Research And Development Of Early Fault Warning Of Fan In Power Plant Based On Multivariate State Estimation Technique

The fan is an important auxiliary equipment for coal-fired plant, its running state has direct effects on the economy and security of power generation. As with the rapid development of coal-fired plants of large capacity and high parameter, it makes greater demands on the fan reliability. Nowadays, industrial equipment state identification technology is developing from condition monitoring and fault diagnosis to early fault warning. Based on the Multivariate state estimation technique (MSET), the early fault warning method for power plant fan is studied in the paper. The valuable time can be obtained by the method to take measures to reduce fault losses or avoid fault, so enormous economic benefits can be brought for power generation enterprises.The fan structure and common faults are studied and analyzed to gained the monitoring signals indicating faults progress, and the currently available monitoring variables of power plant fan are summed up. On the basis of it, the MSET modeling variables are selected by the priciples of’available’,’sensitive’,’simplest’. Then the history data is preprocessed by removing abnormal data, choosing 3 from 1 for bearing temperature, standardizing, and the dynamic process memory matrix construction method was proposed to build the fan MSET normal running state model. Finally, the simulation modelling is carried out based on the normal working history data from a induced fan in a certain power plant, the result indicates that the state modeling for fan based on MSET is of high precision, it can meet the demand of the fan early fault warning.The study on fan state modelling based on MSET shows that rich faults information is hidden behind the difference between observation vector and estimation vector. In order to dig fault information fully and capture the fault process, the similarity function of observation vector and estimation vector is put forward to quantify their difference degree, and based on the importance of each variable for early fault warning, the weight coefficient of each variable was obtained by Analytic hierarchy process(APH). Then the sliding window statistical method is taken to decline the influence of random disturbance, and the reasonable alarm threshold is set by the minimal average similarity value of normal running state. So the early fault warning solution based on the fan MSET model was put forward, if the average similarity for new observation vector is beyond the threshold, the fault alarm can be sent out to remind operators to take measures. Finally, the method was applied on a certain fault and three kind simulated faults of a induced fan in a certain power plant, the result indicates that the proposed method can discover the incipient fault and realize accurate early fault warning for fan.A set of fan early fault warning system in power palnt is developed by the proposed method and B/S structure. The practical application indicates that the early fault warning method proposed in the paper can make it, and provide a feasible solution for early fault waring of other industrial equipment.