Using AI to achieve sustainability and reduce carbon emissions


The drive is on across the world to reduce our carbon emissions and move toward a net-zero future to mitigate the effects of climate change. In the relentless pursuit of establishing their green credentials, companies are setting ambitious decarbonization commitments in the boardroom without the ability to translate these goals into action on the shop floor.

Currently, many companies are focusing on target setting with the challenge that these targets are often not supported by operational pathways to achieving these goals. The Energy and Climate Intelligence Unit (ECIU) reports that at least one-fifth of the world’s largest public companies have committed to meeting net-zero targets. In the last few years, thousands of Chief Sustainability Officers have been appointed to help drive these targets. These companies need to quickly understand the carbon levers in their operations because they have already made public commitments and need to avoid being perceived as “greenwashing.” Setting the next level of concrete metrics that can align and mobilize managers and frontline employees is essential to making measurable progress. Investors and customers of these companies will increasingly expect to see carbon performance communicated in terms of measurable improvements.

One such example is the BMW Group, which has committed to reducing carbon emissions from its productions by 80 percent from 2019 levels by 2030. To translate this target into executable plans, they must understand three key insights – carbon attribution, operational carbon levers, and carbon intensity variance.

Firstly, their carbon attribution, namely, which production activities drive the most carbon emissions? In attempting to quantify their carbon emissions and gauge how much they can reduce them, companies usually take an outside-in or top-down approach of allocation – distributing the total amount across different operational units. The problem with an allocation is it’s artificial and doesn’t connect well to actionable levers. In work we have been doing to help companies understand energy intensity and optimization, we take an inside-out approach because we link emissions to the actual machine and operational activities, which we call attribution.

With this information, you must decide which operational levers have the greatest impact on carbon intensity? For example, the painting process may be responsible for the most carbon emissions but are not a significant lever to change. Instead, the welding process may represent the most opportunity. It is important to understand what the levers are that will change, and such information is often not transparent at the boardroom level because top-down reporting is not connected with actual production activities.

Finally, does the carbon performance (carbon emission to produce each unit of product) vary from day to day, and why? Does it vary from product line to product line, and why? Does it vary from site to site, and why? Variances often point to concrete opportunities for improvement, and on the other hand, may lead to strategic considerations for adopting new product and operational strategies to reduce total emissions.

Most companies do not have enough data transparency about their productions to answer these questions. However, these are perquisites to their ability to meet the targets set. Currently, there is a great divide between the board room and the “control room.” Company boards focus on setting targets, while the factories carry our piecemeal initiatives to improve energy efficiency at incremental paces. It’s time to bridge the divide.

Looking for answers in machine data

Sustainable operations must be data-driven, and industrial AI-based predictive analytics can pave the way. At Atomiton we have successfully deployed our solution across several carbon-intensive sectors in the past few years. It delivers inline predictive analytics that syncs with a plant’s processes to solve these uncertainties. It can even indicate which operational conditions can be maintained without change and which specific levers can make the most impact.

The heart of the solution is the Atomiton Stack (A-Stack), an analytics engine that can sift through large amounts of production data and identify operational patterns. It allows users to easily connect and feed their machine data into analytics models designed to recognize energy consumption trends and discover sustainability levers. The stack comprises three layers: data contextualization, energy modeling, and sustainability analytics. The stack can ingest data from various sources with different protocols, including sensors, operational systems, historians, and enterprise applications such as ERP.

Today for each operational decision made, front line operators understand its implications on safety and quality, but they don’t understand its implications for emissions and sustainability.

A-Stack presents the actionable insights directly linked to companies’ sustainability targets. This helps the operators to see the board room net-zero targets translating into their day-to-day operational activities. For example, firing up a backup gas boiler or electric boiler may be equally good options for a particular upsurge of processing load, but the lower carbon emissions from the electrical boiler, especially if it is powered by “green electricity,” maybe a much more sustainable decision.

Categorizing carbon emissions

Greenhouse gas emissions are categorized into three groups or scopes by the most widely-used international carbon accounting tool, the Greenhouse Gas (GHG) Protocol. Scope 1 covers direct emissions from owned or controlled sources. Scope 2 covers indirect emissions from the generation of purchased electricity, steam, heating, and cooling consumed by the reporting company. Scope 3 includes all other indirect emissions in a company’s value chain.

Some companies focus on scope three, which is hard to control and track, as it requires an understanding of suppliers’ or customers’ carbon footprints. But what companies can control is their operations. For manufacturers, most of that comes from their production systems. For example, 90% of the BMW Group’s Scope 1 and 2 emissions come from its production facilities. This is an area of emission sources that we actively measure, visualize, and optimize with our software. We see that increasingly, as upstream companies start to demand transparency on Scope 3, their suppliers must equip themselves with the ability to measure and improve on their Scope 1 and 2 emissions because these translate into Scope 3 emissions for the upstream customers.

Using AI to understand the process and reduce carbon

Before Vopak started working with Atomiton, it knew its Savannah facility was one of its high energy consumption sites. Most of the energy used was purchased electricity, which generates Scope 2 emissions. Although there was a corporate mandate to reduce greenhouse gas emissions, there was no direct translation into how the terminal staff should operate the plant to achieve this ambition. Vopak provides a range of complex services to its customers with various equipment running at the terminal, conducting thousands of mixing, loading, and heating activities each week. Any concept of reducing carbon becomes abstract when the operators do not know the exact action levers they could use to achieve this.

The software uses AI models to perform electricity attribution and identify variances and operational levers through our work with them. It turned out that 80 percent of the controllable variance of electricity consumption came from one kind of heating activity. This immediately turned into an emissions lever – the hours of heating performed but not required.

However, this lever is not actionable until the analytics inform heating is required for which tank and which pipe segments, based on anticipated operational schedules, conditions, and expected lead time. With the detailed real-time insights from A-Stack, the team can effectively execute the change and significantly reduce Scope 2 emissions by 40 percent.

For boardroom goals to translate into improved carbon performance on the shopfloor, operators must have the tools to understand what does it mean for their day-to-day roles, do they need to change how they work, or do they need to do different initiatives? We call that ultra-transparency, and there is a lack of transparency in this respect. I would emphasize we cannot solve all the problems. Still, for manufacturers, we solve a major part of the dilemma by adding intelligence and visibility to their production and machine data to fully comprehend the carbon implications of their production decisions.

About the author:

Jane RenJane Ren is the CEO and founder of Atomiton, an enterprise IoT software company based in Silicon Valley. Named one of the top 20 Disruptive Influences in Tech by the CFO Magazine, Atomiton provides an innovative IoT software stack that runs next-generation intelligent applications for energy/oil and gas, smart cities, and other industrial sectors.