Wired, optical and pipeline networks support a plethora of critical systems and services in our society today. These include the distribution of electricity, communication, gas and water in diverse and critical infrastructure such as buildings, transport systems, urban water systems, automobiles, airplanes, electrical grids and wireless communications. With the increasing demands for more systems and services, these networks are even becoming more intricate with time. While straightforward maintenance is possible by replacing fault sections on a case by case basis, such methodology can no longer guarantee a sustainable, safe and reliable operation of the utility networks. The availability of techniques for detecting and locating faults that will, or have already, occurred is thu...[ Read more ]

Wired, optical and pipeline networks support a plethora of critical systems and services in our society today. These include the distribution of electricity, communication, gas and water in diverse and critical infrastructure such as buildings, transport systems, urban water systems, automobiles, airplanes, electrical grids and wireless communications. With the increasing demands for more systems and services, these networks are even becoming more intricate with time. While straightforward maintenance is possible by replacing fault sections on a case by case basis, such methodology can no longer guarantee a sustainable, safe and reliable operation of the utility networks. The availability of techniques for detecting and locating faults that will, or have already, occurred is thus more critical and urgent than before. A common element in these diverse networks is an underlying model that characterizes their key properties and this is the acoustic and electromagnetic guided wave channel. Fault detection in network infrastructure can then be formulated as a one-dimensional (1D) imaging or inverse scattering problem based on the wave equation.

In this thesis, multiple signal processing techniques based on an approximate 1D inverse scattering framework are applied and experimentally verified. Specifically, an approximate inverse scattering framework based on the Born and Rytov approximations are utilized to investigate "soft" faults in urban utility networks, including transmission lines and pipelines.To deal with the bandwidth limitation of the measurement due to high attenuation and dispersion at high frequencies or difficulty of high frequencies measurement, super-resolution techniques are introduced. Passive imaging is also considered as well as dispersion estimation with IQML (Iterative Quadratic Maximum Likelihood) algorithm to enable the reconstruction when the measurement is restricted to low signal-to-noise ratios and inaccessibility of multiple measurement locations.

In this thesis, contributions to four areas of ID inverse scattering problem for fault detection in utility networks are presented. The first contribution is developing an analytical formulation, based on the Born approximation, for reconstructing impedance faults in transmission lines and pipelines. Experimental and simulation results show that surprisingly accurate results can be obtained. The second contribution is formulating the 1D inverse scattering of discrete point faults as a sparse reconstruction problem. Convex optimization is applied to super-resolve the profile of discrete point faults in the utility networks. The techniques have applications in determining the structure of faults in utility network components, such as connectors in the transmission line, when the available spectral information is not sufficient to obtain the spatial resolution required. Simulation and experimental results in the transmission line are used to demonstrate the effectiveness of the approach. The third contribution is the extension of the approximate inverse scattering framework where active sensing cannot provide sufficient signal-to-noise ratio. Passive imaging is proposed to overcome this problem and utilizes the high noise power in many utility networks. The fourth contribution is applying frequency domain IQML algorithm to enable dispersion estimation with a single measurement. Traditionally measurements on multiple locations are necessary and therefore this method provides significant reduction in measurement resources. Simulations and experiments in a water pipelines are used to demonstrate the accuracy of the estimation. Finally conclusions are provided and future research directions are suggested.