Application of convolutional neural network in static calibration of wind tunnel balance

As the most important measuring device in aerodynamic testing, the wind tunnel balance is used to measure the magnitude, direction, and point of the aerodynamic loads (forces and moments) acting on the test model. The accuracy of the measurement is directly related to the static calibration of the wind tunnel balance, which establishes the mapping relationship between the balance output signals and aerodynamic loads on the calibration equipment. This paper explores the possibility of improving the calibration performance of the strain-gauge balance in the calibration system AiBCS, developed by Institute of Mechanics of Chinese Academy of Science, using the convolutional neural network (CNN). The applicable conditions, validity, and reliability of CNN in the balance calibration are discussed and evaluated. Results obtained by the CNN-based calibration method and the traditional polynomial fitting method are analyzed and compared. It turns out that the CNN-based calibration method can effectively reduce the load interference between various balance components, yielding a significantly improved performance. Consequently, the deep-learning technology shows great application potential in calibrating wind tunnel balance.