APPLICATION OF RECURRENT NEURAL NETWORK IN RESEARCH OF INTELLIGENT WIND TUNNEL BALANCE

The shock tunnel ground test is vitally important to the research of the high-enthalpy aerodynamic characteristics of hypersonic vehicles, and the high-accuracy aerodynamic measurement is the key technology. When a force measurement test is conducted in an impulse shock tunnel, the flow field is established instantly after the starting process of shock tunnel, at this time, the great impact loads are acting on the force measurement system. The force measurement system is excited under the action of instantaneous impact, and the inertial vibration signal of the system cannot be rapidly attenuated during the short test time. The output signal of the balance will contain the interference due to the inertial vibration, which leads to a bottleneck in the further improvement of the accuracy of the transient force test. In order to improve the force measurement accuracy in the short-duration shock tunnel, the development of high-accuracy dynamic calibration technology is the key method to improve the performance of balance affected by inertial interference. Therefore, in this paper, recurrent neural network is used to train and intelligently process the balance dynamic calibration data, aiming to eliminate the vibration interference signals in the output dynamic signals. The error analysis of the current method is carried out, and the reliability of the current method is verified. The method is applied to the data processing of force test obtained in shock tunnel, and the effect of inertial vibration on the output signal of the balance is effectively reduced. According to the sample verification analysis of the intelligent model, the relative error of each component load is relatively small, where the case of high-frequency axial force component is about 1%. In the verification of wind tunnel force test data, the good results are also obtained, which are compared with those processed by the convolutional neural network model.