Research On Instantaneous Feature Extraction And Condition Recognition Of Shaft System Vibration Of Hydropower Unit

Hydropower units are important power production equipment,and their safety and stable operation are directly related to the safety and stability of the power station and the entire power system.It has become the consensus of academia and the industry to configure condition monitoring and fault diagnosis systems for large hydropower units to support scientific decision-making on unit operation and maintenance.Among them,vibration-based condition monitoring and fault diagnosis are the most effective and the most widely used.A hydropower unit is a large-scale system with strong coupling and complex nonlinearity.Its vibration is affected by various factors such as mechanical,hydraulic,electromagnetic and other factors.In addition,since the hydropower unit undertakes the task of peak shaving and frequency regulation,it starts and stops frequently during operation.The vibration signal exhibits extremely strong nonlinear and non-stationary characteristics.Its vibration signal shows strong nonlinear and non-stationary characteristics.Non-stationary vibration signal processing and data-based fault diagnosis under the background of strong noise are the key and common problems that need to be solved to realize unit condition monitoring and fault diagnosis.For this,traditional methods have been unable to meet the application requirements under the current background.Therefore,researching new theory and methods of non-stationary vibration signal processing of rotating machinery and proposing an effective fault diagnosis framework has important application prospects for enriching the monitoring index system of hydropower units,improving the accuracy of fault detection and diagnosis,and ensuring the safety and stability of the unit.This paper focuses on the instantaneous feature extraction and intelligent identification methods of the vibration signal of the hydropower unit shaft system,Taking full spectrum,holospectrum,complex signal decomposition,deep transfer learning,etc.as the theoretical basis and technical means,on the basis of full investigation and analysis of relevant theories and technical research status at home and abroad,the concept of instantaneous orbit is proposed.The method for extracting the features of the rotor vibration instantaneous axis orbit for different application scenarios is derived.The paper builds a fault diagnosis framework based on rotor vibration signal fusion features and deep transfer learning.The main research content and innovative results of the paper are as follows:(1)Aiming at the problem that the traditional method is difficult to accurately describe the non-stationary vibration behavior of rotating machinery,based on the full spectrum theory,the concept of instantaneous axis orbit is proposed and defined,and the calculation method of instantaneous axis orbit features is derived.Finally,a high-resolution time-frequency representation method of axis orbit is proposed.The features used in the description of the axis orbit,such as the amplitude of the forward/backward components of the axis orbit,the inclination angle of the ellipse orbit,and the shape and directivity index,are extended to transients.With the help of the Vold-Kalman filter order tracking analysis method,the original rotor vibration signals are separated according to the target order,and the instantaneous axis orbit feature extraction is achieved through the fusion of the time-domain waveform amplitude and phase relationship in the orthogonal direction.Based on the extracted instantaneous axis orbit features and inspired by time-frequency analysis technology,the instantaneous axis orbit features are projected onto the time-frequency surface to construct a purified orbit timefrequency representation,which realizes non-stationary changes of the axis orbit Visual representation of the process.(2)In order to realize the feature extraction of rotor non-stationary vibration with no speed signal,a new adaptive chirp mode complex signal decomposition method and a high-resolution complex signal time-frequency representation method are proposed.Traditional complex signal decomposition methods are limited in the vibration signal analysis of rotating machinery due to the phase shift of the results.This paper introduces a method based on Hilbert transform to separate the complex signal into positive and negative components,A new complex signal decomposition algorithm is developed to realize the adaptive decomposition of chirp mode complex signals.Furthermore,based on the decomposition results and WVD,a construction method for the time-frequency representation of complex signals to reduce cross-term interference is proposed,which realizes the high-resolution time-frequency representation of non-stationary complex signals.Simulation and real signals analysis show that the proposed method can be applied to the analysis of non-stationary vibration signals of hydropower units.(3)In order to visually analyze the non-stationary vibration process of the multi-supported rotor system,this paper proposes a multivariate complex signal variational modal decomposition algorithm.Based on the decomposition results of the algorithm,an instantaneous feature extraction method for the axis orbit of the multi-supported rotor system is derived.The three-dimensional instantaneous orbit diagram of rotor vibration is further proposed to accurately describe the non-stationary vibration process of the rotor with multiple supporting surfaces.Inspired by the complex signal decomposition and multivariate signal decomposition algorithms,a new multivariate complex signal variational modal decomposition algorithm is proposed.Inheriting the excellent performance of multi-component center frequency alignment of the multivariate variational modal decomposition algorithm,the multivariate complex signal components can be separated synchronously.On the basis of the single support surface,a method for extracting the instantaneous features of the vibration signal of the rotor with multiple support surfaces is developed.Based on the instantaneous features,inspired by the three-dimensional holospectrum,a three-dimensional instantaneous orbit diagram for describing the overall view of the rotor vibration is further proposed and a drawing method is given.(4)In order to realize the automatic recognition of the vibration state of rotating machinery,a vibration state recognition method based on the fusion characteristics of vibration signals of multiple supporting surfaces and deep transfer learning is proposed.In order to adapt to the requirement of two-dimensional image data for the input of deep transfer learning model based on convolutional neural network,a method of fusing the instantaneous orbit features of rotor vibration signals with multiple supporting surfaces to construct a fused feature map is proposed.Driven by the fusion feature map,based on the theory of deep transfer learning,a pre-training model fully trained on a large general data set is introduced,and a rotor vibration fault classification framework for rotating machinery is designed.Furthermore,a fast identification method of hydropower unit operating conditions based on vibration signal and deep transfer learning framework is proposed.This method does not require operating conditions parameters.(5)In order to meet the needs of big data applications in hydropower stations,a data management method for vibration monitoring of hydropower units is proposed.Combining with the actual needs of hydropower unit vibration status monitoring,a multi-dimensional drive vibration status monitoring data collection and management method has been developed,and a hydropower unit vibration monitoring system oriented to big data applications has been designed and developed.Taking the real hydropower unit's vibration waveform data during the startup and shutdown process as examples,the effectiveness of the method proposed in this paper is verified.