Study On The Vibration Fault Diagnosis Method Of Centrifugal Pump And System Implementation

In recent years, fault diagnosis as a new technology which crosses various disciplines, had developed rapidly and produced huge economic benefits. Both normal operations centrifugal pump and failure centrifugal pump will make a device vibrate. The vibration signal contains a wealth of information about the pump running, and easy to collect, can be applied to monitor and diagnosis the status of centrifugal pump operation. Because the signal of failure centrifugal pumps are a non-stationary signal, so it is necessary to select the appropriate signal processing method which is suitable for non-stationary signal.As the impact of environment and conditions, the vibration signal collected from the running centrifugal pump must contain noise. How to reduce the noise signal? This paper presented the signal denoising method of second-generation wavelet combining new improved threshold function.This method decomposes the failure signals by second generation wavelet decomposition, and does threshold processing on Wavelet coefficients, which decomposed signal by new improved threshold function. While introducing noise reduction evaluation criteria, which based on class separability measure for noise reduction, this paper evaluated the noise reduction effect of complex vibration signals. The denoising result of the simulated and measured vibration signal show that this method combines the advantages of second-generation wavelet and improved threshold function, to better eliminate noise.The denoised vibration signal was analyzed by HHT transform, the computing of complexity, continuous wavelet transform, lifting wavelet packet and chaotic recursion theory, etc. By extracting effective parameters, forming eigenvectors, imputing the classification model the centrifugal pump fault type can be diagnosed.Classification model included RBF neural network, Improved Elman neural network, least squares support vector machines and correlation support vector machines, etc. From the result of learning, training and recognition. It can be found that the model of enhancing the characteristics of wavelet packet and correlation support vector machines has the best recognition effect, but compared with the other compages, there is a little difference in recognition rate. Finally, the Hidden Markov Model (HMM) was highlighted. It overcame the shortcomings of traditional method for just stopping at the static observation; it is very suitable for describing the short-term stable non-stationary signals, the proven results of the failure signal classification are superior to other signal processing methods.2D-HMM as a generalization model formed by external HMM and Internal HMM this is based on each state of external HMM. So it has the advantage of HMM, and describes the signal from both time domain and frequency domain comprehensively. It is very suitable for processing the pump signal which is very Non-stationary and is difficult to repeat reproduction occurred during running. This paper verified the effectiveness of the fault diagnosis method based on 2D-HMM for centrifugal pump, by applying several signal feature extraction methods combining 2D-HMM model.All the above has been concluded in the software made by Visual Basic6.0, under the circumstance of Windows 2000 and has been experienced systematically. The result indicates that the designation and both hardware and software structure of this system are logical and effective. It should be monitored in the centrifugal pump to achieve the purpose of designation.