Airborne Damage Assessment Tool (AirDAT)

Okay, so I saved one of my favorite things for the last use case, and this is storm damage assessment. And I will tell you... this is a, you know... I'll give you a little context of the problem that we're trying to solve.

In our service territory, we deal with a lot of heavy storms. So this year alone, we had two major hurricanes. We had Hurricane Beryl that ran through Texas, knocked out for us about 250,000 customers... but in the broader area, there was over two million customers out across other utilities.

More recently in Louisiana, we had Hurricane Francine, that knocked out over 350,000 customers. Well as you're responding to a storm, there's a lot of things that have to happen.

The first thing you have to do is run a damage assessment. You have to really put eyes on what... what of your asset is actually broken so that you can prioritize it appropriately so you know what needs to be fixed and the sequence of fixing it.

That does two things. It helps you inform your customers more accurately in terms of when they can expect power back on, and it helps you more effectively put your crews to work, especially if you're bringing in crews from outside the company.

So that's the challenge we've been struggling with, and historically, most utilities including Entergy, we struggle in the space. It'll take days to run a storm damage assessment.

During that time — and hurricanes hit in the height of summer — so during that time your customers are without air conditioning, they're without power. It's a situation that gets exponentially worse day and day out.

So the problem we were trying to solve in that case is, how do we shrink damage assessment from days down to hours?

This is the concept that we're working on. We call it AirDAT, which is air damage assessment. And what we're trying to do from this is take what has historically been a ground-based, heavy manual assessment process and lift it up into the air. In other words, use fixed-wing aircraft, equipped with lidar, as well as capturing still imagery, that will be sent out on set patrol areas.

So think about an aircraft that's running 2,000 feet in the air, that's doing anywhere from 100 to 150 miles an hour. You can capture a lot of territory very quickly. Particularly if you look... if you go a couple of miles outside of New Orleans, you'll notice there's a lot of spaces that aren't very easy to get to, there's a lot of swamps, there's a lot of places that vehicles, particularly after a storm, can't get to. From the air, you can capture that very quickly. And as you employ multiple aircraft at the same time, that gives you an even greater ability to cover the service territory.

So we start with a fixed-wing aircraft that's capturing both lidar and still imagery. From that, we feed that imagery into a data model. The model, based on AI, will actually take the imagery and identify assets that are down. So think, down power lines, vegetation issues, poles, transmission lines. It'll produce damage assessments from that and ultimately delivers a full damage assessment file.

Now the aspiration with this — and we're still working on it — is that this process, essentially the second the hurricane's out of the territory, you've got a plane in the air. It's capturing that data, it's running through the model in real time. As that plane lands, we are delivering a full assessment file in hours in what used to take days.

So while we're still working on it, we've come little ways here, and I wanted to show you a little bit of what we found.

Hurricane Francine, which came through here a little while ago, we actually ran a proof of concept of this. And what we were trying to do is really test the technology itself, to find out how accurate is it and can we build a process around it?

Now this is a little bit hard to see I'm sure, but let me walk you through what you're looking at. On the left hand side of the chart you're looking at, essentially, is our data model. And all those dots represent lidar imagery. So the white is the ground, the green dots are actually power lines. That pole that you see the middle, well that's just that. It's a pole. But it's actually a modeled pole. In other words, it's taking GIS data and it's saying your pole should be here and it should look like that.

Now keep in mind, all this imagery, that's happening from 2000ft up, so the computer is actually showing it as you're walking on the ground.

Now what's interesting about that, is while the model is saying your pole should look like that... if you look at the very base of the pole, and it's probably pretty hard to see, but you're gonna see a small string of blue dots. Those blue dots are coming from the base of the pole at a 45 degree angle... unfortunately that's our actual pole. So while the model says it should be upright in that location, the lidar is capturing it and saying, and the model is interpreting it and essentially saying, sorry your pole is actually down right now and it's leaning at a 45 degree angle. And as you look at that string of dots on the right, those are actually our power lines. So that's why the power lines are not connected to the pole in that case. Now it's really hard to see it in the photo, but the photo on the right is actually the still imagery that's being taken by the plane at the same time that we're capturing the lidar. And the still imagery is used to confirm what the model is saying, which is you've got and down pole and you've got wires that need to be fixed.

So, another example of this, and this one shows off a little bit better. The left hand side of the chart shows two poles, in the middle of it you're gonna see a sea of purple dots. Those purple dots, they're actually vegetation. That represents a tree that has fallen on our lines. If you look to the picture on the right, the still imagery is capturing and confirming that you can see a big fat tree sitting on our wires.

So, again, think... think about a storm that's coming in, that's hitting multiple parishes at the same time, it's impacting hundreds of thousands of customers. You've got an ability to send up a plane or multiple planes, to capture this data at a hundred miles an hour, as opposed to individuals on the ground that have difficulty actually getting into the spot. You have the ability to quickly shrink the assessment time and get very actionable data into the hands of our crews so that they can go out, generate work orders that are prioritized, that will help get our power back on quickly.

And the other benefit of this is you've got a much more complete assessment early in the process. So we can proactively communicate with our customers, that hey, it may take X days to turn your power back on. And while that may not be a great answer, it's a better answer than a wide range where they can't really make a decision based on that.

So, it's still early in the process right now. We have a viable proof of concept, it showed very promising results. Our work in front of us next year is to build an end-to-end process around this, where we can actually turn this to drive the outcomes that we're looking for. Our aim right now is to have this in place by next hurricane season.

So, with that, I’ll leave you with a couple of thoughts and comments. The first is... hopefully you got a sense that there is real value to be had from AI solutions today. We're really excited about the technology. It works. We're using it to solve real-world problems. And, most excitingly, it's expandable. It's something that can scale to other use cases very easily. Right now our problem is not identifying use cases, it's really prioritizing use cases.

It’s a very exciting time to be in the industry right now, and we’re excited about what we have in front of us. With that, I appreciate the time.