Reducing the carbon emissions of the hard-to-abate sector with AI
Computing hears how reinforcement learning can make the hard-to-abate industries more energy efficient– with a potentially transformative impact on global CO2 emissions.
Yesterday, Computing set out in this article how decarbonising cloud is contingent on the decarbonisation of construction materials - particularly steel and concrete. One start-up seeking to do just that is Carbon Re, an AI and climate tech company spun out of Cambridge University and UCL. The company recently raised £4.2m in seed funding and is building the world's most advanced AI platform for industrial decarbonisation, leveraging reinforcement learning to reduce the emissions of critical industries.
Carbon Re aims to reduce the carbon emissions created by the foundation industries - cement, glass, metals, paper, ceramics and chemicals. What unites these industries is the necessity of immense temperatures for manufacturing. They are referred to collectively as "hard-to-abate," because they provide building blocks of economic development along with about 22% of global CO2 emissions.
Aiden O'Sullivan is CTO and Co-founder at Carbon Re. He's also an academic and pioneer in the application of AI-to-energy systems. He leads the Energy and Artificial Intelligence Lab at UCL and is also a fellow at the Alan Turing Institute and Programme Committee Chair in AI and Climate Change at the Centre for AI & Climate. He sets out the fundamental problem he and the team set out to solve.
"You can't just electrify everything because you need these really high temperatures to get these chemical reactions. What we want to do is facilitate the materials transition. How do we make these materials in a way that reduces carbon emissions? We're taking that approach to steel and glass but starting with cement which is the largest emitter in this space."
Transforming a reactive model
From the perspective of a customer, the Carbon Re Delta Zero SaaS platform is very easy to use. An API connects the platform to the plant so data is fed into what is effectively a bespoke live digital twin of the plant. No software is installed on the plant side. The AI agent then makes recommendations for optimisation on a minute-by-minute basis. Conditions and inputs in the cement industry are highly changeable, but the agent adapts. O'Sullivan explains why cement manufacturing is so volatile.
"It's an incredibly low margin industry. For example, if the price of Indonesian coal changes they might buy Australian coal. That fuel travels further so conditions change. Maybe it gets wet, maybe it's a different kind of seam. The energy content of that will change. Putting two tonnes of coal into the kiln one day might be a very good idea. The next day that might be totally insufficient. Or it might be far too much.
"These industries are reactive. The way it would manifest is you would put in the two tonnes of coal, you see what happens and four hours later, you make a decision about whether that was the right thing to do and then decide what to do next time. What our digital twin allows them to do is to simulate the impact of these decisions in real time. Our reinforcement learning agents will make a recommendation as to the right settings and then we keep monitoring that process as we go along. It changes a reactive model to a predictive one."
Reducing fuel consumption reduces carbon emissions, and the quantities involved are so immense that small changes can have a considerable impact.
A small change in fuel consumption leads to huge savings of carbon emissions because they burn the dirtiest fuels, the most energy dense fuels to get those high temperatures."
How has such a traditional industry responded to AI?
"Because it's an engineering and technical discipline, they're often quite interested in trying new technologies," says O'Sullivan. "There's been quite a lot of interest in what we want to do. We're running three trials with plants all over the world and are in discussions for plants in the UK and Europe."
Climate tech hasn't been ambitious enough - yet
There have been challenges building the service. O'Sullivan acknowledges that getting good quality data out of factories and industrial plants isn't easy.
"It's something you have to overcome to start solving the problem. We needed preparatory learning about industrial computing systems, distributed control systems, the different ISOs and the different standards that they have. This was the groundwork that enables us to then do the AI work. There's a huge amount more in terms of complexity that we're dealing with just in terms of getting the data."
ChatGPT this is not, and not just in terms of the complexity of data training the models. The ethics of generative AI are complex, but the work of O'Sullivan and Carbon Re is the one of the most powerful examples of the benefits that the reinforcement learning branch of AI can confer on all of us. Making the 'hard-to-abate' industries more energy efficient not only saves those industries money but could significantly reduce global carbon emissions quite fast. For the industries themselves it looks like the proverbial no brainer.
"Fuel costs are about 25% of overall costs for these industries," says O'Sullivan. What we're planning to do is improve the energy efficiency of the process. There's no upfront investment. It's purely digital. We want to scale to having gigatons of impact. For that we need scalable, easy to deploy products that we can get into a cement plant in six weeks. There are about 3000 cement plants in the world so there's 3000 problems to solve."
Despite the enormous commercial potential of Carbon Re, O'Sullivan says he will define success by how many companies follow them into the space. He is cautiously critical of the AI community as a whole for its lack of ambition on climate.
"The AI community in general hasn't done enough on climate change. We've been focused on solving problems that we know we can solve rather than engaging in problems that actually need to be solved. There's so many startups coming into the computer vision space because you more or less know that the technology will work. There's not that ambition to take AI to a space where you're not sure that it's going to work. That's why I want to see more ambition in climate tech. If Carbon Re is a success it will inspire other companies to do similar things. That's what I see as us being a success; not by taking over the world but by seeing more companies like us in the space solving other real problems.
"There's an awful lot to be done."
You may also like
/news/4338523/tatas-uk-gigafactory-project-takes-major-step-forward
Components
Tata's UK gigafactory project takes major step forward
Sir Robert McAlpine to build multi-billion-pound factory
/news/4336662/kings-speech-promises-regulation-most-powerful-ai-technologies
Legislation and Regulation
King's Speech promises regulation of 'the most powerful AI technologies'
But no specific AI bill
/podcasts/4333508/national-grid-analogue-digital-ctrl-alt-lead-podcast
Public Sector
National Grid is turning analogue to digital - Ctrl Alt Lead podcast
'We can't do what we've always done, just more efficiently'