Robot dog Spot to transform into Sparky with AI enhancements
Image courtesy Avangrid
Iberdrola-owned energy company Avangrid is coordinating a pilot project with Levatas and Boston Dynamics to advance substation inspections, using an advanced version of Spot the robotic dog, now using artificial intelligence (AI) and nicknamed Sparky.
Avangrid, which owns and operates eight electric and natural gas utilities, serving more than 3.3 million customers in New York and New England, said in a release that the project will deploy the robot dog to complete visual and thermal inspections at two substations of its Connecticut subsidiary, United Illuminating (UI).
The project will take place at UI’s Singer and East Shore substations and test a variety of AI models, developed by Levatas, to read analogue gauges, record thermal images and detect damaged equipment.
To do so, the robot dog—nicknamed Sparky by the Avangrid team—is outfitted with a camera that has a 30 times optical zoom and an infrared camera capable of taking thermal readings to compare the transformer and breaker phases.
There is also an option to attach an acoustic sensor that can detect, locate and visualise changes in sound signatures, malfunctioning equipment and other noise anomalies in real time. The robot also has a core processor to enhance autonomous navigation and communications.
At the Singer substation, the project will test how quickly and accurately the robot can detect and read several of the substation’s analogue gauges. At East Shore substation, the project will test the robot’s ability to inspect transformers, circuit breakers and capacitor banks.
Have you read:
Spot the robotic dog sniffs out faults in National Grid’s interconnectors
High voltage birdspotting: Elia uses drones for collision prevention
Initially, the robot will be controlled by an onsite operator using a tablet, which can be used to both drive the robot and create autonomous missions. Avangrid is also working to install software that would allow for remote operation as well.
“We’re very proud to once again be among the utility industry pioneers pushing the boundaries of technology,” said Pedro Azagra, Avangrid CEO.
“Last year, we announced the establishment of an in-house team to build unique machine learning models to increase reliability. With this pilot project, we are now exploring using AI in another aspect of our business where there is great potential to bring high value to our customers and stakeholders. This type of innovation will help us be more efficient, target our investments and increase reliability for our customers.”
“It’s amazing to see this technology, which was inconceivable a few years ago, in our hands bringing value and benefits to our customers,” added Catherine Stempien, Avangrid Networks President and CEO.
“While there will be many benefits, most important is that we expect Sparky will increase the frequency of our substation inspections so that we can see how our equipment is functioning during different seasons, times of the day and energy loads. With this increased amount of data, we will have the potential to proactively identify unknown issues and trends before they cause outages that impact our customers. This is a great example of innovation and technology helping us do more.”
From Spot to Sparky
According to a spokesperson from Avangrid, Sparky improves on Spot in that it will be trained to live and operate within an energised substation environment and will be taught to identify abnormal conditions within that space.
Using machine learning algorithms, Avangrid has collaborated with Levatas to create an insulator inspection model that will use one of Sparky’s cameras to look for, identify and report insulator anomalies.
It will have the ability to digitise analogue gauges to keep track of important equipment health indicators and create alarms and notifications.
Sparky also has a thermal camera that it will use to inspect for thermal hotspots on substation equipment, giving operations personnel the opportunity to manage those issues promptly and efficiently.
The spokesperson added that, as with AI models, they expect the machine learning algorithms to improve with time, making the solution more valuable.
“The gauge reading model, for example, has been successful in flagging health indicators out of pre-defined thresholds and acceptable range. We are pleased with the results so far and are excited to continue with Sparky’s integration into our day-to-day operations and maintenance programs.
“Currently, we do not have a specific timeline for deployment of Sparky. At this time, we are focused on developing its application model, including its operational space, data management and communication protocols.”
