Flexible rotor unbalance fault location method based on transfer learning from simulation to experiment data

Abstract In the study of the high-speed dynamic balance of flexible rotors, rotor unbalance positioning is a challenging topic. Particularly for slender rotors, the axial position of the unbalance has an important influence on the high-speed dynamic balance. The unbalance at different axial positions is not the same or even opposite in different rotor mode vibration behaviors. If the unbalance position of a rotor can be identified, the actual unbalance of the rotor can be reduced from the root. This balance method has the same effect in each vibration mode of the rotor; hence, low-speed dynamic balance can be realized to replace high-speed dynamic balance, considerably saving on costs. Deep learning based on few labeled samples can achieve good results for the identification of unbalanced positions; however, there are infinite potential positions of unbalance in the actual rotor. It is difficult to collect sufficient labeled samples to train a reliable intelligent diagnostic model. Fortunately, a large number of rotor vibration datasets labeled with different unbalance positions are available using the rotor dynamic model, and the unbalance position data calculated using the dynamic model contain diagnostic knowledge related to the rotor unbalance position data measured in the rig. Hence, inspired by transfer learning, this study proposed a transfer learning method using dynamic model simulation and experiment data for flexible rotor unbalance fault location. Cross-domain deep transfer recognition of rotor unbalance position was realized.