Research On The Simulation Credibility Assessment Of Marine Electric Propulsion System

Oil pipeline transport is the main form of oil resources distribution. With securityincidents caused by aging pipelines arousing great attention, security threat to nationaleconomy can not be ignored. Hence, it is critical to research pipeline inspection technologythat provides technical support for pipeline breakage condition. This thesis aims to probe intohow three-dimensional shape of pipeline defect can recover from Magnetic Flux Leakage(MFL) signal, which is the technical difficulty of pipeline inspection.Magnetic Flux Leakage testing is the common technique of oil pipeline inspection.Based on the application of MFL testing in pipeline inspection, this thesis first outlines thecomposition of pipeline inspection system, roles of various parts as well as rationale of MFLtesting, further elaborates the theoretical model of leakage magnetic field, and analyzesthrough magnetic dipole model the impacts of different defects parameters on MFL signal.The intelligent recognition technology of MFL signal is put forward on the basis oftheoretical studies. Intelligent recognition technology mainly contains MFL signalinterpolation, feature extraction, parameter identification as well as category judgment. Themagnetic sensor arrangement makes MFL signal discontinuous in the width direction. In thelight of discontinuity of original MFL signal, interpolation processing by cubic splineinterpolation to original MFL signal is conducted, after which the surface continuity andspatial resolution of signal are improved. In the next place, the characteristic quantity of MFLsignal and its specific meaning through mathematical description are explored. With extractedcharacteristic quantity acting as the input of defect parameter identification, soft measurementon defect width is carried out by support vector machine, which proves to be a goodmeasurement effect through the detection of sample set. Thus the author proposes geometricprimitive that approximates real defects using a variety of rules in view of two-dimensionalreconstruction, and gives high priority in the effect of support vector machine in defect typeidentification, achieving category judgment of MFL signal and enriching its additionalinformation.Three dimensional reconstruction technique of pipeline defect is also come up with inthe thesis. The technology, on a basis of continuous two-dimensional reconstruction,decomposes three-dimensional reconstruction of defects into two-dimensional reconstructionof some columns. By means of introducing the method of least squares support vectormachine in machine learning, this thesis discusses its basic principle and common kernel function, respectively employing the genetic algorithm and particle swarm algorithm tooptimize penalty parameter and regularization parameter and make comparison of algorithmperformance. Followed are the reconstruction verification on semicircular and triangle defectsand comparison of their reconstruction results by BP neural network, least squares supportvector machine as well as its optimized one by particle swarm in the two-dimensionalreconstruction of defects.Finally, simulation models of hemispherical defect, conical defect, cylinder defect andtriangular prism defect are established by finite element method, through whichthree-dimensional MFL data of three defects are obtained via data mapping and saving. Thesedata are then used in the experiment to verify three-dimensional defects reconstructionproposed in this paper, whose results demonstrate the effectiveness of three-dimensionalreconstruction technique through comparing and analyzing the shape of three real defects andreconstruction results.