Research And Application On Condition Assessment And Maintenance Optimization Of The Turbine Driven Boiler Feed Water Pump Set

The turbine driven boiler feed water pump set is an important auxiliary part of the thermodynamic cycle system in thermal power plant.Through it,the pressure of boiler feed water can be increased,the condensate water can be converted to boiler feed water,and the attemperation water could be supplied to superheater and reheater.In addition,the running consumption of the turbine driven boiler feed water pump set is also very large,accounting for about three percent or more of the unit power.So the safe and economic running of the turbine driven boiler feed water pump set is closely related to the safety and economy of the whole unit directly.However,compared with the main equipments such as boiler,turbine and generator,condition assessment of the turbine driven boiler feed water pump set has not been given special attention.So it is a weak link of equipments’ condition monitoring in the thermal power plant,and it may lead to drop load operation or even unplanned shutdown of the unit when it breaks down.In recent times,the data mining technology is developing and maturing rapidly.Faced with the existing and rapidly growing data in many kinds of control systems and information systems in the thermal power plant,it is becoming more and more necessary and urgent to take use of these silent information and make effective knowledge mining from it.Also,new opportunities will be brought to the condition assessment and condition-based maintenance of a large number of equipments in power plants.Based on data mining technology,this paper is composed of the following parts:(1)The history data in SIS is converted to the input parameters of classification models rationally by a feature extraction method based on statistics and a feature selection algorithm named Relief.After that,classification models for the distinguishing between normal condition and fault condition of a boiler feed water pump turbine(bfpt)which ever had a fault of blades fracture and a boiler feed water pump(bfp)which ever had a fault of balance disk rubbing in two power plants are made respectively through five classification algorithms.Then the models are verified in actual cases.According to the results,BP neural network,support vector machine and combined classification have better performance than others.And the potential risks of faults can be identified 4 to 10 weeks in advance by these models to avoid the drop load operations of the units or even the occurrences of insecurity events.Accordingly,a new thought is provided for the fault predictions of more kinds of equipments.(2)The performance prediction of the turbine driven boiler feed water pump set is studied through the numerical prediction algorithms.And the key parameters of preformance predictions for bfp and bfpt are analyzed respectively.Compared with the traditional performance calculation methods,the dimension of parameters needed for preformance prediction is reduced significantly.The prediction results are in good agreement with the verified data and can meet the practical requirement of the engineering.During this study,the parameters are sorted followed by the analysis of each parameter’s influence on performance prediction.Finally,the performance predictions of bfp and bfpt are accomplished by taking the parameter sets which have larger impact on performance predictions of bfp and bfpt respectively as the feature vectors.(3)The practical applications of fault prediction and performance prediction for the turbine driven boiler feed water pump set in the power plant are discussed.The condition assessment including fault prediction and performance prediction for the turbine driven boiler feed water pump set is combined with the theory of risk-based maintenance.Accordingly,the failure possibility of the turbine driven boiler feed water pump set can be evaluated quantitatively and the risk rank can be determined automatically.So the corresponding maintenance advices could be made according to the risk rank.This method has a guiding significance for the maintenance optimization of the turbine driven boiler feed water pump set.(4)The main research conclusions in this paper are generalized and summarized.According to the problems and shortcomings of the fault prediction,performance prediction and maintenance optimization methods in this research,the directions and ideas for the follow-up research are further prospected.