MULTIVARIATE TIME SERIES MODELLING OF THERMOGRAPHIC DATA FOR SUBSTATION INSPECTION

  • Alastair Straker

Student thesis: Phd

Abstract

This thesis aims to investigate long-term condition monitoring of power substation assets with thermal imaging techniques. Due to limited inspections and measurement techniques which are very susceptible to noise from environmental factors and human error, the inspection process can fail to detect fault pre-cursors in equipment. This thesis aims to contribute to this problem by monitoring equipment continuously, which collecting data on the electrical load and weather conditions that influence the measurements, in order to characterise the thermal response of equipment using LSTM modelling techniques. Two scenarios are presented in the form of two experiments, comprising of electrically loaded overhead lines and cable terminations. These are energised for multiple day periods with realistic load patterns, with wind effects emulated by an industrial fan, while thermal images, electrical load and environmental conditions are collected. The data are used as input to linear regression and LSTM recurrent neural networks. The work contributes the use of multiple low-cost non-calibrated thermal imaging sensors for data collection, and the novel application of LSTM recurrent neural network modelling methods to produce accurate time-series models of the thermal output of points of interest on substation equipment in a laboratory environment. It also contributes a large-scale experimental rig facilitating the long-term monitoring of high voltage power equipment at high currents in laboratory environments, enabling multi-directional thermal imaging monitoring. Lastly it provides a case study into the thermal behaviour of 66 kV cable-sealing ends when energised long-term. Recommendations for further work are outlined, including extending data collection for longer periods, conducting long-term monitoring outdoors and utilising the generated models in implementing fault-detection methods.
Date of Award1 Aug 2022
Original languageEnglish
Awarding Institution
  • The University of Manchester
SupervisorFrank Podd (Supervisor) & Joaquin Carrasco Gomez (Supervisor)

Keywords

  • time series
  • substation
  • machine learning
  • condition monitoring
  • thermal imaging
  • lstm

Cite this

'