Design of an Air Temperature Sensor Using CFD and Neural Network-Based Error Correction

Air temperature sensors often exhibit measurement inaccuracies due to solar radiation effects, leading to overestimations of actual air temperature by up to $1~^{\circ }$ C. To address this issue, a novel temperature sensor with enhanced radiation shielding and ventilation was developed. The proposed design incorporates an S-shaped flow diversion device, two silver-plated mirrored aluminum plates, two hemispherical flow diversion structures, and a sensor probe. Computational fluid dynamics (CFDs) simulations were conducted to evaluate radiation shielding efficiency and airflow performance. Radiation-induced errors were assessed under varying environmental conditions, and a neural network algorithm was implemented to develop a correction model. Experimental results showed that, prior to correction, the average radiation-induced errors for the new sensor, the 41 003 model, and the 43 502 model were 0.443°C, 0.533°C, and 0.317°C, respectively, with maximum errors of 0.996°C, 1.42°C, and 0.905°C. After correction, the average radiation-induced error of the new sensor decreased to 0.147°C, with a maximum of 0.485°C. The 41 003 and 43 502 models achieved reductions to 0.171°C (maximum: 0.598°C) and 0.118°C (maximum: 0.37°C), respectively. The new sensor demonstrated a mean absolute error (MAE) of 0.434°C, root mean square error (RMSE) of 0.351°C, and correlation coefficient ( ${r}$ ) of 0.855 in corrected radiation-induced errors compared to experimental data. In contrast, the 41 003 model exhibited an MAE of 0.604°C, RMSE of 0.471°C, and r of 0.832, while the 43 502 model achieved an MAE of 0.268°C, RMSE of 0.218°C, and r of 0.901. These findings highlight the effectiveness of the new sensor design and correction model in mitigating radiation-induced measurement errors, providing improved accuracy for atmospheric temperature sensing applications.