Perceiving flow fields and aerodynamic characteristics of turbomachinery via sparse detection data: a graph data mining model

Accurate state perception is the foundation for the operation and maintenance of turbomachinery. To tackle the challenge of developing operational strategies from poorly interpretable and sparse detection data, a graph data mining model has been proposed, facilitating rapid and accurate perception of flow fields and aerodynamic characteristics from sparse detection data under varying operating conditions. Besides, an aerodynamic characteristics graph compatible with both structured and unstructured meshes has been designed to leverage large flow field datasets and node spatial topology relationships, thereby enhancing accuracy. The proposed model accurately perceived flow fields and aerodynamic characteristics of a centrifugal compressor, with mean squared errors of 1.05 × 10−5 and 2.24 × 10−4, respectively. Moreover, comparative analysis with several traditional models demonstrate that the proposed model possesses superior accuracy. Finally, two sensor layout configurations that can effectively balance accuracy and cost are recommended. The proposed model employs sparse sensors to perceive the flow fields and aerodynamic characteristics of turbomachinery, creating an effective approach to enable intelligent flow control.