learning framework-based thermal image localization method for pneumatic system leak detection. Infrared images captured in the pneumatic system were collected, and then a deep-learning localization method based on the YOLO framework was established. Then, the recognition accuracy of this method was calculated. The results indicate that the accuracy has reached 99.5% of mAP_0.5 and 86.27% of mAP_0.5:0.95, indicating that this work is a meaningful study that can apply intelligent computing to engineering.