Research Of Power Plant Fan Failure Prognostic System

With the development of large-scale and the increasingly number of the power plant equipment in power plant, failure increases in frequency and positions correspondingly. The accident is bound to threaten the personal security. It is essential to improve the reliability of the key equipment of power plant. Ensure the normal operation of equipment is increasingly important. The statistics coming from the State Electricity Regulatory Commission Power Reliability Management Center shows that the major equipment malfunction failures caused unplanned downtime have reached two times a year. The key equipments of the power plant are closely related to the whole plant production safety, such as generator, steam turbine, and fan, etc. Condition monitoring and early fault warning of the key equipments plays more and more important role in reducing accidents.Plant fan is an important auxiliary equipment of power plant, which generally refer to forced draft fan, induced draft fan, primary air fan and dust-discharging fan, etc. Induced draft fan transports high temperature flue gas with impurity. In the poor working conditions, it has higher fault ratio. The failure directly threaten the safety of the boiler combustion and equipment,and affects the economic effect of power plant.In this paper, failure prognostic system takes induced draft fan of Tuoketou Power Plant’s 600 MW unit as an example. The fault symptom reflects on vibration. In view of vibration of induced draft fan, a failure prognostic system is proposed, which is based on multivariate state estimation technique(MSET) and sliding window residual statistical method with the weighting matrixes. Simulations are performed based on MATLAB. The system consists of three parts. They are data preprocessing, data modeling, condition monitoring and fault early warning. The first step is data reduction with principal component analysis and rough set. And a brief knowledge system is achieved. Building predictive modeling using MSET is the second step, and results are compared with the least squares support vector machine combined with PSO. Sliding window statistical method with the introduction of weight vector analyzes the residual of estimation value and the actual value. The change of the two curves can reflect the residual’s changing trend. When the curve exceeds the setting threshold, the equipment is on abnormal operation. Then failure prognostic system is realized.