Date of Award

5-31-2018

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Systems Engineering

First Advisor

Eric Sandgren

Abstract

Microwaves promulgate through a waveguide, reflecting off its inner surface, of course. It is known that any irregularity (discontinuity) in the surface will alter the course of the reflected signal, and that the reflected signal can be recaptured and processed algorithmically to generate a waveform characteristic of the discontinuity. The work of this dissertation is based in this phenomenon. The aim is to develop methods for detecting and then characterizing pipe wall thinning (PWT) discontinuities in pipes. The work includes both simulations and practical experiments. New possibilities were discovered as the work progressed. Prior work revealed a promising initial setup and as the work proceeded, adaptations and refinements optimized the results. The initial phases of research were conducted using Computer Simulation Technologies (CST) software and a standard set of PWT specimens so that the simulations could be calibrated. The modeled (and later the practical) setup consisted of a vector network analyzer providing a sweeping frequency range of microwaves from 1.4 GHz to 2.3 GHz. The frequency range was determined by the inner diameter of the pipe (6 in., 152.4mm). The pipe length was 1m, with PWT specimens located along its length. All theoretical pipes had 0.5in (12.7mm) wall thickness. The PWT specimens were set as percentages of that thickness. The initial simulations were made with only microwave signals used to detect and characterize the PWT specimens, which yielded inconclusive location results. Still, the measurement errors were relatively consistent, so a second phase was added to the protocol, to more accurately locate and characterize the PWT specimens. The more successful technique is straight-beam ultrasonic (UT) probes guided to a starting position based on the predictions of the microwave phase. During the practical experimental phase, straight beam UT was used and could both accurately locate the PWT specimen, and draw the profile of a rectangular or semicircular cross-section for characterizing the shape of the PWT. In the process of determining how to apply UT to the system, angle-beam UT was also tested, and an optimization technique was developed using CST simulations. The predicted locations of the PWT derived from the microwave phase fell within a half-width (W/2) ahead of and W/2 beyond the actual location. This led to the development of an innovative sensor positioning equation to calculate the excitation points (EPns) to use when positioning the UT probe. In addition, the EPns are passed to a genetic algorithm to fine-tune the angles used for the UT probes. In practical application, these same manipulations worked well with straight-beam UT probes. While developing the protocol, it was realized that pattern recognition using correlation analysis could be applied to the microwave waveforms in order to characterize discontinuities. To verify this, 71 different CST simulations were used to develop a database to which unknown PWT specimens could be compared. This process was successful both in simulation and experimental identification of unknown specimens. The experimental verification used only three positions, three profiles and three depths of PWT. A further advance that grew out of this pattern recognition work was the development of a neural network that can learn and build the pattern recognition database on its own. In experimental terms, development of the database would be prohibitively expensive. The 108-sample training dataset was developed using CST, was then learned by the neural network, which itself optimized the weights and topology for the network. The network was then able to predict the characteristics of unknown samples that were introduced to the system, with acceptable accuracy. This advance is certainly worth further development work.

Download

Included in

Engineering Commons

Share

COinS