A General Super-Resolution Approach Integrating Physical Information for Temperature Field Measurement

In industrial measurement, temperature field measurement typically relies on thermocouples and spectroscopic techniques. These traditional methods often suffer from insufficient precision, resulting in prevalent low-resolution measurements in real thermal scenarios. To address this challenge, we propose a novel general super-resolution approach for temperature field measurement in various thermal scenarios, leveraging the low-resolution (LR) data obtained from sensor array technology. The method incorporates skip connections and multi-path learning, along with physical information loss, to enhance accuracy. To validate the effectiveness of the approach, simulations across three two-dimensional thermal scenarios are conducted: the heating process in silicon chips, the thermodynamic process of hot and cold water mixing, and the convective heat transfer phenomena involved in metal sheet dissipation under airflow. The results show that the learning model can accurately predict the HR temperature. The proposed approach offers a pathway for generating HR solutions, bypassing traditional time-consuming simulation processes while ensuring data accuracy. By utilizing a fixed model and a lightweight physical loss function, we simplify the deployment process, facilitating applications in computational fluid dynamics (CFD) solutions, engineering measurements, and related fields.