Research On Multi-Source Fusion Technology In Load Recognition Of Rotor System

This project is proposed aiming at the features of mechanical equipment rotor system such as complex structure, bearing large load, driving factors of high power motor, vibration problems, tedious forces of affecting load and fault,load identification of the system is carried out to ensure the reliable and safe operation of the mechanical equipment.Rotating machinery is widely used. Monitoring of the rotor system which is the core component of the rotating machinery is particularly important. In the rotor system driven by electric motor, the load characteristics can be responded to the response signals when the system is subjected to the load of torque.Vibration signal is easy to collect, which can well reflect the function time and intensity information of the load. Since the connection of the electromagnetic torque, the stator current signal of the motor also can be used as an important information for the identification of the load. Thus, these two kinds of signals can be used as the information sources of the load identification.In this paper, four different types of load excitation tests are carried out, which are static, impact, linear and sinusoidal, and the vibration signals of the rotor system and the current signal of the driving motor are acquired. Based on the characteristic of vibration signals, an improved threshold processing function is put forward to reduce the vibration signal. Based on the characteristic of the spectrum of motor current signal, the singular value decomposition is put forward to weaken the power frequency component and highlight the characteristic component. After preprocessing the two kinds of signals, the recognition of the types of load and the quantitative identification of the load are carried out.Under different types of load,correlation dimension of fractal theory is used to extract quantitative characteristics of time domain vibration signals to reveal the mapping relationship between the load type and the vibration characteristic and the rotor system load type identification method is proposed based on the vibration signal. Aiming at current signals, wavelet packet decomposition is proposed to extract the energy characteristics distribution and get feature vector to reveal its mapping relationship with the load type by using neural network. For the limitation of single type signal, multi-source fusion technology based on Bayesian estimation to establish the mapping relationship between the load type and the two response signal characteristics, which provides a new method for the identification of load type.Aiming at the quantitative identification of load, the support vector machine machine learning method is introduced. Based on a single vibration signal and current signal, the steady loads are used as the training sample to set the coupling relationship between the amplitude of the response signal and the load value with regression support vector machine.For the test sample, the amplitude of signal input of the selected feature points can forecast the numerical load to fit discrete load formulas or graphs. For the contingency of using a single signal in the quantitative identification of load,multi-fusion technology combined two kinds of signals is put forward in the quantitative identification of load and proved its reliability.In this paper, the recognition of the load type fill the gaps in the field of load identification. In addition, the identification methods proposed in this paper do not need to consider the parameters of the system itself to avoid the cumbersome system modeling process, which can be used to solve other similar problems.