Digital Twin-Assisted Multiview Reconstruction Enhanced Domain Adaptation Graph Networks for Aero-Engine Gas Path Fault Diagnosis

This article proposes a digital twin (DT)-assisted multiview reconstruction enhanced domain adaptation graph networks (MRDANs) to improve the diagnostic accuracy and adaptability to performance degradation of the aero-engines gas path system (AGPS). First, the DT model with sufficient multicondition data to lay the foundation for subsequent experiments is obtained. Then, convolutional neural networks (CNNs) are used to expand the view of multiple feature spaces. Further, a graph-based multiview reconstruction (MR) method is designed for feature extraction. This approach simultaneously considers the topology and node feature on the graph by constructing a learnable adjacency matrix to tune the topology in the reconstructed graph and placing random walk kernels on different graphs. Next, graph neural network (GNN) is used to perform feature extraction on the reconstructed graph, while the proposed feature harmonized constraint (FHC) is combined with domain adaptation. Finally, the comparison experiment is given, exhibiting that, the proposed framework performs better fault feature extraction ability and domain transfer ability in gas path fault diagnosis.