Optimization of dense modal identification for aerospace components based on VMD

This paper focuses on aerospace engine parts, specifically addressing the challenge of recognizing structure-intensive modal frequencies, and proposes a novel method for extracting these frequencies. The research primarily investigates a signalf2/f1> 0.5 delay processing approach using a BP neural network. To mitigate the endpoint effect associated with the Variational Mode Decomposition (VMD) method, two signal extension strategies were evaluated: extreme point extension and BP neural network prediction extension. Simulation tests on various signal types demonstrated that the BP neural network prediction extension method provides the most effective suppression of the endpoint effect. Consequently, this method was applied to pre-process the delay of simulated signals, which were subsequently decomposed using the VMD algorithm. Experimental validation confirmed that this combined BP-VMD approach effectively extracts frequency information from signals, thereby enhancing the decomposition accuracy of the VMD algorithm.