Using computer vision to increase the efficiency of asset inspections
Utilities spend billions a year on inspections, gathering data and information to identify capital replacement, and maintenance work. The find rate for most inspections are relatively low, meaning that much of the time - and significant cost - is spent manually reviewing photos that do not yield work. AI can help utilities speed up the manual review process, leading to faster work initiation, improved safety and reliability, and less wasted time and operating and maintenance (O&M) expenses.
Every utility is required to inspect their transmission and distribution system. Some utilities do annual inspections (mainly for high risk areas), while others will inspect their entire grid over 3-5 years. Combined, U.S. utilities have 200,000 miles of transmission lines, 1.2 million transmission poles, 5.5 million miles of distribution lines, and 186 million distribution poles scattered across the United State’s 3.8 million square mile territory. This means finding asset defects are like looking for needles in a haystack.
Most inspection cycle “find rates,” or the rate at which inspections yield assets with defects, range from 5% to 20%. The more often a utility inspects their assets, the lower the find rate. Utilities doing annual inspections could have a find rate of 5% or less; meaning manual reviewers spend 95% of the time reviewing data that do not have defects. Reviewers are potentially wasting time as they sift through thousands, or even hundreds of thousands of photos before they find the assets with defects.
Computer vision can help utilities prioritize review - resulting in faster defect identification and cost savings
Machine learning, or more specifically computer vision, has the potential to reduce the time that it takes to identify asset defects as well as significantly reduce cost.
Software solutions such as Noteworthy AI Inspector employ computer vision models that are specially trained to identify issues such as corrosion, leaky transformers, broken and flashed insulators and more. Armed with these tools, utilities can prioritize their review on the problem assets that the models identify so that review teams can focus on the 5% that matters and waste less time on the 95% that don’t.
Utilities can spend over half of their overall aerial inspection budget on reviewing inspection data. But because computer vision models can prioritize and reduce the overall amount of photos that reviewers need to analyze, utilities can in turn reduce the number of hours required to analyze inspection data which can yield real, hard dollar cost savings.
Faster defect identification will help utilities improve safety and reliability
Most manual reviewers analyze photos and video as they come in from ground or aerial survey missions - not according to the asset’s risk level, or the probability that the photo yields an asset defect. Utilities therefore end up with a significant backlog of photos, video, and data to review.
An unfortunate side effect of this backlog is that issue remediation suboptimally isn’t scheduled as quickly as possible, which can lead to both safety and reliability concerns.
Because computer vision can shorten the time spent on review, human reviewers can more quickly generate work orders so that technicians can more quickly repair asset defects. And faster repairs most certainly improves grid safety and reliability.
Aerial inspection “find rates” are quite low - ranging from 5% to 20% - meaning that reviewers may be spending up to 95% of their time reviewing inspection data on assets that don’t actually have any defects. Next-generation software applications that employ specially-trained computer vision models can help quickly to surface issues so that reviewers can prioritize their work and reduce the total number of hours spent on review. Once the defects are confirmed, remediation work orders can immediately be initiated, resulting in improved safety and reliability on the grid.