A blended method incorporating a multi-model for pulse wind tunnel aerodynamic identification considering large-scale aircraft. (1st January 2023)

• Record Type: Journal Article
• Authors: Li, Shichao; Wang, Qinchao; Ma, Guilin; Gao, Hongli; Guo, Liang; Liu, Zihao; Shao, Kaibo
• Is Part Of: Measurement science & technology. Volume 34:Number 1(2023)
• Publication Date: 2023-01-01
• Subjects: impulse wind tunnel -- deep learning -- aerodynamic force identification -- strain balance -- force measurement system; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac900e
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: Pulse wind tunnel aerodynamic tests are an important part of hypersonic aircraft design and development. With the development of hypersonic aircraft technology, the aircraft model in pulsed wind tunnels has gradually become large-scale and heavy-loaded. During the force measurement test, due to the influence of the increased size of the aircraft model, the real aerodynamic signal of the force measurement system is submerged by the interference signal. During the effective test time of several hundred milliseconds, it is impossible to obtain high-precision aerodynamic signals of the force measuring system using traditional signal-processing methods. Therefore, for the in-depth study of aerodynamic identification of short-time hypersonic pulsed wind tunnels, there is an urgent need for an identification algorithm that can adapt to the force measurement system of large-scale aircraft. In this paper, a new aerodynamic identification algorithm that combines a traditional signal-processing method and a deep neural network is proposed and applied to pulse combustion wind tunnels. The algorithm is mainly divided into signal preprocessing and deep learning. First, the original signal is decomposed into different sub-signals via variational modal decomposition (VMD), and then the real aerodynamic signal is obtained via convolutional neural network (CNN)-long short-term memory (LSTM) training. For different interference signals in the pulsed wind tunnel test, this algorithm innovatively designs VMD preprocessing and optimizes hyperparameters, to extract signal features for subsequent deep learning and to filter out interference components. To ensure the consistency between the validation and the application, the algorithm was verified using the suspension test bench during training, and satisfactory results were obtained. Finally, the algorithm is applied to the suspension force measurement system of a pulse combustion wind tunnel, and the aerodynamic identification results present satisfactory accuracy.