Representatives from Carbon Re discuss the development of their company’s artificial intelligence (AI) driven approach to lower cement sector CO2 emission
Global Cement (GC): Please could you introduce Carbon Re?
Sherif Elsayed-Ali (SE-A): Carbon Re was formed in October 2020 to develop artificial intelligence solutions that help cement plant operators reduce fuel-derived CO2 emissions. We believe that, while cement and concrete are essential for the future, the associated high levels of CO2 emissions will not be an option.
Carbon Re is a company born out of the pandemic and I think that the past 12 months have been something of a wake-up call. There were many warnings about such a pandemic and yet humanity was badly prepared. Climate change is much the same. We have also been warned about climate change for a very long time, yet meaningful action is only just starting to take place. Carbon Re wants to help cement producers take steps to reduce their CO2 emissions now, while helping them with their bottom lines too.
GC: What is the Carbon Re approach?
Daniel Summerbell (DS): Carbon Re is developing an approach that uses artificial intelligence (AI) to reduce CO2 emissions from cement plant fuels. It is based on reducing the mass of CO2 emitted per useful heating value (kgCO2/UHV), essentially the CO2 footprint of the heat that is actually delivered to and used by the process. For example, heat used to dry a wet fuel, for example biomass, is not considered. Neither is the fuel that heats the excess air, the amount of which can vary substantially between different types of fuel. You may have to run a lumpy refuse-derived fuel (RDF) with more air, for example, than a finely ground coal. By considering the kgCO2/UHV value we can objectively analyse the actual reduction in CO2 of using a specific alternative fuel, rather than simply assuming that they lead to lower emissions.
By itself, kgCO2/UHV has the potential to get lost in the pantheon of cement process indicators and it can be fairly counter-intuitive. What combining kgCO2/UHV with AI does, however, is remove layers of complexity and allow priorities to be identified and brought to the attention of plant staff. This way, Carbon Re can help cement plants to identify which of their familiar
parameters should be improved, which will, in turn, lead to a lower kgCO2/UHV.
GC: What stage is the development currently at?
DS: The approach is based on initial research performed at cement plants operated by Hanson UK, part of Heidelberg Cement. We are currently developing the approach for wider use and are negotiating agreements with producers around the world.
GC: What information is fed to the AI?
SE-A: Carbon Re’s approach makes use of existing process parameters, including, but not limited to, O2 levels, air flow rates, cement chemistry, fuel parameters, production rates and many more. Some of these have very clear relationships with kgCO2/UHV emissions. Others, for example time of day or the personnel working a particular shift, may also have relevant relationships with this indicator. However if the AI indicates that a particular parameter could be important, we can look in more detail to establish whether or not there is a relationship and if so, if it is causal. Of course, relationships between process parameters and kgCO2/UHV emissions will also vary from plant to plant.
What’s clear is that the more data you have, the greater the opportunities to reduce kgCO2/UHV. Thankfully, many cement plants are equipped with sensors that record vast quantities of data, more than can reasonably be acted upon by plant personnel. They can end up with long lists of priorities that may, or may not, be in a helpful order. Carbon Re’s AI-based solution is a low-cost way to properly identify the best steps towards reducing fuel-derived CO2 emissions from cement plants.
GC: How does the AI handle the incoming data?
SE-A: Our approach is led by our co-founder, Aidan O’Sullivan at University College London. It is based on deep reinforcement learning, an area of AI that has seen huge progress in recent years as the methodology behind things like DeepMind’s famous AlphaGo.
GC: What does the plant have to install to make use of Carbon Re’s system?
SE-A: Carbon Re’s system is based in the cloud, so there is very little, if any, ‘infrastructure’ to install at the plant. The cloud-based system analyses the incoming data and identifies specific, quantified, recommendations for the operators, which it presents as a priority list on an intuitive browser-based dashboard. We use base AI models developed specifically for cement processes, with a layer of fine tuning for each individual plant.
DS: The AI is intended to be used in the plant at three levels. The first is at the operator level. We imagine a situation where the AI indicates priorities for discussion, say at the plant’s routine morning meeting. The relevant staff would discuss whether or not these are realistic and, if so, how to achieve them. The operator would then be set targets for the day. The following day, there can be feedback on how each of the changes affected the plant’s performance with relation to kgCO2/UHV. Did this have the desired result? If not, why not? The results are fed back to the AI and the system updates its knowledge base.
Perhaps there is a priority that couldn’t be achieved. In this case, we start to look at the second level of use: Why couldn’t the operator achieve the priority? Maybe they simply ran out of time... or could there be some underlying problem with the plant? Such issues can be reported to plant management via a second dashboard on a monthly basis. If a priority keeps being recommended by the AI, it shows that the priority has greater value to the plant from a kgCO2/UHV perspective than we might perceive. It could even be as simple as a sticky valve. Such a ‘small thing’ might get lost at the bottom of a traditional list of priorities and it could take months to resolve. However, AI might show that fixing it now could bring unrealised benefits for the plant.
