A few weeks ago, I had the opportunity to meet the co-founders of buzz solutionsa company that provides a platform for utilities to analyze, store and manage the growing load of high-resolution inspection imagery of transmission and distribution assets.
The CEO, Kaitlyn Albertoli, studied international relations, finance and business at Stanford, ran a non-profit organization around sustainable food with 60 people overseeing the needs of 300 people. Albertoli is from San Clemente, a seaside town in Southern California, and the San Onofre nuclear plant was located there. Much of its upbringing has been the controversial and charged attempts to level it up and ultimately shut it down. She became fascinated with energy and in particular the renewable energy space.
The CTO, Vik Chaudhry, comes from a diametrically opposed environment, New Delhi in India. His undergraduate at Delhi College of Engineering was in Civil and Environmental Engineering. In his senior year, he built a quadcopter from scratch with pollution monitoring sensors and flew it around hotspots in New Delhi. He found that it was often over 500 AQI, the equivalent of smoking 60 cigarettes a day. This air quality could be directly linked to a lack of electrification. He moved to Stanford for a master’s degree focused on the energy industry, energy engineering, and the application of AI and machine learning to demand response and energy efficiency, with a large project for the assessment of the location of wind farms by drone. Cisco picked him up to run ML and AI for a few years as he and Kaitlyn started Buzz Solutions.
They met in a 2017 Stanford class, now called Venture Creation for the Real Economy. The first three weeks were spent setting up the go-to-market strategy, the next three weeks building your 5-year projection, and the last three weeks building your pitch deck. Stanford brings in CMOs, CEOs, and VCs every three weeks to provide input.
They originally focused on siting wind farms and were soon directed to drone inspections of wind farms. When they spoke to energy companies, the narrative was always the same: have you seen what’s happening with more frequent inspections and the use of drones there? This led them to leverage lists of Stanford alumni to talk to inspection crews from 35 major power companies. They found they were capturing 10 times more images than they historically were much more frequently and using far more drones. They were getting hundreds of thousands of high-resolution, high-zoom images with plans to expand their transmission and distribution infrastructure to millions each year.
There is far more complexity, far more components, and far more observable problems in electricity transmission and distribution infrastructure than in wind farms. Serious faults could cause downed power lines, sparks and possibly wildfires. The concerns were leading to regulatory pressure for increased inspections and maintenance. And, of course, the infrastructure is aging.
But the data-evaluation process was manual, with linemen and trained engineers watching their computers instead of solving problems. It was a machine learning opportunity. They started trying to help utilities identify hotspots and points of failure that could lead to major outages. And that was before the 2017 wildfire season that damaged so many lines in California. They were there, they were ready and they were able to.
The increase in inspections is motivated by two major factors. The infrastructure is aging, with the average age of components such as insulators being over 40 years and limited asset tracking. Much of what utilities are trying to figure out is what assets they have, where, how old they are, and how fast they are degrading. With the arrival of renewable energies on the network and the climatic impacts, the components undergo increased constraints resulting in increased degradation.
As renewables come onto the grid, retrofitting is a requirement and identifying which components need to be retrofitted is a challenge. But you don’t tend to hear about the network unless there’s a problem like wildfire outages, power outages, or outages caused by east coast storms. The network has been in the spotlight in recent years and particular attention is being paid to this aging and critical infrastructure in the United States.
It is not possible to train 10 times as many people to process 10 times as many photos. Utilities were struggling with the problem. They were trying to hire engineers and linemen to analyze the data, but the delay between imaging and analysis was increasing. This meant an increase in risk, as degrading components degraded further in the meantime. Public services are increasingly interested in solutions for managing and analyzing big data.
Buzz Solution’s market timing was excellent. High resolution cameras on smaller, inexpensive drones allowed for much more footage, cheaper and safer. Many drone players have entered the market over the past 5-8 years, learning from their mistakes. Like DJI drones, they’ve all gotten smaller, more powerful, smarter, and the sensors have drastically shrunk. Fist-sized sensors are now thumb-sized. They are highly accessible platforms that can be controlled by anyone.
And in 2016-2017, open source visual recognition machine learning toolkits became very actionable. Google launched ImageNet and ResNet which quickly became a standard backbone for image processing solutions. The fact that the algorithms are open source is essential, but there has been great innovation in computing, with inexpensive cloud-based graphics processing units (GPUs) accessible in seconds with just a few clicks. The result was an actionable toolkit without a PhD in machine learning and a decade of use.
Historically, line inspections were done with helicopters or by walking the lines with binoculars, and that was only a quarter or a third of the lines a year. There are also on-demand inspections for major storm hazards, storm damage assessments, high winds, or wildfire hazards. Utilities identify sensitive areas subject to higher risk or where critical failures could occur, and these areas are inspected more frequently. There are both high-level overviews and more granular inspections. Obviously, more granular and more numerous images require more evaluation.
For a single transmission tower, there may be 30 to 90 frames per tower, usually in the 40-60 range. Even for distribution poles, the wooden or concrete ones delivering electricity to buildings, there are 4-12 frames captured, usually in the 8-10 range. And there are many towers. There are about 120,000 miles of high-voltage transmission lines in the United States alone, with pylons roughly every five hundred yards, which suggests about 600,000 pylons. The distribution network is much larger, with approximately 5.5 million kilometers of cables and many more poles.
The volumes of data are staggering, the need to inspect them is increasing dramatically, and there aren’t enough people to do the job. Enter Buzz Solutions. The company is now saving 50% in image evaluation time and is trending towards 80% in effort savings.
The next era of inspections emerging is that of neural network chips on drones, initially for automated flight around towers, identification of protruding components and image capture, and eventually immediate fault detection. the Skydio drone, for example, already uses an ML chip for its autonomous flight and imaging. With FAA regulations easing and more companies gaining approval for operations beyond line-of-sight, there is a lot of work in this space.
Buzz Solutions is well situated for this. They view their product as an AI orchestration platform. It supports multiple data streams and produces results. An extended use case might be a processing unit for a drone in the field. Various utilities and organizations are testing this now. The first step is to recognize the assets with the drone following the power line itself, flying around the pole and imaging them. The next generation will be heavier machine learning models with problem recognition and circling the poles to inspect components. If an insulator or conductor is damaged, the drone could circle it and take more images, and potentially send an immediate alert for priority repair.
Buzz Solutions is responding to customer requests for this advanced feature and will roll it out when the FAA is more lenient on range and more compute on the drones themselves. Vik’s love for building and flying drones will be satisfied once again.
Another use case being explored is fixed-wing vertical take-off drones that have longer ranges and higher resolution cameras so they can capture higher resolution images at higher speeds. Quadcopters will still be needed for in-depth, granular inspections, but many drone companies and third-party utilities are investigating different form factors of UAVs for different use cases. It’s one of many places where drones are replacing helicopters more cheaply, part of the shrinking market for manned rotorcraft that’s disrupting the industry.
So ends the summary of the first half of my conversation with Vik and Kaitlyn, the co-founders of Buzz Solutions. Stay tuned for part two, coming soon.
Do you appreciate the originality of CleanTechnica? Consider becoming a Member, supporter, technician or ambassador of CleanTechnica — or a patron on Patreon.