Transmission and Distribution AI
Electric utilities collect imagery and video to visually inspect transmission and distribution infrastructure. These data help utilities identify infrastructure defects and prioritize maintenance decisions. The ability to collect these data has outpaced the ability to analyze it. It is common for utilities to manually review these images to complete the inspection. This is time consuming, costly, and subjective. It’s likely some of the data review tasks can be automated by leveraging machine vision, machine learning techniques, and artificial intelligence. However, model developers need training data to create these systems. Today, there is very little overhead inspection imagery publicly available to support academic, private industry, and utility research on the topic. The Electric Power Research Institute is addressing that industry gap by collecting, cleaning, labeling, and share utility inspection imagery.
Publicly-available image and video datasets to advance machine vision, machine learning, and AI.
EPRI has collected over 150,000 images of utility infrastructure. Most of the data is imagery and is from an aerial perspective. These images have been collected through EPRI field projects or contributed from utility collaboration.
Before EPRI can share data, they must be edited to remove any Personally Identifiable Information, filenames, and location metadata. Additionally, the contributing utility may request additional obscuration.
EPRI has labeled some imagery and made available. This data has been labeled according to EPRI’s Overhead Transmission and Distribution Taxonomies, available at the following links: Distribution | Transmission
EPRI has shared data under Creative Commons licensing. The two licenses most used are below.
A summary of the datasets can be found here.
If you’d like access to datasets, please send an email titled “T&D AI Imagery Access” to AI@EPRI.Com
Many AI topics are explored in the Distribution Asset & Analytics (P180.005) project set. Some of the more current and related research includes Natural Language Processing (NLP). Please view the NLP page for more information by clicking the card below.