Research On Condition Monitoring And Trend Prediction Technique Of Centrifugal Compressor K5403

Centrifugal compressor K5403 was the key equipment of Baling Petrochemical Co., Ltd.. Once that the equipment failured, not only seriously affected the production, brought major economic losses to businesses, but also caused fatal accidents and other casualties. Therefore, to achieve predictive maintenance to equipments, this thesis took centrifugal compressor K5403 as the research object, and carried out research on condition monitoring and trend prediction techniques.The main contributions of this thesis are summarized as follows:1. The purpose, significance and course of development and trends for condition monitoring and trend prediction technique were presented. The importance of using condition monitoring and trend prediction technique in centrifugal compressor K5403 according to the maintenance status and user needs was stated.2. According to the structural characteristics of centrifugal compressor K5403, the fault mechanisms of the common forms of the fault types ( for example: imbalance, misalignment, rub-impact movement, association loosening faults of rotor support system, a variety of faults of rolling bearing, surge, and so on.) were analyzed respectively, and the appropriate fault features were given. At last, a number of commonly used fault diagnosis method based on analysis of the vibration signal were introduced.3. Combined with the actual needs of enterprise, the condition monitoring and trend prediction system of centrifugal compressor K5403 was developed. The key technique such as Object-oriented technique, virtual instrument technique and configuration technique were introduced. System requirement analysis and object-oriented design based on UML were researched.4. Based on the non-linear and non-stationary characteristics in running status of centrifugal compressor K5403 and the advantages of neural network, the trend prediction method of iterative multi-step based on PFANN prediction was put forward in this dissertation. Compared with BP algorithm, PFANN had higher generalization ability, faster convergence and more accurate training accuracy. As the result of using GA, the training could converge to global optimal solution. Compared to the general multi-step prediction method, the method of iterative multi-step prediction had a higher prediction accuracy in the medium and long term prediction. 5. Based on the principle of system design and beyond the technique talked before, the condition monitoring and trend prediction system of centrifugal compressor K5403 was designed from the hardware and software respectively. Talking into the higher efficiency of data acquisition and anti-interference of transmission line, the PCI-6224 data acquisition card and signal conditioning module were selected. In accordance with the modular design of the system functional design, the system was divided into three parts: data acquisition, condition monitoring and condition analysis and diagnosis. The realization of each module was introduced finally.6. Finally, to sum up and outlook this thesis, the research content and innovation were introduced, and the inadequacies of this thesis were analyzed, and the further request was put forward to the next research works.