- Guidance on technologies and approaches for automating distribution inspection
- Learn through laboratory and field experience how to collect the right inspection data
- Understand the feasibility of using automated image analysis for defect detection
- Quantify performance of artificial intelligence predictions with objective processes and datasets
- Provide objective data to inform deployment decisions such as inspection processes, vendor selection,and purchasing decisions
The video below introduces the new supplemental project.
Download meeting materials here.
Download full project description here
Background, Objectives, and New Learnings
Distribution infrastructure maintenance presents significant challenges for inspectors and assessors. Three of the main challenges are:
- There is a lot of infrastructure to inspect
- Assets are spread over a large geographical area and can be hard to access
- There are many distribution assets, each with their own unique degradation and failure modes.
Because of these challenges, traditional distribution practices have relied on a segment by segment prioritization or even a “run-to-failure” approach. However, emerging technologies may enable more routine inspection while managing costs. Automated Unmanned Aircraft Systems (UAS) operations have a greater ability to collect imagery than ever before. In addition, computer systems built on artificial intelligence (AI) are becoming increasingly capable of analyzing this inspection imagery. Utilities investigating opportunities to advance their distribution inspection practices could leverage these advances in technology.
While some utilities are beginning to integrate these technologies, there are questions related to the best approaches for deploying these technologies in automated inspections. For example, there is no industry consensus on the best way for capturing an image given the unknowns, such as number of images per structure, perspectives, and camera settings that produce the best inspection results. This project intends to develop image capture specifications in EPRI’s laboratories and validate those specifications through field data collection on host utility circuits.
Collecting the imagery is only half of the inspection process. Images must be reviewed in order determine any actionable information. Today, many processes rely on human review which can be time consuming and costly. The use of machine learning and AI for inspection imagery is an attractive data review solution. After the images are captured, computer systems could perform automated analysis utilizing AI to identify issues. This may reduce the time and cost to review inspection imagery. However, these AI systems frequently operate as a black box, meaning it is difficult or impossible to articulate how the system determines outcomes. For these systems to be deployed and trusted, a rigorous performance assessment will be needed prior to deployment. Even after the initial deployment, it will be important to continually validate performance against a known control set as a form of ongoing quality and performance control.
In the Advanced Distribution Inspection project, EPRI plans to quantify the performance of this new, automated method. Work will begin in EPRI’s labs where the team can develop, test, and refine overhead distribution inspection requirements. Then, the team will evaluate how well these specifications translate to the field. Field validation is an important step as the distribution environment poses unique challenges. Lastly, EPRI will label the field collected data to train and evaluate AI models. With the results of this research, utilities can make data-driven decisions, with confidence that they are applying the technology where it makes the most sense.
Public benefits from this research include the potential to reduce the cost of distribution inspection that may result in increased reliability.
Participants may benefit by gaining exposure to emerging technology and approaches, receiving guidance on image capture techniques, and understanding the current performance of AI defect detection systems.In addition, this work builds a foundation for futureAI research and accelerates the innovation of AI solutions.
EPRI invites project participants to engage in the following project tasks:
Develop Image Capture Guidance - Stage defects and capture imagery in the laboratory to develop, test, and refine overhead distribution inspection requirements.
Capture Imagery in the Field-Capture distribution asset imagery using UAS.These images will be used for multiple purposes, including AI training, validation of image capture guidance, and objective evaluation of analysis systems.
Validate Image Capture Guidance-Analyze imagery collected in the field to validate EPRI developed image capture specifications.
Evaluate AI Prediction Systems-Evaluate and quantify performance of automated image analysis systems. This includes developing a common taxonomy, exploring synthetic datasets, evaluating image labeling methods, and analyses of multipleAI models’ performance.
This research is expected to produce the following:
- Technical report describing the methodology for image capture during distribution inspection;
- Technical report describing the analysis and results of AI model performance for automated image analysis; and,
- Distribution inspection taxonomy and AI training datasets.
The non-proprietary results of this work will be incorporated into EPRI Distribution Systems R&D program, and made available to the public, for purchase or otherwise.
Price of Project
The project cost is $60,000.
This project qualifies for self-directed funds (SDF) and tailored collaboration (TC).
Project Status and Schedule
This research is expected to take 18 months to complete, with milestones, updates and results delivered throughout the project timeline.
Who Should Join
Utilities, inspectors, and AI leaders that are looking to make data-driven decisions on how best to deploy new technologies for distribution inspection.
For more information, contact EPRI Customer Assistance at 800.313.3774 (firstname.lastname@example.org)
Member Support Contacts
Brian Dupin, 650.906.2936, email@example.com
Barry Batson, 704.905.2787 firstname.lastname@example.org
What if I don’t use drones, what value is this?
- The value of this project is that collaborators don’t have to be drone experts. If you think you may use drones in the future, the results from this project can guide your decisions on how best to use them.
We don’t even do inspections, why would I care about this project?
- The goal of this project is to investigate a new method that may improve the current ROI to perform distribution inspection. But at the end of the day, this is research into an inspection option that utilities can deploy where it makes the most sense.
Will I get software that I can use?
- No, this project develops a methodology that you can use to improve or build your inspection program with UAS. This allows the research to be applied to many utilities and modified to fit your needs.
Will you do an inspection on my equipment as part of this project?
- This project does have field inspection tasks to it. The objective of those field tests are to validate the image capture specifications and methodology that EPRI developed. Also, the imager from this field collection will support AI training sets.
Who is the internal customer that would benefit the most from this project at a utility?
- It’s likely reliability and maintenance managers that are looking for the lowest cost inspection solution.
Do I have to contribute data as part of this project?
- Participants do not have to contribute any data to the project. For those that may want to contribute, EPRI has a process to protect and anonymize utility data for the benefit of collaborative learning.
What are the time requirements on a utility advisor?
- EPRI encourages a minimum participation through quarterly webcasts and deliverable review. With lab and field testing, EPRI commonly invites participants to witness the tests. However, there is no requirement from the utility advisor.