Linear and nonlinear combined aerodynamic reduced order model based on residual network framework

Residual convolutional neural network (ResNet) is a rising technology in the field of artificial intelligence due to its advantages of easy optimization, high accuracy, information fidelity and no gradient disappearance. Based on the powerful ability of machine learning to learn from big data featur...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Europhysics letters 2022-06, Vol.138 (6), p.63002
Hauptverfasser: Qi, Hui, Yu, Jiaming, Jiang, Jingjiang, Guo, Jing
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Residual convolutional neural network (ResNet) is a rising technology in the field of artificial intelligence due to its advantages of easy optimization, high accuracy, information fidelity and no gradient disappearance. Based on the powerful ability of machine learning to learn from big data features, a residual network aerodynamic reduced order model (ROM) framework based on CFD sample data was proposed. In this letter, a binary airfoil is taken as the research object, and the displacement information representing the movement characteristics of the flow field can be transmitted by two ways: one is that the input information is continuously captured and extracted by the deep convolution layer; the other is that the input information is directly transmitted around the convolutional layer. Then the characteristic information of the two approaches is integrated and activated through the output layer to complete the regression prediction of the wing aerodynamic characteristics. Finally, the ROM frame is used to predict the strongly nonlinear hysteresis loops of airfoil under pitching motion, and the results show that the ROM frame is effective. In addition, the framework can be applied to qualitative interpretation of complex flow phenomena and aeroelastic analysis.
ISSN:0295-5075
1286-4854
DOI:10.1209/0295-5075/ac765e