Rotor Vibration Fault Feature Extraction And Diagnosis Based On EMD And Chaotic Analysis

With the progress of science and technology, rotating machinery are constantlymoving towards large-scale, complex, high-speed, continuous and automated direction.Though these developments improve production efficiency, they put forward higherrequirements for safe operation of the equipment.Fault diagnosis for rotating equipment,especially for rotor vibration, is an important measure to ensure the safe and reliableoperation of the equipment. This paper studied on the method of rotor vibration faultfeature extraction and diagnosis based on EMD and chaotic analysis, established athree-band hybrid feature extraction model, and did deeply researches in some of the keyissues.For the common four kinds of nonlinear vibration fault, including rotor massimbalance, rotor misalignment, oil whirl and rubbing, this paper analyzed differentvibration fault theoretically based on the fault model, concluded the characteristics ofeach fault in the time domain and frequency domain, and made clear the criteria ofvibration fault, therefore provided a reference for subsequent fault diagnosis.Then, basedon the EMD signal decomposition method, this paper reconstructed three-band signalcomponents to analyze the feasibility of extracting high, medium, low frequency relativeenergy and energy entropy as fault feature, and draw the defects for misalignment andrubbing of the feature extraction scheme. By the chaotic numerical characteristicsanalysis of high-frequency components of fault signal, it analyzed the feasibility oftaking the correlation dimension and the largest Lyapunov exponent as thehigh-frequency component characteristics to distinguish between misalignment andrubbing. Subsequently, it formed the eigenvectors by collecting rotor signal samples andextracting their tri-band energy, energy entropy and chaos numerical characteristics ofhigh-frequency components. Finally, it constructed two-classifier and multi-classifier byusing support vector machine and decision directed acyclic graph scheme respectively.After training and diagnosing the collected samples and analyzing of experimental results,the paper obtained the superiority of the program.