Study On Fault Diagnosis Methods For Reciprocating Compressors By Using Feature Analysis Of Vibration Signals

For all petrochemical enterprises, it is the most considerable work to maintain the stable operation of reciprocating compressors which play a very important role in petrochemical production. Therefore, condition monitoring and fault diagnosis for reciprocating compressors are of great practical significance and economic value. Based on the engineering research project for ethylene hypercompressor, National Natural Science Foundation of China (Grant No.50805014) and the Key Project of Chinese Ministry of Education (Grant No.109047), some effective works for fault diagnosis of reciprocating compressor are carried out by using feature analysis of vibration signals. The main works of this dissertation are listed as follows:1. The working mechanism of reciprocating compressors is illustrated in detail. Failure modes and mechanism of the key components are summarized. Under the analysis of the characteristics of the response signals produced by certain excitation sources, vibration signal models of the key components are established. Impact response and periodicity are indicated as two key features of reciprocating compressor signal, and the application of these key features in reciprocating compressor fault diagnosis is preliminarily discussed.2. Frequency-domain energy distribution characteristics of reciprocating compressor vibration signals are discussed. Based on frequency-band energy variation, a criterion on the decision of response signal characteristic bands is proposed. The features of these characteristic bands can be extracted by comb filtering and envelope analysis. Application in reciprocating compressor condition monitoring verifies the validity of these methods.3. EMD denoise criterions based on the characteristics of reciprocating compressor vibration signals are proposed. The envelope peak factors of the denoised signal are obtained by using improved Hilbert transform method. A new optimal method based on improved NS-DPSO is applied to select the optimal feature subset, which is employed to construct eigenvectors to identify different working conditions of reciprocating compressor by using optimal multi-class SVMs. The engineering application in working condition classification of reciprocating compressor valves verifies the effectiveness of these methods.4. The properties of cyclic auto-correlation function are studied. Under the discussion of second order cyclostationary method applied in condition monitoring and fault diagnosis of reciprocating compressors, cyclic auto-correlation function is indicated to be one sort of special time-frequency distribution method. EMD denoise and reconstruction are used to reduce the cross-terms of cyclic auto-correlation function. The conception of cyclic information entropy is proposed and a minimum distance classifier based on cyclic entropy distance is used to achieve the fast classification of reciprocating compressor working conditions.5. The algorithm of Local Wave time-frequency coherence is proposed, which is applied in the fault diagnosis of reciprocating compressor cylinder. Firstly, double channel signals of cylinder and valve are synchronously sampled. Secondly, cyclic cross-correlation function is used to eliminate the time delay between the two signals. Thirdly, Local Wave time-frequency coherence method is utilized to remove the interference components, which produced by valve vibration, from cylinder signal. Finally, more concise and precise working condition information for cylinder and inner components can be obtained.6. An on-line condition monitoring and fault diagnosis system for reciprocating compressors is developed. Overall architecture of the system is designed and primary monitoring parameters are determined. Hardware structure and the functions of software modules are introduced in detail. At last, a diagnosis example is used to testify the effectiveness of the system.