@Article{cmc.2026.085491,
AUTHOR = {Juhui Zhang, Zongyi Xing, Chenxiao Cai},
TITLE = {Motor Temperature Detection System Based on Multi-Scale Feature Screening and Thermal Image Reconstruction},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/cmc/online/detail/28285},
ISSN = {1546-2226},
ABSTRACT = {As urban rail transit systems continue to expand, greater technical demands are placed on the reliability assessment and temperature monitoring of key bogie components under dynamic operating conditions. To address the complexity of state detection caused by motion blur in high-speed targets, this study proposes a motor temperature detection system based on dynamic infrared video analysis. The system evaluates motor operating conditions by rapidly screening motor images and applying super-resolution reconstruction techniques. First, the system’s configuration and operating principles are described. Next, an infrared image screening algorithm is developed based on temperature feature extraction and matching. The algorithm incorporates sliding window grouping and template matching to quickly identify motor targets in infrared video. Finally, the Perception ResNet Attention Generative Adversarial Network (PRAGAN) super-resolution network is employed to improve the spatial resolution of thermal images. It utilizes a multi-scale residual encoder, Visual Geometry Group (VGG)-based perceptual loss, and an uncertainty-based adaptive loss-weighting strategy whose update rule is given explicitly. The system was deployed and validated on Nanjing Metro Line 1. Experimental results demonstrate that the system achieves a temperature measurement accuracy of 2°C and operates reliably at a train speed of 55 km/h. It maintains high measurement accuracy while minimizing equipment complexity, indicating strong engineering applicability.},
DOI = {10.32604/cmc.2026.085491}
}