Research On Feature Selection And Extraction On Rotor Condition Monitoring

In the process of production, rotating machine is the central part and the quality of its running state directly affects the production process. If there is a fault in the rotating machine, it may cause a chain reaction and serious consequences. It can be effective in preventing accidents that taking real-time status monitoring on rotating machine and analysising and diagnosising the fault. Feature selection and extraction are the key technologies of the condition monitoring and fault diagnosis of the equipment and the effect of the condition monitoring and fault diagnosis can be directly influenced by the fault recognition of the characteristic parameter. By the feature selection and extraction technology the characteristic parameter with excellent fault identification can be achieved. In this paper,various characteristic parameters in the time domain and frequency domain are studied and the feature selection and extraction technology executed by the genetic algorithms and genetic programming are investigated.Firstly, most failures of the rotor machine are caused by the vibration, so the vibration mechanism of three typical rotor failure: imbalance, misalignment and bearing part loosening are investigated and their fault features are obtained. The characteristic parameter in time domain and frequency domain, specially the properties of dimensionless characteristic parameters and its principle of fault identification are studied.Second, the feature selection and extraction based on the genetic algorithm and the genetic programming are explained. The basic component and the procedure of the algorithm are described in detail. The combination of the genetic algorithm and the genetic programming is proposed to solve the problem of the calculation of the genetic programming.The feature selection and extraction will be completed by the genetic algorithm and the genetic programming in sequence.Third, based on the tendency of the networking of the condition monitoring system, the vibration signal collecting and disposal installation with wireless transmission is designed and its composition and function of each module are specified. Through the testbed for measuring rotor vibration, three typical rotor failures are simulated, investigating the characteristics of the time-domain waveform and power spectral of the rotor vibration. And fault features from the experiment are compared with the fault feature in theory.Finally, the values of characteristic parameters in the time domain and frequency domain are achieved by processing experimental data and their fault recognition especially for dimensionless characteristic parameters are studied. The genetic algorithm is used to undertake the feature selection of dimensionless characteristic parameters achieved in the experiment to remove characteristic parameters whose fault recognition is weaker. Then feature extraction for characteristic parameters which have been chosen are taken by the genetic programming,to construct a new dimensionless characteristic parameter. Meanwhile dimensionless characteristic parameters and characteristic parameters in frequency domain are mixed as the input for genetic programming, achieving the mixed characteristic parameter. At last, fault recognition of the new characteristic parameter and conventional characteristic parameters are compared.