Smart Network Diagnostics: Online Partial Discharge Monitoring of Electrical Networks
Sonia’s research focus on the project was based on the insulation diagnosis of cable systems, through on-line partial discharge technique and deep neural network algorithms. To this regard, the main contribution to Sonia to MEAN4SG’s scope has been the Smart diagnosis for electrical networks.
This research was carried out at OCT (Ormazabal Corporate Technology – Spain). It focussed on the insulation diagnosis of cable systems, through on-line Partial Discharge (PD) detection techniques and deep neural network algorithms.
There are systems available on the market for the detection and measurement of PD sources. However, these systems tend to be expensive and often require interruption in service for installation and measurement. Nowadays, in a smart grid context, on-line techniques are more suitable, and utilities are working alongside manufacturers to develop more affordable monitoring solutions. Smart condition monitoring is the key solution to diagnose equipment incipient failure, and this requires intelligent algorithms to extract meaningful information from raw data, make predictions and provide accurate, real-time asset health insights. Therefore, deep learning methods could be an excellent option for these requirements.
The accurate localization of a partial discharge source can reduce the response time for a maintenance team dramatically. In terms of partial discharges in cables, the pinpointing of a PD source to within a few meters equates to a maintenance team being able to go directly to the place and excavate a small area. Within the substation environment, a PD source could be more difficult to localize, and as such, the identification of PD type is useful.
The automatic identification of PD sources on-line using an attractive techno-economically optimized system based on deep learning methods, which is possible with the ongoing advancement of GPUs (graphics processing units) for embedded applications and algorithms written in freely available programming languages such as Python or R, is an attractive option for network operators, maintenance providers and also brings added value to equipment manufacturer’s products. Therefore, the development of this technique is well aligned with the scope of MEAN4SG.
Fundación para el fomento de la innovación industrial (FFII) (12.2017 – 03.2018)
Fundacion CIRCE (03.2019 – 05.2019)
- Answer to the issues in Accurate Fault Location by synchronizing measurements performed in UDEX. Results: Method developed taking into account network configuration
- Answer to the issues in Accurate Fault Identification by s Deep learning method development for PD identification. Results: Novel method for PD identification through image classification
- Answer to the issues in Techno-economically viable solution by the use of low cost system for acquiring data and identification algorithm. Results: Low cost system used successfully with DL algorithm implemented on low cost optimised computing system.
- 4 publications. The thesis manuscript is been writing with the results and a publication in IEEE journal is planned to be submitted.
Deep Neural Network for partial Discharge sources recognition in distribution network
- “Measuring PD propagation in complex MV distribution network configurations”. Sonia Barrios, Ian Gilbert, Aritz Hurtado, Patrick Mulroy, Iñaki Orue (CIRED) [Repository Link: https://zenodo.org/record/3856012]
- “Partial Discharge Classification Using Deep Learning Methods—Survey of Recent Progress”. Barrios, Buldain, Comech, Gilbert, Orue (Energies) [Repository Link: https://zenodo.org/record/3855657]
- “Practical measurements of partial discharges in a smart grid laboratory”. Sonia Barrios, Ian Gilbert, Aritz Hurtado, Patrick Mulroy, Inaki Orue (2018 First International Colloquium on Smart Grid Metrology (SmaGriMet)) [Repository Link: https://zenodo.org/record/3856419]
[To be updated]
Master degree in Electrical Engineering
Dr. Iñaki Orue
Dr. Ian Gilbert
Ormazabal Corporate Technology (OCT)