Supercharging utility inspection operations with AI image analysis; pricing and benefits.

The topic of AI image analysis is popular.

We work a lot with AI at Scopito, but because every algorithm is customer-specific, trying to paint of picture of the generic benefits, is hard.
However, this article is an attempt at providing you with an idea of what a large-scale AI image analysis project could look like.

These are insights from a pilot project with the goal of inspecting 800.000 poles yearly, powered by a new AI image analysis algorithm. We at Scopito are working on this in collaboration with a large south American utility.

Goal

The utility’s goal is to reduce downtime and extend asset life, by inspecting 800.000 of their 5-6 million poles and towers on a yearly basis.

Solution

As with most large-scale inspection projects involving AI image analysis, this one starts with a 6-month pilot project, at the end of which, the utility will be ready for full-scale operations.

The first two months are allocated to data collection. This includes preparations, flight planning and the capture time. This is also accounting for bad weather and other obstacles.

Once data is captured, the manual analysis will take 2-3 weeks. The algorithm will be trained on the annotations created during this time.

When the training-data is ready, it will take 2-3 months to train the algorithms, after which it is expected to have an accuracy-level of above 90%.

Scopito will be developing the algorithms based on images of 0.04% of the utility’s total assets.

The faults we are looking for include, but are not limited to:

• Insulator damage
• Encroaching vegetation
• Broken cross-arms
• Bird’s and bee’s nests
• Guy wires sagging
• Woodpecker holes
• Top cap issues
• Rotten wood
• Thermal anomalies
• Failing concrete

The areas of interest will depend on the faults found during the manual annotation.

The implementation of AI image analysis algorithms will not only allow the utility to scale their operation in the coming years, but also extend the lifetime of their assets significantly as well as build a digital representation of their asset fleet.

Increased analysis speed

The speed of AI image analysis is essentially what makes it possible to complete an operation of this size. Having a human workforce complete analysis of pictures of 800.000 structures, would simply not be feasible.

It is expected that Scopito’s algorithm will be able to analyse at least 1.000* images/hour. At a fast pace, a human will be able to analyse 175 images/hour in Scopito.

If we assume there are an average of 2 images per structure, that amounts to 1.600.000 images/year.

This would take a human 380 days to complete, assuming he works 24/7 without breaks.

In comparison, it would take Scopito AI 66 days.  

*AI processing capacity is easily scaleable without introducing significant costs.

Added accuracy with AI

In addition to working faster, AI image analysis provides consistent results. It cannot be subject to distractions, fatigue, or social distancing, the way humans are.

Depending on the quality and consistency of data it is trained on, an algorithm can become extremely accurate in its annotations. The accuracy also grows with the amount of data it processed, and is expected to reach >95% once the operation is running at full capacity.  

Extend asset lifetime

AI image analysis contributes to extending the lifetime of utility’s assets in more ways than one.

Because it works faster, and more reliably than its human counterparts, it can detect issues early on, which means maintenance can get to work as early as possible.

In time, as the algorithm has processed enough data, it will be possible to identify trends and in term predict the likelihood of future failures on specific assets. This is most commonly referred to as predictive maintenance.

Visual fault presentation

These benefits are not generic for all types of AI image analysis, but rather they come from working in a visualization platform like Scopito.

When utilities, like this one, are managing assets at scale, visualization starts to become exceptionally important. Getting an overview of 5-6 million assets, without a visual representation, is unmanageable. Once you add AI image analysis to that mix, the amount of data only grows, and the clear overview becomes harder to find.

That is why Scopito is dedicated to giving our clients the full overview at every level; from the color-coded map of all their assets, to the overview line-map with encroaching trees and faults clearly marked, all the way down to the asset-view.

We specialize in presenting our client’s heavy data in the most logical and intuitive way, and without that, AI image analysis isn’t worth its weight in gold.

---

---