Virtual fuel sensors can save environment and money

No new measurement sensors need to be installed. By collecting data from the ship's energy system, the vessels fuel consumption can be predicted using machine learning.
It is one of the conclusions of a thesis presented at Linnaeus University on December 13th.


When Fredrik Ahlgren describes his dissertation, he first speaks of losses, of energy that is of no use. He mentions the heat contained in the exhaust gases is wasted through the ship's chimney. Heat that could be converted into electricity. This will reduce emissions by up to five percent.

In his dissertation, he has conducted a comprehensive analysis of the entire energy system on board the Birka cruise ship, which runs between Stockholm and Åland.

“We collected engine data: temperatures, pressures and exhaust gases and put it into a theoretical model of an ORC (Organic Rankine cycle) system, that can produce electricity from waste heat. This way, we learned how to design such a plant in real operating conditions. Because she was a cruise ship, she often drove slow and far from the design point. Now we were able to optimise the facility after the operating scenario.”

In today's ships, even in the older ones, there is a lot of data from the machinery system stored in local databases.

"The data is used to make weekly reports, watch trends, see if service needs to be done on the engine, but it could be used for so much more. It is important to learn to take care of it, " Fredrik Ahlgren says.

One and a half years ago, when he was quite far into hos work with the thesis, he got an idea of how to use machine learning to predict the ship fuel consumption.

"I took noon-reports for as much as up to four days and paired it with other high-resolution data such as temperatures and engine speeds. In this way you can create virtual fuel sensors. In this way you can show the real-time fuel consumption on the bridge without having a fuel sensor installed, alternatively view each individual motors’ consumption only from a common mass flow meter! I compared with a mass flow meter and managed to show that it works."

Fredrik Ahlgren's dissertation shows that machine learning is not only an effective tool for accurately measuring fuel consumption, it is also likely less expensive than other methods, such as installing additional mass flow meters. Knowledge of how the energy system works is of course of incredibly valuable before major investments like ORC systems are installed. This is especially true for ships that have dynamic operating scenarios with a lot of starts and stops.

"Machine learning is a hot topic right now and it is also the part of my work that has gained most of the interest. The dissertation is already sent to a hundred interested in the industry. Right now I'm discussing with shipowners about continuing the project in some form. So there is definitely a lot of interest."