A novel dual attention network for sparse reconstruction of turbine blade surface fields

Zuobiao Li, Fengbo Wen, Zhongqi Liu, Yuxi Luo, Zhiyuan Zhao, Dongsheng Wen and Songtao Wang

Energy, 2025, vol. 317, issue C

Abstract: Achieving accurate and robust reconstruction of turbine blade surface field from randomly distributed sparse sensors has been a longstanding challenge, which is of high importance for cascade experimental data processing and gas turbine operation and maintenance. As a solution to this problem, this work proposes a novel dual attention network, which consists of a spatial analysis network (SAN) and a multi-scale shifted windows transformer network (MS-SwinT), for arbitrarily positioned sensors of any number that exceeds inferior limits. The new network is examined against a case study of 3D flow in a classical axial turbine blade geometry with complex wave structures. Upon integrating multilayer perceptron with shared weights and attention units, permutation invariance is achieved, and the prediction results are independent of the input order of the measurements. The number of sensors Ns and the isentropic exit Mach number (Ma2s) are the pivotal determinants of the flow field reconstruction quality. The proposed method requires 32 measurement points (Ns = 32) per side to ensure the reconstruction precision, corresponding to a super-resolution ratio of rsup = 1024. Besides, MS-SwinT determines the reconstruction quality of the joint model at small Ma2s, whereas the deviation in boundary conditions inferred by SAN emerges as the primary factor for the reduced performance at higher Ma2s. The joint model can enhance the spatial resolution of experimental data. Individually, the SAN facilitates the inference of boundary conditions from limited measurements to assist the control systems in decision-making. MS-SwinT serves as a powerful tool for virtual experiments, improving design efficiency by reducing the need for physical experiments and data acquisition costs. The case study demonstrates the great potential of the proposed method in improving the quality of surface field construction with sparse and random data, which is of high value for future applications in gas turbines.

Keywords: Gas turbines; Flow field reconstruction; Super-resolution; Attention network; Randomly distributed sensors; Permutation invariance (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0360544225002865
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:energy:v:317:y:2025:i:c:s0360544225002865

DOI: 10.1016/j.energy.2025.134644

Access Statistics for this article

Energy is currently edited by Henrik Lund and Mark J. Kaiser

More articles in Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().