Research On Defects Inversion Based On Intelligent Methods

Pipeline transportation is a special mode of transport created by the development of the oil production,which plays a very important role in the national economy and social development.But at present,due to various reasons such as time span and technical shortness,our country’s pipeline has problems such as the aging,corrosion and wear.If it is not handled in time,serious accidents will happen such as leak even explosion.For reducing the damage of pipeline fault,studying pipeline defect detection technology has important significance.Magnetic flux leakage detecting technology is a hotspot of current pipeline inspection technology,which is one kind of the nondestructive testing.In magnetic flux leakage detecting process,the last part is inversion,which is used to judge the existence of defects and calibratethe property of defects.In this paper,inversion methods of defects are designed based on many kinds of intelligent methods.By studying the structure and arithmetic of RBF neural network,the effects of its transfer function,spread constant and the clustering center on its model generation are discussed.A model based on RBF network for inversion is designed.With the test of defect data training,comparison test with the BP network and the test with abnormal data collected in inspection condition,the application characteristics and applicability of RBF neural network used in defect inversion is analyzed.By studying the concept and structure of support vector machine(SVM),the effects of its Kernel function,punishment factor are discussed.And a model of inversion algorithm based on SVM is designed.With the result of comparison with BP network,results of tests of sample quantity of defect data changing and results of testing model with abnormal condition,characteristics and stability of SVM model applied to defect inversion is discussed.By studying the development and concept of random forests,the influence of its tree parameters and the number of trees discussed.Then a model of inversion algorithm based on itis designed.Through its defects inversion contrast test with regression trees,defect data sample selection test and field interference tests,the applicability and stability of random forest in defect inversion analyzed.By studying the development and ideas of Adaboost,improved models named RBF_Adaboost and SVM Adaboost is designed.Comparing with RBF model and SVM model and results of inversion with abnormal data,it is proved that RBF_Adaboost and SVM_Adaboost are stable and accurate.In this paper,the feasibility of various intelligent methods is discussed for defect inversion,and defect inversion methods based on these algorithm is designed.Finally in this paper,by comparing the methods,it is concluded that random forests inversion suits various situations with little requirement,and Adaboost enhancement algorithm is suitable for occasions that accuracy requirement is high and time is enough,etc.