TY - JOUR
T1 - Deep Neural Networks and Proper Orthogonal Decomposition-Based Parameterized Reduced-Order Model and its Application in Transonic Axial-Flow Compressor Blade
AU - Tan, Chunlong
AU - Gao, Hangshan
AU - Li, Lei
AU - Li, Honglin
N1 - Publisher Copyright:
© The Author(s), under exclusive licence to Springer Nature B.V. 2025.
PY - 2025
Y1 - 2025
N2 - Modern turbomachinery blades are facing increasingly pronounced aeroelastic challenges with the increase of unsteady loads and the widespread use of lightweight materials. Conventional coupling methods fail to analyze this issue efficiently due to tremendous dimensionality difference between fluid and structure domains. To address this critical bottleneck, a novel parameterized reduced-order model (PROM), based on deep neural networks (DNN) and proper orthogonal decomposition (POD), was proposed and validated in this study. The framework operates through two synergistic phases. The first stage was dimensionality reduction, in which, POD was employed to extract flow field modes and determine corresponding mode coefficients. The second stage was parameters mapping, where a DNN model was constructed and trained to learn the nonlinear relationship between design parameters and mode coefficients. Finally, the efficacy and robustness of the PROM approach are demonstrated using Rotor 67, a typical transonic axial-flow compressor. The results show that the proposed PROM has an excellent performance in flow field prediction and the maximum relative error less than 5%. Moreover, a well-trained PROM can accurately determine the pressure distribution over the surfaces of compressor blade in just 0.03 s, effectively enabling real-time simulations. This advancement holds significant promise for enhancing aeroelastic analysis in turbomachinery blade design.
AB - Modern turbomachinery blades are facing increasingly pronounced aeroelastic challenges with the increase of unsteady loads and the widespread use of lightweight materials. Conventional coupling methods fail to analyze this issue efficiently due to tremendous dimensionality difference between fluid and structure domains. To address this critical bottleneck, a novel parameterized reduced-order model (PROM), based on deep neural networks (DNN) and proper orthogonal decomposition (POD), was proposed and validated in this study. The framework operates through two synergistic phases. The first stage was dimensionality reduction, in which, POD was employed to extract flow field modes and determine corresponding mode coefficients. The second stage was parameters mapping, where a DNN model was constructed and trained to learn the nonlinear relationship between design parameters and mode coefficients. Finally, the efficacy and robustness of the PROM approach are demonstrated using Rotor 67, a typical transonic axial-flow compressor. The results show that the proposed PROM has an excellent performance in flow field prediction and the maximum relative error less than 5%. Moreover, a well-trained PROM can accurately determine the pressure distribution over the surfaces of compressor blade in just 0.03 s, effectively enabling real-time simulations. This advancement holds significant promise for enhancing aeroelastic analysis in turbomachinery blade design.
KW - Axial-flow compressor
KW - Deep neural networks
KW - Parameterized reduced-order model
KW - Proper orthogonal decomposition
UR - http://www.scopus.com/inward/record.url?scp=105003885077&partnerID=8YFLogxK
U2 - 10.1007/s10494-025-00656-5
DO - 10.1007/s10494-025-00656-5
M3 - 文章
AN - SCOPUS:105003885077
SN - 1386-6184
JO - Flow, Turbulence and Combustion
JF - Flow, Turbulence and Combustion
ER -