M.S. in Engineering
Electrical & Computer Engineering
Henry M. Rowan College of Engineering
National Science Foundation; National Energy Technology Laboratory
Gas pipelines; Nondestructive testing
Electrical and Computer Engineering
Signal inversion in nondestructive evaluation (NDE) applications is a critical step before remediation decisions are made. The accuracy and confidence of the signal inversion results therefore play a key role in evaluating the effectiveness of the NDE procedure. Conventional NDE signal inversion algorithms that employ artificial neural networks treat all geometric regions of the NDE signal equally. Consequently, when the inversion algorithm is presented with input data that is significantly different from the training data, the performance of the network deteriorates significantly. This thesis presents a superior alternative for NDE signal inversion. Different geometric regions of the NDE signature are assigned different confidence levels; separate neural network inversion algorithms are applied to each region and the results are combined. The neural network inversion algorithm consists of radial basis functions that implement geometric transformations of the input NDE signals. It is shown that this "divide-and-conquer" strategy yields robust results, especially when applied to test data that the neural network has not seen before. While the algorithm is exercised theoretically using simple 1-D and 2-D defect geometries, the technique is also validated using NDE inspection images from a suite of test specimens representative of the in-line inspection of gas transmission pipelines.
Bram, Justin Gary, "A "divide-and-conquer" strategy for NDE signal inversion in gas transmission pipelines" (2006). Theses and Dissertations. 842.