Beyond this, the third level looks at potential to alter the plant itself. The AI might say, ‘Keep parameter X constant at level Y.’ It may be a winning recommendation in principle, but cement production is inherently unstable and it could be that the operator finds it impossible to actually keep parameter X in check. This could be due to a physical shortcoming or bottleneck that is not intuitively obvious to the staff.
The possibilities of the third level then become really exciting. If a digital twin model of the plant has been established, the staff can look at the problem far more deeply and forecast the effects of making physical changes. What will it be worth to the plant, both in terms of return on investment and kgCO2/UHV? Too often such projects have return on investments that are fairly vague. In contrast, the Carbon Re approach informs the project manager of the potential benefits of a given project in a far more comprehensive manner. Appropriate actions become clearer.
GC: Could the AI control the plant?
DS: We have deliberately stepped away from the AI controlling the plant at this stage. The level of complexity this would add is enormous and we are not sure that the sector is ready for this approach just yet.
GC: It seems that the Carbon Re’s approach can unmask previously unknown correlations. What are some of the more surprising ones?
DS: It turns out that where you burn your fuel has a major influence on kgCO2/UHV, which had not been picked up on by the overall metric previously. For example, in one plant we found that if you increase the amount of secondary recovered fuel (SRF) in the main burner past a certain point, you actually have to burn more coal to heat the material up so it will burn. This is a case of increasing alternative fuel use not bringing a net CO2 benefit. In fact, it can make the situation worse!
GC: What benefits would a typical plant expect to see when it uses Carbon Re’s AI solution?
DS: We expect that a typcial plant would be able to achieve a 16 - 19.5% reduction in fuel-derived CO2 emissions, which would represent around 5 - 8% of the total. How this translates to cost is hard to assess, as fuel prices vary and, as we have established, the relationships between kgCO2/UHV and fuel are non-linear. We have also been working on a grinding optimisation AI-led system that can reduce the cost of the energy needed to grind materials within the plant by up to 8%, which could provide the basis for an entire separate article.
SE-A: Highly-qualified AI experts are thin on the ground. Even the largest and most ambitious cement producers would struggle to develop a functional and effective AI system in-house. Therefore, our aim is to be the provider of that AI expertise for the cement sector. We want to bring both environmental and cost benefits to producers around the world.
GC: What is the timescale for commercial use?
SE-A: We are working, as discussed, with our early adopting partners right now in a number of locations around the world, including in North America, the Middle East and Asia, as well as countries in Europe. We hope to launch a commercial product to the general market in early 2022.
GC: Do you think that the rapidly-rising price of emitting CO2 under the EU Emissions Trading Scheme will encourage uptake of this solution in Europe in particular?
DS: While EU ETS prices have risen dramatically, they are now consistently over Euro40/t, there are still quite a lot of free allowances available to cement producers. This dampens the effect of rising prices at present. However, as the number of free allowances comes down, EU ETS prices will go from being an area of concern to a serious financial consideration, as well as those affected by the recently-announced Carbon Border Adjustment Mechanism (CBAM).
So, in reponse to the suggestion that Carbon Re’s solution is necessarily a Euro-centric concept, I don’t agree. This is because the solution allows you to both reduce CO2 emissions and cost, be it from optimising efficiencies in the fuel mix, air flow, less maintenance or any one of many other parameters.
I do agree that, in the longer term, Carbon Re’s approach will go from providing an incremental improvement to becoming an essential add-on for a cement plant. We want to become established now and embed our solution in the decision-making process, so that we can help cement producers get over that transition.
GC: What are the biggest hurdles to implementation in the coming years?
SE-A: As you just touched on, we want to implement this solution across the world, not just in one place. This means that we will have to apply the system across different sized producers that work to different regulatory requirements in different languages using different measurement units, that work with different types of fuel to make different types of cement for different markets. This will challenge our approaches for sure but it’s something we look forward to working on as we grow.
GC: What is the greatest opportunity for the company over the next few years?
SE-A: There is increasing action by governments to reduce CO2 emissions, as demonstrated by China’s commitment to CO2 neutrality by 2060, the EU’s commitment to zero CO2 by 2050 and the recent policy changes from US President Joe Biden. The use of emissions trading schemes will only grow globally and pressure from the public and investors is now clearly geared towards sustainable investment and business practices. As cement is essential for future growth, this will require low- or zero-CO2 cement and concrete products. We can facilitate this transition, while improving cement producers’ bottom lines too. We look forward to helping them.
GC: Thank you for your time today, gentlemen.
SE-A/DS: You are very welcome indeed!