Abstract:The field balancing of flexible rotors is one of the key techniques to reduce vibration of large rotating machinery. Although in recent decades the balancing theory has been thoroughly studied and various balancing techniques have been well developed, the present balancing methods are still remain for further improvements in accuracy and efficiency. Firstly, most balancing methods need large numbers of trial runs to obtain the vibration responses of trial weights in different correcting planes. Secondly, the vibration response in each measured section is always taken from a single sensor, and thus are lack of comprehensive vibration information of rotor. In fact, the movement of rotor is a complex spatial motion, which can't be objectively and reliably described just with a single sensor in each beating section. In order to overcome above shortcomings of traditional balancing methods, this paper presents a new field balancing method for flexible rotors, which is based on adaptive neuro-fuzzy inference system (ANFIS). The new method successfully applies the information fusion, ANFIS and computer simulation together. It integrates and fully utilizes the information supplied from all proximity sensors by holospectrum for enhancing the balancing efficiency and accuracy. A fuzzy model is established to simulate the mapping relationship between vibration responses and balancing weights by using the ANFIS. The inputs into ANFIS are the amplitudes and phases of integrated vibration responses, while the outputs are the mass and azimuth of balancing weights. A fuzzy set with three membership functions (MFs) is used to describe the magnitude of vibration amplitudes or of balancing weights. Another fuzzy set with five MFs is used to describe the quadrant of vibration phases or of balancing weights. Based on the historical balancing data, a combination of least-square and back-propagation gradient descent methods is then used for training ANFIS membership function and node-parameters to model input (vibration response)/output (balancing weight) data. The simulation study shows that the ANFIS can obtain satisfactory balancing result after a single trial run. At the same time, with the help of computer simulation, different correction schemes can be compared and rapidly simulated to direct balancing operation. Finally, the effectiveness of the new method was validated by the experiments on balancing rig and in the field balancing practice of several 300MW turbo-generator units.

Application of adaptive neuro-fuzzy inference system in field balancing