Distributed analytics for asset management
Using AI to predict how power transformers will fail, and what to watch out for
This task will create a decision support tool for asset managers. The utility asset manager will be able to put results of transformer inspections into the tool and receive:
- A health value for the transformer ("condition-based rating")
- A comparison of the transformer's health with other transformers from all over the world
- Possible faults, based in particular on gas levels
- A prediction on how the transformer's health will degrade in the next 1-2 years, based on how similar transformers have behaved
- Advice on what gas levels to monitor, and how to watch them
A central part of the task is to conduct the analysis without a centralised dataset, and without being able to see the asset information of other transformers. The analyst is working blind to protect the privacy of the transformer owners! This is achieved using TAC, a system provided by the challenge sponsor, VIA.
The steps we are taking in the hackathon are:
- To identify typical condition curves
- To match these curves to the test transformer
- To map the changes in health over time for the matching transformer, in particular looking at gas levels
- To predict how the test transformer will change with time by generating a heatmap and plotting the test transformer to the heatmap
- Visualisation and business value!
We have four datasets that we can use:
- A global repository of transformers that we access using VIA's privacy-preserving TAC system (including fault information)
- An anonoymised database of Swiss transformers
- A list of faulty transformer readings, from the IEC60590 standard
- A list of normal transformer readings, from the IEC60950 standard
The decision support tool will be combined with online and offline asset management approaches. It will be combined with load and geospatial data to create an exploitable tool.
Prototyping has been completed and now the project will be taken forward for funding