Blind Signal Separation Algorithm And Its Application Method Research In Rotor Fault Diagnosis

Fault feature extraction and pattern recognition is the most crucial problem for the reliability and accuracy in the fault diagnosis of rotating machineries. This dissertation addresses the fault diagnosis of rotating machineries, with the purpose of enriching machine fault diagnostics and requirements of engineering application of fault diagnosis of the key equipment in mechanical engineering, by means of constrained blind source separation methods. It is necessary and important to diagnose machine fault accurately and effectively, so as to provide maintenance strategy and deduce economic losses. It is not only of great theoretical significance, but also of great engineering value. This dissertation explores the applications of the theories with second order blind separation, median filtering, adaptive particle swarm optimization, denoising source separation in the feature extraction, representation and vibration signal for the rotary machinery. The main research works can be described as follows:(1) Noise reduction usually is conducted before analysis of mechanical fault feature, which could damage effective signals.This article proposes an algorithm of blind source separation based on the second-order statictics.The method focuses on noise separation rather than noise removal.So there are no harms to effective signals. This idea might provide a new way for noise reduction. The algorithm of blind source separation based on the second-order statistics blind identification is applied to mechanical vibration data.The results show that the algorithm is effect,noises are separated and re-moved, and accurate the rotor fault feature are picked up.(2) The performance of existing nonlinear mechanical failure signal separation methods is affected by the non-linear contrast function that is selected according to the distribution of original signals. To solve this problem, a blind source separation algorithm based on adaptive particle swarm optimization is proposed, which takes the negentropy of mixtures as a contrast function. The inertia weight factor depends on the negentropy, which can improve the contradiction between the convergence speed and the performance of separated signals. The simulation results was verified the effectiveness of the proposed method. Finally, Some mixed rotor vibration signals were separated successfully using the proposed method.(3) Signal processing methods are commonly used to analyze the structure of signals according to the criteria of spectral distribution. However, the causal relationship between components and sources are not revealed. Under the condition that only observed signals are known, the mixed signals can be separated into several components by denoising source separation (DSS) method according to statistical feature. The sources of observed signals are revealed by these independent components, thus it provides a direct reference to condition monitoring and active control of vibration and noise. The basic theory of DSS and denoising functions based on different criterion are studied, and the separation performance of four types of denoising function such as energy function, slope function, kurtosis function and tangent function are quantitatively compared by means of simulation of typical mechanical signals. The results show that the algorithm based on tangent function is more suitable for extracting nonlinear coupling information of mechanical equipment. The DSS method based on tangent function is used to extract running information feature of rotor, and the quantitative analysis results show that some mixed rotor vibration signals were separated successfully using the proposed method.(4) When the rotary machinery is running, the vibration signals measured with sensors are mixed with all vibration sources and contain very strong noises. It’s difficult to separate mixed signals with conventional methods of signal processing, so there are difficulties in machine health monitoring and fault diagnosis. The principle and method of blind source separation were introduced here, and it was pointed out that the blind source separation algorithm was invalid in strong pulse noise environment. For the vibration signals in strong pulse noise environment, they were de-noised with the median filter method firstly, and then the de-noised signal was separated with the blind source separation algorithm. The simulation results was verified the effectiveness of the proposed method. Finally, Some mixed rotor vibration signals were using the proposed method. Thus, a new separation approach for vibration signals in strong pulse noise environment was provided.