Abstract:
This thesis focuses on fault diagnosis of wired electrical networks using reflectometry. To
develop diagnosis algorithms; we have studied both, direct problem (based on numerical
simulations of electrical networks) and the inverse problem (which allows obtaining the
network parameters from its measurements). For the direct problem, a very High Resolution
Finite Difference Time Domain (HR-FDTD) method can be used to simulate how a given
electrical transmission network is affected by hard/soft faults with enough fidelity. However,
such process requires heavy and costly computational resources. A valid alternative to the
FDTD based model relies on the use of a quicker and more efficient analytical based
approach. For instance, the wired electric networks supporting a TEM propagation mode can
be easily modeled by solving the transmission line telegrapher equations. Since the precision
of the optimization algorithm in identifying the fault location and topology requires a model
sensitively similar to the considered network, the key step remains the precise evaluation of
the p.u.l RLGC parameters of the cable. A further improvement of the model accuracy
consists in the modeling of the parasitic inductance associated to the T-junction.
Finally, the ultimate goal of the highly realistic forward modeling being the inversion, it is
important to know that reflectographs obtained by forward modeling are generally difficult to
interpret, and they are not self-explanatory. For this purpose, multiple iterations are required,
and thus, an efficient optimization algorithm is introduced in this dissertation, which is able to
treat the inverse problem, and to identify any abnormality affecting the network in a short
computational time. The methodologies and algorithms proposed in this thesis are validated
either by numerical simulations or by real measurements.
