The Application Of Neural Network In Magnetic Flux Leakage Signals

The inspection of the pipeline is an important means to assure the integrity and safe operation of the nation's vast gas pipeline network. A sensor assemblage that employs magnetic flux leakage (MFL) principles is often used to inspect the gas transmission pipelines. Detecting and characterizing defects of the pipeline is the overall goal of the inspection procedure. The paper describes a method to map the second-dimensional profile of the defects by training the neural networks.Firstly, the basis theory of MFL detecting together with the structure and the working procedure of the pipeline MFL detecting instrument is introduced. The defect characterization network requires an extensive data set for training. Since experimental data is limited, the numerical models simulating test are employed for generating training data. The MFL signals used for training the neural network are generated using a finite element model.Then the finite element analysis software (ANSYS) which is applied on the simulation of magnetic flux field is briefly described. Once these signals are generated, the neural network may be trained.Secondly, the detailed discussions of artificial neural networks are presented, including the artificial neural element model, the structure of neural network and the learning algorithms of neural network. A more popular neural network, namely, the radial basis functions neural network is described. This focuses on the essential and learning algorithms of RBF neural networks.Finally, the thesis describes the application of RBF networks for characterizing defect in natural gas transmission pipelines, and describes an extension of the concept where an RBF network is used to characterize the complete defect profile. Results obtained to date have proven the feasibility of using neural networks for solving inverse problems in nondestructive evaluation.The objection of the paper is mapping the defect profiles by training the RBF neural networks, the results indicated the method is helpful to the predicted goal, and it can establish a foundation for intelligent defect discrimination.