Remaining Life Assessment Method Research For Centrifugal Pump Approaching Overhaul

Centrifugal pump is widely used in oilfield production site.Regular maintenance or experience judgment is adopted in the overhaul on site,which lacks the remaining life assessment of centrifugal pump approaching overhaul.Therefore,it is of great significance to carry out the research on the remaining life assessment of centrifugal pump approaching overhaul.In this paper,the centrifugal pump is taken as the research object,and the degradation index construction,overhaul standard threshold determination and remaining life assessment of centrifugal pump are studied.The main contents are as follows:(1)A method to construct degradation index of centrifugal pump based on multi-source feature fusion is proposed.The root mean square of vibration speed,the peak to peak value of vibration displacement and the amplitude of blade passing frequency,which reflect the running state of three key parts,namely bearing,pump shaft and impeller,are extracted from the vibration characteristics as the vibration indexes.The outlet pressure which reflects the degradation performance of centrifugal pump is extracted as the characteristic parameter.The vibration index and the characteristic parameter index are combined into a multi-source feature set to perform feature fusion using KPCA,and the first kernel principal component is obtained as a degradation index that reflects the degradation process of overall health state of centrifugal pump.Oilfield data is used for example analysis and the trend index of a centrifugal pump is calculated to be 0.9169,which verifies that the degradation index based on multi-source feature fusion can effectively characterize the degradation process of centrifugal pump.(2)A model based on optimized probabilistic neural network(PNN)is proposed to determine overhaul threshold for centrifugal pump.The kernel principal component analysis is used to reduce the original data dimension and optimize the structure of PNN.The genetic particle swarm optimization algorithm is used to optimize the smoothing factor of PNN to improve the effectiveness and fault tolerance of the model.The overhaul threshold model of centrifugal pump is established based on the optimized PNN,and the dynamic overhaul threshold curve is established based on the field data.By analyzing the overhaul case of a centrifugal pump in May 2016,the overhaul was performed in time just when the degradation index of the centrifugal pump exceed the overhaul threshold,and the overhaul date was advanced by about two months,indicating that the proposed overhaul threshold model can judge whether the centrifugal pump should be carried out overhaul or not,and avoid the blindness of overhaul work.(3)A model based on the optimized extreme learning machine(ELM)is proposed to evaluate the remaining life of centrifugal pump.The genetic particle swarm optimization algorithm is used to optimize the input weight and bias of the ELM to improve the prediction accuracy.Based on the optimized ELM,the remaining life assessment model is established,and the degradation index obtained after multi-dimensional feature fusion is used for prediction.The mean square error and determination coefficient of the prediction results are 0.0003 and 0.9984 respectively,indicating that the optimized ELM model has smaller error and higher accuracy than other models.Based on the prediction of the degradation index and the overhaul threshold,the remaining life of centrifugal pump is evaluated.The analysis of the September 2018 overhaul case of a centrifugal pump shows that the remaining life assessment model proposed in this paper can evaluate the remaining life of centrifugal pump approaching overhaul to perform condition-based maintenance,and avoid unplanned downtime accidents caused by insufficient remaining life.