Saving Grace

As costs rise, the market tightens and legislation becomes more stringent, fuel efficiency has never been more important. We spoke to GreenSteam Co-founder and CEO/CTO Daniel Jacobsen about how his organisation is using AI to help the industry meet those demands

These are challenging times for ship owners and managers. Environmental legislation, coupled with energy efficiency directives and a difficult global market, mean that being compliant and competitive matters like never before.

At the heart of competition and compliance are operational efficiency and effectiveness, two factors united by a single commodity: fuel. Using fuel optimally reduces costs, decreases emissions and limits a vessel’s environmental impact. Does that come at the cost of operational efficiency? Not anymore.

Knowing and understanding the factors that contribute to a vessel’s performance over time, identifying where each has lost fuel or used it sub-optimally is key to building a road map to optimisation.

The task has become more pressing in light of IMO2020, which comes into force on January 1 2020 and limits sulphur emission from marine fuels to 0.50%. This will cause fuel costs to increase by at least 50%.

To that end, knowing your vessels and how they work is critical to optimising performance and fuel efficiency. The key to unlocking that value is AI and its ability to analyse data and learn from it. At the forefront of that in the marine market is an organisation called GreenSteam, whose AI solutions are designed to analyse vessel performance and deliver insights that result in actionable intelligence and measurable financial gains.

Daniel Jacobsen is CEO/CTO and co-founded GreenSteam in 2007 after completing a PhD in applied mathematics from the Technical University of Denmark (DTU). Along with his fellow co-founders, he saw tremendous potential in applying AI and machine learning to marine fuel efficiency challenges.

After several years of intense testing with innovators and early adopters, they launched their first product – GreenSteam Optimiser – in 2010 and immediately signed their first customer. Since then, GreenSteam has seen dramatic growth, opening offices in Poland and the UK, while achieving commercial success across all vessel types and classes. We caught up with Daniel to see how GreenSteam is changing the industry one vessel at a time with some of the industry’s most advanced tech.

Daniel, you are applying advanced tech to solve one of the industry’s most complex problems. In a nutshell, how does it work?

To solve problems that are so complex that only machine learning can provide the solution, you need three things: domain expertise, machine learning or AI expertise, and lots of data. In combination with our growing customer base, we now have plenty of all three: a team of naval architects and navigators, a team of machine learning/AI experts, and large datasets – ever increasing as the marine sector is discovering the value of digitalisation and data.

AI models are somewhat like humans in that they can learn from experience, which is what data really is. The difference is that they can learn faster and are able to capture relationships and discover effects that humans never could, because of the complexity and sheer number of dimensions. 

Using these models – that keep learning 24/7 by the way – our solution teams work closely with our customers to build tools and solutions that enable serious fuel and operational savings to be implemented. 

“You need three things: domain expertise, machine learning or AI expertise, and lots of data”

How does that work in practice for shipping? 

There are so many factors that impact a vessel’s fuel and operational efficiency – for example wind, waves, fouling, load, vessel type, and sea temperature. Each vessel is actually unique and our AI learns to understand each individual vessel, as well as relating performance across many vessels. 

These factors can be grouped into what you could call conditions, controls, and costs. Conditions include wind, waves, and currents. Controls include anything you can change, such as trim and speed. Costs can be fuel or power but can also include commercial costs, such as charter rates that you want to take into account for, say, speed or route optimisation.

We work with any data and formats that our customer has, and we also offer our own autologging installations. The first step is to clean the data, for example identifying and removing implausible noon reported fuel consumptions. The AI models then ingest the data and finally, our tools and solutions use these models to guide and advice the user as to what could be done to optimise effiency.

As an example, consider fouling. The user can see how much resistance and cost is added by fouling and how this is developing over time, to assist with decisions on when and how to clean the hull. The tools also give the user early warning if a coating is starting to break down. 

Does this improve with time as well?

Yes, it does. The more data you add, the more the AI learns and the better and more accurate the solutions are. The state-of-the-art AI that we apply and constantly improve is even able to transfer knowledge from one vessel to another, while also continuing to learn per vessel. You might think of it like a self-driving car that drives down one street, shares that street with the rest of the fleet and, in return, gets a map of the city. The potential for savings, operational efficiency and environmental improvements with this type of ‘collective AI’ is tremendous.

How do you measure performance improvements?

Our users get baselines that are more powerful than what traditional approaches have offered. In those approaches, one typically isolates small amounts of data where everything is ‘nice’ – still water, no wind, and so on, to obtain a ‘baseline’ – typically measured during sea trial. To measure ‘performance’, you again need to isolate these small amounts of ‘nice’ data so that you can compare with the original ‘nice’ state. The problem is that you end up throwing out most of your data, sometimes more than 95% of it, and you lose all the information that is contained in that data – simply because it is very rare in shipping that everything is ‘nice’.

Instead, our users see baseline performance based on all the data. This is possible because our models learn the cause and effect between any factor and condition, and the costs (e.g. fuel). When our users see performance, this is again using all the available data and comparing with a previous state, also using all of the data. Thus, it is much more reliable and informative and thus useful.

As an example, consider trim – static or dynamic. To calculate baselines and performance we ‘ask’ our models: ‘In this period of time, what would the cost have been if the vessel had been operated at the best trim?’ This yields a simple percentage index, in this case called ‘trim performance’, that the user can then track. If, after some crew training, this number edges closer to 100%, the user knows that the training was effective. 

“we will see AI transforming shipping deeply, even changing core business models”

So in order to baseline, you need a reasonable amount of historic data?

Yes, typically a couple of years’ worth, ideally more in case of fouling so that the data includes one or more dry dockings or hull cleanings. 

One of the most valuable aspects of machine learning is that it will allow you to manage uncertainty and risk. Our solutions will give advice, but if a model finds itself in a new situation and is uncertain how to act or proceed due to a lack of relevant data, it will tell the user that and show exactly where the uncertainty lies. This can then even be used as an intelligent exploration tool, guiding the user towards collecting the most relevant data in the future with the least effort. 

How to you think AI will change shipping long term?

As we have seen in other industries, I think we will see AI transforming shipping deeply, even changing core business models. And just as in other industries, the early adapters and innovators will gain decisive competitive advantages from taking advantage of the powers of AI before the rest of the industry. Take insurance as an example. Here, early AI adopters are gaining market shares by customizing risk scores based on data from cars, fitbits and so on. Or manufacturing, where companies ahead on the ‘industry 4.0’ track are winning. In shipping, the companies that are the first to take advantage of AI will similarly gain market shares over their competitors. These companies are also the ones that see that sharing of data with ‘AI enablers’ like GreenSteam, while it can feel uncomfortable at first, brings a return of value that is simply game changing. I am excited about being part of this transformation especially considering that – despite the room for improvement – shipping is the most environmentally friendly means of transporting goods and owns much credit in the reduction of global poverty that we have witnessed over the last decades. If you want AI to change the world for the better, shipping is a great place to start.