An optimized VMD and global spatiotemporal model for diagnosing inter-rotor rub-impact faults in dual-rotor turboshaft engine

Abstract The next-generation dual-rotor turboshaft engine adopts a bearing cavity structure with flexible stators to reduce weight. However, under complex operating conditions, this structure promotes coupled nonlinear vibrations and rub-impact issues between the rotors. Inter-rotor rub-impact in a dual-rotor turboshaft engine is often obscured by rotor-to-stator rub faults and environmental noise, making its identification challenging and posing a significant threat to the safe operation of helicopters. To address this, we propose an optimized VMD and global spatiotemporal model for diagnosing inter-rotor rub-impact faults in a dual-rotor turboshaft engine. This method combines signal processing with deep learning techniques to effectively uses multi-sensor data to accurately identify inter-rotor rub-impact faults. Specifically, the raw signals are first decomposed using a VMD optimized by the TTAO algorithm, followed by signal reconstruction based on selected criteria. The reconstructed signals are then input into the GSN in the form of multi-channel feature maps for fault diagnosis. Finally, the effectiveness and superiority of the proposed method are validated and compared using fault data collected from a dual-rotor rub-impact test device. Experimental results demonstrate that the proposed method can distinguish inter-rotor rub-impact among various rub-impact faults, achieving an average diagnostic accuracy of 99.61% and excellent classification performance.