Research On Incipient Fault Diagnosis Methods For Rotating Machinery Based On VMD And Optimized MSVM

Rotating machinery, which serves as essential engineering equipment in modern industrial production, has been widely used in chemical industry, petroleum, metallurgy, electric power and many other important fields which closely related to national economy and people’s livelihood. Once the rotating equipment breaks down, the whole system will paralyze, causing huge economic loss and irretrievable casualties, which has become a serious impediment to steady development of national economy. Research has shown that rotating machinery incipient fault has a longer incubation period. So it is significant for preventing the economic and property loss caused by rotating machinery breakdown to predict the exact time, location and fault category when the fault is in the early embryonic stage or it is a small failure and to use this to guide equipment maintenance, control rotating machinery equipment fault and ensure rotating machinery operation safely and reliably.Vibration signals of early faults of rotating machinery are susceptible to strong background noises, complex propagation, acquisition equipment, signal transmission channel and signal degradation, which often makes the fault information contained in the vibration signal weak and submerged in strong noise; Rotating machineries mostly have complex structures, unstable operation conditions, coupled vibration of multiple parts and large vibration noises, which make the collected vibration signals have strong characteristics of instability and nonlinearity, fault symptom not obvious, and fault characteristic too weak to extract; Early fault samples of rotating machinery are rare as there is no long-term and systematic collection. Besides, there is no explicit relation between the fault characters and fault categories, so fault identification is difficult.Aiming at these problems, such as the enhance of rotating machinery fault signal, nonlinear feature extraction, fault diagnosis under small fault samples, this thesis in-depth studies the incipient fault diagnosis method based on variational mode decomposition and optimized multi-kernel support vector machine. The concrete research content is as follows:(1) Aiming at the problem of strong background noise and weak fault information in vibration signal from rotating machinery, an adaptive maximum correlated Kurtosis deconvolution method is put forward to enhance the weak fault signatures. By regarding related kurtosis as evaluation indexes, and giving full consideration to the characteristics of impact ingredients, an iterative algorithm is used to implement convolution operation process. Shannon entropy served as the objective function, use grid search method to automatically search the optimal filter and order cycle, which makes the maximum correlated Kurtosis deconvolution method more effective and adaptive, so the weak fault signature submerged by noise can be detected effectively.(2) Aimed at the problem that the characteristics of early fault of rotating machinery is faint and nonlinear fault feature is hard to extract, the feature extraction method of multiband and multi-scale sample entropy feature based on adaptive Variational Mode Decomposition is proposed to show early weak fault characteristics of rotating machinery. To characterize the condition state and achieve the recognition of early faults of rotating machinery, multi-scale sample entropy of different frequency bands modal is extracted and made up of the sensitive feature vector set.(3) Aiming at the selection of some key parameters in variational mode decomposition, an adaptive VMD decomposition method is proposed. The method that can ensure the decomposition accuracy and guide the determination of the optimal value of K by the correlation between the original signal and each modal decomposition decomposed by VMD is put forward of; the smaller the balance constraint parameters in VMD decomposition, the more wide the bandwidth of various modal components, and the easier to overlap center frequency and alias modals; it’s vice versa, but this increases the difficulty of calculation. Through the simulation analysis, the idea that general equilibrium constraints are desirable for sampling frequency in practice is put forward; Analyze some performance of adaptive variational mode decomposition: orthogonal performance analysis, energy saving degree analysis, the equivalent filter attribute analysis and abnormal information interference. The simulation results indicate that adaptive variational mode decomposition is superior to the EMD, EEMD and LMD method in orthogonal performance, energy saving performance; Using fractional Gaussian noise through numerical simulation experiment to analyze EMD, LMD and AVMD ’s equivalent filter attribute, compared with EMD and LMD, AVMD can be more closer to the packet decomposition, and it is a kind of adaptive decomposition method and can provide much higher time-frequency resolution than EMD and LMD; especially for the signal disturbed by abnormal information, AVMD still has good effect and can decompose signal adaptively.(4) Aiming at the lack of early fault samples of rotating machinery, a fault diagnosis method of rotating machinery based on optimized multi-kernel support vector machine(MSVM) is put forward. Through introducing the weighting factor which combines different kernel functions, construct multiple kernel based on the global kernel function and local kernel function, realizing the rapid mapping from the input feature vector to kernel space, so that the algorithm has better generalizability and stronger model-explain ability; use immune genetic algorithm to obtainoptimal parameters of MSVM, which can overcome the uncertainty of parameter selection of MSVM, and then improve the stability and generalizability of MSVM in small-sample-size fault diagnosis of rotating machinery.At the end of the thesis, the work of this paper is summarized, and expectation of feature research is presented.