Skip to main navigation Skip to search Skip to main content

Système de réalité augmentée pour la détection et l’analyse temps-réel d’équipement de réseau électrique

Translated title of the thesis: Augmented reality system for real-time detection and analysis of electrical equipment
  • Olivier Beaudoin

Student thesis: Master's thesisMaster in Engineering: Information Technology Engineering

Abstract

Hydro-Québec’s electricity distribution network is aging and requiring more and more maintenance. Following a large-scale inspection of the utility poles and transformers in its network, Hydro-Québec has established a semi-annotated image bank containing examples of poles and transformers in good and bad condition. In order to automate the visual inspection of equipment in its network, an intelligent model will be built using neural networks. The task of this model is to recognize certain elements of the network (pole, transformer) and to give a status rating to them. In addition, the model must also read LCLCL nameplates in order to perform the correction of another database. This model can be run locally on an Android mobile device or remotely, without affecting its performance. The Android mobile device provides GPS data and accelerometer data to increase the quality of model detections. The extracted segmentations are analyzed in order to deduce a status rating according to the type of equipment. The poles are rated primarily according to the angle and the transformers according to the presence or not of an oil leak. If one of these two indicators is detected by the application, the technician performs a detailed inspection to potentially detect other problems. The geo-referenced database of Hydro-Québec utilities contains errors about positioning , type of equipment and registration. The mobile application reads the LCLCL plate affixed to the pole and uses the GPS data of the device to search the current database. In a case where the GPS coordinates or the sequence of letters and numbers differ from what the database contains, the technician has the possibility of issuing a correction. The use of synthetic data has made it possible to make some progress in research but is problematic for more complex scenarios. Lack of tagged data pushes towards using adversary networks to improve generation of synthetic data, automate tagging and more.
Date25 Jun 2020
Original languageFrench
Awarding Institution
  • École de technologie supérieure
SupervisorChristian Desrosiers (Supervisor) & Luc Vouligny (Co-supervisor)

Cite this

'