Insulation breakdown is a key failure mode of high voltage (HV) equipment, with progressive faults such as electrical treeing leading to potentially catastrophic failure. Electrical treeing proceeds from defects in solid insulation, and cables are particularly affected. Research has shown that diagnosis of the fault can be achieved based on partial discharge (PD) analysis. Nonetheless, after diagnosis of a defect, engineers need to know how long they have to take action. This requires prognosis of remaining insulation life. The progression of a defect is far less well understood than diagnosis, making prognosis a key challenge requiring new approaches to defect modelling. The practical deployment of prognostics for cable monitoring is not currently feasible, due to the lack of understanding of degradation mechanisms and limited data relating defect inception to plant failure. However, this thesis advances the academic state of the art, with an eye towards practical deployment in the future. The expected beneficiaries of this work are therefore researchers in the field of HV condition monitoring in general, and electrical treeing within cables in particular. This research work develops a prognostic model of insulation failure due to the electrical treeing phenomenon by utilising the associated PD data from previous experiment. Both phase-resolved and pulse sequence approaches were employed for PD features extraction. The performance of the PD features as prognostic parameters were evaluated using three metrics, monotonicity, prognosability and trendability.The analysis revealed that features from pulse sequence approach are better than phase-resolved approach in terms of monotonicity and prognosability. The key contributions to knowledge of this work are three-fold: the selection of the most appropriate prognostic parameter for PD in electrical trees, through thorough analysis of the behaviour of a number of candidate parameters; a prognostic modelling approach for this parameter based on curve-fitting; and a generalised framework for prognostic modelling using data-driven techniques.
|Date of Award||20 Jun 2018|
- University Of Strathclyde
|Supervisor||Stephen McArthur (Supervisor) & Victoria Catterson (Supervisor)|