Graph convolutional networks modeling for aerodynamic system stability with spatiotemporal topology fusion based on digital twin

Digital twin, as a key direction in the digital transformation, is becoming an important supporting technology for intelligent operation and predictive maintenance of aeroengines. In response to the stability identification of the aerodynamic system, this study proposes a digital twin model based on the flow field topological structure graph convolution, focusing on the topology fusion modeling methods and the recognition strategies of unsteady flow characteristics. This method utilizes the Mapper algorithm in variational mode decomposition (VMD) to extract topological features from high-dimensional time-series data, constructing an adaptive graph structure generation module that reflects the complex dependencies in spatiotemporal sequences. Combining the advantages of time convolutional networks (TCN) and long short-term memory networks (LSTM), a TCN-LSTM time feature extraction module is designed. A spatiotemporal information fusion prediction model is constructed by the graph convolutional networks (GCN) module for the aerodynamic system. The model represents the root mean square error reduction approximately by 81.6% over the benchmark TCN_GCN model. With a prediction performance metric R 2 of 0.9955, the compressor stability state is well achieved by the high-precision prediction. In addition, the study applies spatial domain topological structure, topological entropy, and measure entropy analysis methods to provide an in-depth description of the aerodynamic instability process of dynamic system. The system instability warning can be detected in advance at 251.6 r, 211.6 r, and 3.6 r by instances. This research combines topological theory with graph neural networks, developing a collaborative intelligent data processing framework. It aims at extracting the spatiotemporal structure and predicting the state of dynamic systems, providing multi-dimensional information for system recognition.