Neural Networks Based Diagnostics for Mistuned Bladed Disk

Abstract

Neural networks based diagnostics is studied in this paper for the fault/alert assessment of the mistuned bladed disk. By randomly varying the blades frequencies (± 2% of the nominal frequencies), we experimentally collected 1000 sets of data. Each set of data is composed of blade frequencies, maximal vibrating magnitude for each blade with the corresponding frequency. The goal is to investigate the mapping relationship from the blade property (blade frequency) to maximal vibrating displacement and the corresponding frequency. Assuming the system is forced by a traveling excitation, the vibration/frequency signature is derived. The challenging part stems from mistuned bladed disk system, where every blade performs differently for the same input excitement. The neural networks based diagnostics system demonstrate its capability to identify the mapping relation for this complex system. The simulation is based on a finite element model of bladed disk system and the results exhibit the effectiveness of the proposed algorithm.