How machine learning can understand the growth of weeds on ships

Using machine learning to improve the performance of ships affected by fouling – the growth of weeds on the hull.

All boats and ships grow weeds on the bottom (fouling) while they are in the water. How quickly these weeds grow depends on a range of factors, including the quality of the paint on the boat, the type of antifouling paint used, the speed a ship travels, the environmental conditions the ship faces – and more. Yet as these weeds grow, the ship’s performance is affected, with more power and fuel needed to travel at the same speed.

For environmental, financial and performance reasons, gaining a better understanding of fouling is necessary. As a result, Professor Giles Thomas from UCL Mechanical Engineering led a research project to use machine learning to better understand the fouling process.

How to pinpoint the impact of fouling on ships

“The idea of using machine learning for this is to try to isolate out all the variables impacting a ship’s performance,” explains Professor Thomas. “A ship might use more fuel one day because it goes into a big current or a headwind. Machine learning is the best way to do it because you've got so many variables that you don't understand the relationships between them.”

The team explored a range of machine learning techniques to see which one could give the most accuracy for this project. As many ships collect data through high-frequency Continuous Monitoring (CM) systems, there were extensive datasets readily available to train the machines to interpret and learn from the data. Professor Thomas partnered with Maersk to access the data from five container ships operating for 12 months between Europe and South America during 2018. Using data from the route in these warmer waters meant higher marine growth was likely, which would in turn create a noticeable fouling effect. The team also fed additional information into the machine learning process, such as Copernicus weather data, short term weather forecasts, wind speed, true wind direction, and days since the last hull clean.

After applying the data to different machine learning models, the team found the best way to hone in on a way to predict power usage. In fact all models gave a consistently low levels of errors, demonstrating how machine learning in general has great potential in this area. The best performing model was the Random Forest model, where multiple decision trees are trained in parallel, with the predictions from all trees then pooled to create an average value.

Accurate predictions that will help the environment

“With machine learning, you use a certain amount of data to train your system,” explains Professor Thomas. “Then you test it against data that you haven't used in that training to see how accurately you can predict the power or fuel consumption. The key thing in this project was how accurately we were able to predict it. There was a much higher accuracy than what has been done before.”

The research found an average increase in power requirement of 5.2% for a fouled ship compared to a clean ship. This was the case across all speeds and in all given conditions. For one of the ships used in the research, the team calculated that the additional fuel used per day due to fouling equated to almost 6,000kg, at a cost of £2,500.

The model will enable shipping companies to predict each ship’s deterioration in performance due to fouling. Since the cost of halting shipping operations in order to clean the hull – and the cost of the cleaning itself – is expensive, shipping companies can use this data to find the best time to undertake cleaning.

This research comes at a crucial time for both the shipping industry and in relation to global environmental agendas. Carbon dioxide emissions from shipping have increased by 70% since 1990, and the International Maritime Organisation (IMO) has a target to lower the 2008 greenhouse gas emission level from shipping by 50% by 2050. Research that enables smarter monitoring processes like this will make a tangible difference. “The good thing is that it’s helping the operators save fuel costs,” says Professor Thomas. “And that has a knock-on environmental benefit, because it means that they're burning less fuel and they're reducing emissions.”

---