Fault Diagnosis For Turbine Vibration Based On Genetic Algorithm For Optimization Of Two-stage Neural Network

In the field of electricity production, steam turbine has a pivotal position, for thislarge and complex equipment, the computer systems with its high-speed, accuratelycalculate and long-term stability advantages play a huge role in the monitoring andfault diagnosis of steam turbine. The face of the increasingly complex operationrequirements, the core issues of turbine vibration fault diagnosis system encounteredis how to deal with complex environmental and informational condition. Throughtheir meta-analysis, a new effective fault diagnosis system is expected to resolve theproblem.Through the analysis and summary of the actual principle of the steam turbinefault, this paper took a common steam turbine vibration fault as the research object.First the architecture of the turbine vibration fault diagnosis system were summarizedand the application and feature of mechanical vibration’s signal data in time domainanalysis, spectral analysis, orbit of shaft center analysis and other commonly usedmethods were presented. The common underlying cause of turbine vibration failure,as well as the production of site-specific mechanism of fault symptoms was studied.And the paper summarized the vibration signal characteristic of long-term stability inunit operating conditions, and continued to collect data rotor vibration faults and faultsymptoms. A more detailed of rotating machinery and vibration signal characteristicswere analysis, including the imbalance, misalignment, rotor rubbing and othercommon faults. Second, according to the use of steam turbine equipment vibration’sspectral signal data, the BP neural created a simple diagnostic model by its theoreticaladvantages of fault diagnosis. Then the BP model was optimized by the geneticalgorithm in global optimization features, the failure was narrowed the scope to asmaller categories range. Finally, the Bayesian network was brought into GA-BPturbine vibration fault diagnosis system. The Bayesian network analyzed the result ofGA-BP turbine vibration fault diagnosis system, combined with other informationsuch as orbit of shaft center, phase etc. The Bayesian network constructed a geneticoptimization of two-stage neural network turbine’s vibration fault diagnosis expertsystem. The system completed the study of the theory in modeling, reasoning andconclusion of the analysis process.