Loading dissolved gas analysis (DGA) and oil quality dataset

Is AI the Next Frontier for Transformer Assessment?

March 25, 2019
Artificial intelligence can be the tool to guide future developments and applications in the T&D industry.

Industry Challenge

Transformer condition assessment is complex. Transformers can fail in many ways — each with different signatures in the condition data. Experts must examine a variety of data to identify signatures, determine condition, and suggest appropriate corrective action. Today’s business environment requires experts to assess transformer condition and risk with reduced manpower, reduced domain expertise, and an increasing amount of raw data.

Periodic dissolved gas analysis (DGA) and oil quality assessment are routine practices while on-line DGA monitors provide a wealth of data at much greater frequency. The challenge is how to effectively analyze this data; artificial intelligence (AI) could play an important role, but research is needed to assess the technology.

EPRI’s Response

The Electric Power Research Institute (EPRI) is expanding its expertise and a proven history in transformer analytics to new and emerging analytic methods, such as AI. AI is rapidly emerging for utility analytics and has the potential to provide valuable, actionable results. However, utilities need to be confident in the effectiveness of these AI solutions.

AI’s success hinges on availability of rich data sets that are well curated and properly labeled for training. It is unlikely that any single utility has enough data to build a solution. Given this, it is beneficial to leverage collaborative data sets curated by knowledgeable asset experts. EPRI’s approach is to facilitate the AI community to develop solutions by:

  1. Acquiring DGA and oil quality datasets from in-service and failed transformers.
  2. Labeling and curating that data so it is usable for AI systems.
  3. Separating the data into training and evaluation sets.
  4. Providing the training dataset to the AI community for solution development.
  5. Educating the AI community on transformer fundamentals: failure and degradation modes and the relationship to the available data.
  6. Quantifying AI solution performance using performance metrics.
  7. Comparing the performance against traditional and physics based methods such as the IEEE, IEC and EPRI Power Transformer Expert System (PTX) methods.

Rather than develop AI solutions, EPRI intends to enable solution development by providing large datasets to a wide range of AI developers and independently evaluate their performance.

Progress, Results, & Next Steps

EPRI has collected and classified more than 640,000 DGA records from 46,600 transformers from 27 utilities worldwide. A team of world class experts has been assembled for education of the AI community, training dataset curation, and AI solution evaluation.   Engagement with several AI solution providers has been initiated.

How to Use the Research

T&D utilities will be able to use the results to confidently deploy AI solutions. Additionally, the learnings will help guide future developments and applications of AI in the T&D industry.

How to Get Involved

Utilities can help by supplying historical DGA and oil quality data from in-service/failed/retired transformers for creation of a more robust data-set. Additionally, they can make EPRI aware of interested AI solution providers.

About the Author

Bhavin Desai

Bhavin Desai is a senior program manager responsible for EPRI’s T&D Asset Management Analytics research.

Voice your opinion!

To join the conversation, and become an exclusive member of T&D World, create an account today!