Health State Assessment And Remaining Useful Life Prediction Of Oil Rig Drilling Pump

This thesis comes from the project of Sichuan Science and Technology Department:2019 Sichuan intelligent manufacturing and robot major science and technology project "intelligent drilling machine development and application".Oil drilling pump is the core equipment of drilling rigs.In order to monitor the operation status of the drilling pump and ensure the safe and reliable operation of the drilling rig,this thesis carries out research on health status assessment and remaining life prediction research of oil drilling pumps,to manage the oil drilling pump in a correct and reasonable way,so as to realize the active maintenance of the oil drilling pump,which is of great significance to reduce the economic cost and improve the reliability and safety of the oil drilling rig operation.This thesis mainly completes the following work:Firstly,according to the non-linear and non-stationary characteristics of oil drilling pump vibration signal,the characteristic information was extracted from the drilling pump vibration signal under limited conditions.Firstly,the locations of vibration signals were analyzed for the many core components of the drilling pump and the small distance between them,and then the time and frequency domain characteristics of vibration signal were extracted.In view of the complex frequency components of vibration signals,empirical mode decomposition was performed to further extract the time-frequency characteristics of the mode components.Based on all the time-frequency domain characteristics,a high-multidimensional and multi-domain feature set was constructed to describe the health status of the drilling pump.Then,aiming at the problem of the correlation and redundancy of the above-mentioned high-dimensional features of the vibration signal,an improved local tangent space permutation algorithm was used to reduce its dimension.In the algorithm,the label information and local set coefficients were used to dynamically construct data point neighborhoods,the maximum likelihood method was used to estimate the low dimensional embedding dimension of high-dimensional data,so as to reduce the aliasing between different types of data,and to achieve the effective simplification and fusion of high-dimensional features.On the basis of the above-mentioned low-dimensional features,a method for evaluating the health status of drilling pumps based on Softmax and Mahalanobis distance(MD)was proposed.First,the low-dimensional feature data was clustered through the GG(Gath-Geva)clustering algorithm to classify the health status of the drilling pump,and then the trained Softmax classifier was used to identify the health status of the test sample.The Mahalanobis distance between the output state sequence and the ideal state was calculated,and was normalized to obtain the health index,next,the range of the health index under each health state was obtained,so as to realize the division of the health state of the oil drilling pump pinion shaft and the qualitative and quantitative evaluation of the health state.Finally,in order to solve the problem of poor fitting of multi-stage support vector regression(SVR)model in small samples,based on the support vector regression method,combined with health state partition and cluster sampling,this thesis selected the appropriate super parameters,and took the pinion shaft as the experimental object,and the effective and accurate remaining life prediction of oil drilling pump was realized under the simulation and example verification.