Fault Diagnosis Method For Centrifugal Pump Based On 2D Hidden Markov Model

Fault diagnosis is a new multi-subject-crossed technique. It has been developed rapidly in recent years and brought huge benefit. Fault of centrifugal pump always brings the equipment vibration, abundant operation condition information is contained in the vibration signal and it's easy for acquisition and diagnosis. Fault vibration signal of centrifugal pump is non-stationary signal, so the processing method which is suitable for non-stationary signal has to be chosen. The time-frequency analysis methods such as windowed Fourier transform, and wavelet transform have their own limitations.Hidden Markov Model (HMM) overcomes the traditional diagnosis method's defect of just staying on static observation. It's very suitable for describing short-time-stationary, non-stationary signal and more efficiency than other signal processing methods in classifying fault signals. As generalization model of HMM, Two-dimensional Hidden Markov Model (2D-HMM) is composed of external HMM and internal HMM which is based on each external state. So it not only has HMM's advantage, but also can comprehensively describe the signal from two sides. It's extremely suitable for processing the signals which have strong nonstationarity and weak reproducibility produced by centrifugal pump.The research theme of this dissertation is the 2D-HMM centrifugal pump fault diagnosis, combined with research of science and technology fund project offered by education department of Jilin province (No.2007047).The basic algorithm and implementing technology for vibration signal filtering, and fault classification basing on 2D-HMM are proposed by uniting theoretical analysis and experiment test and proved to be effective through 2BA-6A centrifugal pump experimental system.First, research background and significance of subject selection is introduced. Not only parameters and performance of the 2BA-6A centrifugal pump experimental system but also experimental setting and method is included in first part of the thesis. Second, starting with the HMM's theory and algorithm, the model types,mainly parameters, topology structure are Subsequently introduced. And primary arithmetic of the 2D-HMM are discussed in detail. The problem in the practical application of the model and the upswing measure for these problems are also set forwarded. Finally the comparison research of HMM and 2D-HMM from the model structure, parameters description and arithmetic complexity is also presented.Then, fault diagnosis method based on 2D-HMM for centrifugal pump is proposed. Its feasibility, characteristic, basic steps and realization method of diagnosis are also introduced. On the basis of summarizing common faults of centrifugal pump and their vibration characteristics, parameter setting such as model selection about 2D-HMM in centrifugal pump fault diagnosis is provided.Finally, the vibration signals are analyzed by the introduced signal feature extracted methods through 2BA-6A centrifugal pump experimental system. And the fault diagnosis method based on 2D-HMM for centrifugal pump is proved to be effective.