Digital twin-assisted multiscale residual-self-attention feature fusion network for hypersonic flight vehicle fault diagnosis

Hypersonic flight vehicle (HFV) with long term exposure to poor operating environments will inevitably experience performance degradation and potential failures. Currently, data-driven approaches have been commonly used for fault diagnosis. However, it is a challenge to obtain reliable and adequate data to identify HFV faults. To cope with this issue, this paper put forward a digital twin-assisted multiscale residual-self-attention feature fusion network (MRFFN) for hypersonic flight vehicle fault diagnosis. Firstly, a mathematical simulation is performed to establish the DT model of HFV. Then, the constructed DT model is employed for simulating multiple fault states of HFV to generate an approximation to the real system state data. Finally, a novel MRFFN is designed for training and validation utilizing the data derived from the DT model. The comparison performance demonstrates the MRFFN is superior to other intelligence methods in its ability to accurately identify hypersonic flight vehicle faults.