Data recovery of 2D lifetime-based phosphor thermometry using deep neural networks

Abstract Thermographic phosphor (TP) thermometry has been widely used as one of the newly developed non-contact surface temperature measurement methods. However, temperature information is frequently lost locally because chemical bonding coatings are easily damaged during the measurement. This limits its application, such as in the case of jet impact, high-speed motion, high vibration, etc. We proposed to use of deep neural networks (DNNs) as a tool for recovering lost data. In this study, we captured the dynamic two-dimensional (2D) temperature field of jet impingement cooling a high temperature plate by TP thermometry. Different parts of the temperature information, including jet impact area and non-impact area, were then removed to assume the coating was damaged. A prediction model was established by the DNNs using the 2D space and time coordinates as the input dataset and using the temperature information corresponding to the coordinates as the output dataset. The removed temperature information was then recovered using the developed regression model and compared with the raw temperature field to evaluate predicted results. The results showed that the removed data can be successfully recovered using the established DNN prediction model, in which the predicted accuracy was greater than 94.94%. While in the central and the boundary area of the jet, the model has a relatively poor performance, which is mainly due to the large temperature gradients. These results indicate that the established DNN model can be used to recover the lost temperature information but it is limited for applications in regions with drastic temperature changes.