Inverse problems in non-destructive evaluation of gas transmission pipelines using magnetic flux leakage

The critical issue in non-destructive evaluation (NDE) is the characterization of materials on the basis of information derived from the response of energy-material interactions without affecting the quality or utility of the test piece. This is commonly referred to as an "inverse problem". Inverse problems are in general ill-posed and full analytical solutions to these problems are seldom tractable. Practical solutions however employ constrained search techniques to find the optimal solution from the set of possible solutions. NDE signals recorded by the sensors are influenced not only by internal defects and degradations, but they also tend to be marred by the noise originating from various sources. Hence inverse problems in NDE need to deal with filtering techniques to obtain a noise-free signal. In this dissertation, NDE of natural gas transmission pipelines is considered.; Natural gas is transported to consumer sites through a vast network of pipelines. In order to ensure integrity of the system, the pipelines are periodically examined using Magnetic flux leakage (MFL) technique. The MFL inspection assembly is commonly referred to as "pig". In the MFL inspection tool, permanent or electromagnets are used to magnetize the pipe-wall in an axial direction and an array of Hall-effect sensors is usually installed around the circumference of the pig to sense the leakage flux caused by anomalies in the pipe-wall. The signal picked up by the sensor array is recorded and subsequently analyzed by trained analysts. Traditional methods involving manual analysis of this signal are time consuming. The performance of these methods cannot be standardized and is subject to change depending on the levels of skill and training of the analyst. The gas pipeline industry is therefore interested in automated methods for analyzing the MFL inspection signal in order to improve accuracy and decrease the turn around time between the actual pigging and receipt of inspection results.; MFL signals obtained from the pig are contaminated with noise from various sources. The variation in magnetic properties of seamless pipes introduces a quasi-periodic noise called as seamless pipe noise (SPN). Lift-off variations in sensors due to motion of pig inside the pipe and noise from electronic system hardware contribute to additional noise in the signal. Hence signal interpretation is carried out in two steps: (1) Noise removal and identification of regions of interest (ROIs) enclosing potential defects. (2) Inversion of the signal in the ROI to predict full 3-dimensional depth profile. This dissertation discusses the conventional methods of noise removal and their limitations when applied to the MFL signal. A new method based on higher order statistics (HOS) is introduced and described; its advantages over conventional methods are discussed.; Two approaches, namely direct and iterative inversion methods are presented in the second step of inverting the MFL signal to predict the defect depth profile. Both methods are based on the use of radial basis function neural network (RBFNN).; The challenge of high dimensionality (of the order of few thousands) is addressed by modifying traditional approaches described in literature. These modifications in both direct inversion as well as iterative inversion are new contributions to the field.