Research On Feature Extraction And Automatic Recognition Of Axis Orbit For Rotor System

With the rapid development of modern industrial technology,rotating machines are developing towards large-scale,high speed,light and intelligent mode.And rotating machinery is more and more complex.Rotor system is the key part of rotating machinery.Due to various random factors,it is an inevitable failure in long time running.The chain reaction caused by the shafts often results in huge economic losses and disastrous consequences.Therefore,it is of great significance for fault monitoring and fault diagnosis of rotor system.In recent years,because of the rapid development of Internet,the intelligent fault diagnosis of rotor system is also faced with higher requirements.Feature extraction and automatic recognition of axis orbit is a commonly used method in rotating machinery fault diagnosis.The axis orbit is composed of two vibration signals which are perpendicular to each other.Its shapes contain a lot of information,which can reflect the operation status of the equipmentvisually and intuitively.The traditional methods mostly use time-frequency processing and signal processing technology.The efficiency of fault diagnosis is low,and the rate of false positives is high.In this paper,on the basis of traditional signal processing,the technology of image processing,invariant moment feature extraction and automatic recognition is introduced,and the better experimental results are obtained.In addition,a detailed test analysis is made for the difference between the axis orbits of two rotors for double-span rotor system.It will undoubtedly provide new research ideas and data support for the research of rotating machinery fault diagnosis technology.The main work of this paper is as follows:(1)For the research of feature extraction and automatic recognition of axis orbit for rotor system,the paper establishes the foundation for shaft orbit automatic identification research,with signal refining,digital image processing,invariant feature extraction,manifold learning theory,fractal theory,and image pattern recognition.(2)In this paper,several edge detection algorithms and image feature extraction methods are compared,and a fault diagnosis method is proposed,which combines the invariant moments and the fractal box dimension as the combination feature vector,and use the neural network to learn and recognize the faults.This method has a good recognition effect for common fault types.And due to the introduction of fractal box dimension as a part of the feature vectors,the method is more sensitive to oil whirl and oil whip.(3)The original signal of axis orbit is often messy,which affects the recognition accuracy and recognition speed.In order to improve the recognition effect,an automatic recognition method of axis orbit is proposed with the combination of morphological image processing and Hu invariant moment feature vector.Firstly,the axial displacement signal is processed by a proper filter to eliminate the high frequency noise.The mathematical morphology is used to restore the clean axis orbit,including the dilation,erosion and skeleton operation.Hu invariant moments of the skeleton axis orbits are calculated.Finally,BP neural network is trained by using Hu invariant moments as the feature vectors to recognize the faults of the rotor system.The experimental results show that the skeleton axis orbit is purer,and the difference between the actual axis orbit and the sample is reduced.The calculation speed is improved obviously.It is a reliable and effective method for the recognition of the axis orbit.(4)In this paper,a fault diagnosis method based on three-dimensional(3D)axis orbit and LTSA manifold learning is proposed for the problem of fault feature loss of two-dimensional(2D)axis orbit.The 3D axis orbit is transformed into a 2D manifold diagram.Compared to the 3D axis orbit,the 2D manifold diagram is more convenient for analysis and recognition,it has a simple and intuitive feature differentiation,and the spatial topological relation of the data points of the 3D axis orbit is retained.(5)In this paper,a detailed test analysis is made for the difference of theaxis orbits at different positions of the double-span rotor system.The experimental data include four states of normal,misalignment,oil whirl and oil whip.A detailed analysis is included in the failure mechanism,structural features,fault location and selection of measurement points,which provides the experimental basis and theoretical guidance for the subsequent fault diagnosis of the double-span rotor system.