Research On Analysis Method Of Current And Vibration Signal Of Rubbing Fault In Rotor System

As a core component of large rotating machinery,rotor system is prone to various faults under high-speed and heavy-duty operating conditions.Rotor rubbing fault is one of the typical faults.When it appears,it will reduce operation efficiency of the mechanical to make the used energy loss.More seriously it will bring about the heavy safety accident and economic loss.Therefore,domestic and foreign experts have done a lot of systematic and theoretical research on the identification and diagnosis of rotor system rubbing fault.Usually the starting point is to extract the fault characteristics of the rotor vibration signal.However,the sensor that collects the vibration signal is inconvenient to install due to environmental conditions in which the device is located,and the interference of mechanical resonance and noise on the useful signal will have an impact on the diagnostic effect.At this moment,a flexible and efficient signal acquisition method is urgently needed.The motor current signal analysis(MCSA)is based on the stator current of motor as the starting point for signal analysis to diagnose the fault of the motor body and the dragged load,which can make up for the shortcomings of the signal analysis method based on direct contact to measure signals.In this research,with reference to the dynamics related theory of rotor system,the motor current signal is mainly used to explore diagnosis methods of rotor system rubbing fault.According to the mechanics simplified model of rubbing rotor,the differential equations of bending and torsion coupling motion of the rotor isderived.Combined with the theory of electromagnetics and motor structure,the electromagnetic torque is used as the link to connect the load torque.In the MATLAB/Simulink software environment,the electromechanical coupling model of rubbing rotor system is established.The rotational speed,eccentricity and stator stiffness are controlled separately to simulate the situation.The frequency spectrum of the corresponding current signal is obtained by Fourier transform and the frequency modulation phenomenon is analyzed.The modulation law of the rubbing in the current spectrum is found.The rotor test rig is set up for field test to complete the acquisition of motor current signal when rotor is rubbed,and the simulation results are verified.The coupling characteristics between rotor system with rubbing fault and the motor stator current are researched.In order to study the special fault of rotor single-point rubbing,the stator current signal and vibration signal are combined to diagnose fault from the perspective of simulation and test.Firstly,how the stator current signal reflects the dynamic and static rubbing characteristics is deduced.At the same time,the weak fault characteristics of the vibration signal are extracted by the variational mode decomposition(VMD).According to the current spectrum,the vibration spectrum is combined to predict the rubbing state of rotor.Before studying rubbing fault,it is necessary to classify the common faults of rotor system.The current signals under different rotor operating states are decomposed by 4 layers wavelet packet decomposition for extracting the energy features as the training and recognition feature vectors of neural network,then identify the operating states by RBF,ELM and PSO-ELM respectively.The average correct rate of identification of PSO-ELM network is 96.67% and the whole test samples of rubbing fault are classified accurately,so its performance is obviously better than the other two networks.Therefore,PSO-ELM is more suitable for identifying the rubbing fault mode of rotor system.To investigate the degrees of rotor rubbing is a further study of the rubbing fault diagnosis of rotor system.The EEMD decomposition of the current signal under different rubbing conditions is performed,and the energy specific gravity of IMF component containing the main fault information is used as the feature vector.GRNN and PNN neural networks are used separately for identification and classification.The correct rate of recognition of GRNN is 93.33%,and the recognition accuracy of PNN is 96.67%.Therefore,PNN network can be selected as an effective mean to identify the degrees of rubbing.