With a large maritime footprint and a stake in anything up to 2,000 different vessels on the water on any given day, the incentive for a company such as Shell to be at the forefront of efforts towards safer and more efficient shipping are obvious.
“Some of the more complex data we receive from our assets we wouldn’t have been able to understand or make any insight into without the help of Machine Learning or Artificial Intelligence,” explains James Helliwell, fleet technical excellence analyst for Shell Shipping and Maritime tells The Naval Architect.
In 2016, Shell started upon a strategy of digitalisation. On a wider level this strategy includes the development of technologies such as 3D printing and VR, and like many companies Shell is exploring the possibilities for remote surveying.
But it’s with the capability to start processing the data from onboard sensors that AI is proving particularly advantageous. Until very recently, the reporting of onboard data had scarcely changed in more than a century; it was principally the noon report.
Onshore analysis of vessel performance was reliant on crew going around the engine room and physically reading the gauges and sensors, writing the data down on a clipboard and then radioing or phoning it back. Only in the last five years, with the advent of the Internet of Things and the constant streaming of data from ship to shore has such a tool been possible.
Shell Shipping and Maritime has developed a draft trim and optimisation program that is currently being deployed across 50 oil product carriers and 12 LNG carriers in its fleet. The program uses a software tool known as JAWS (Just Add Water System) which initially worked off a statistical model using noon report data, before high frequency data sensor data became available.
“The thing that surprised us is that a data driven approach hadn’t been developed before,” says Helliwell. “Historically, CFD hull models were used to try and determine the optimum hydrodynamic condition. By using real data from the vessel in service, we can also capture the changes in vessel performance over time, such as the changes in hull fouling, to get more accurate results.”
Over the past year or so, as part of ongoing collaboration with the University of Southampton, Shell has been moving beyond the statistical model towards an AI-based tool which takes some of this data and feeds it through a neural network to make predictions. In other words, using ML to model the powering of a vessel through the interaction of the different variables.
With the aid of specialist human analysis of the raw data to help ‘train’ the neural network, it has been able to estimate within 2% error of the actual measured shaft power. Inevitably, there are certain problems that can arise with the reliability and accuracy of some sensors, particularly on older vessels, and these still require intervention until the neural network has a better understanding of such anomalies.
Nonetheless the model is currently undergoing final validation with the aim of deploying the AI-enhanced JAWS across the Shell fleet later this year. For the present JAWS is intended as a decision support tool for crew, but unsurprisingly the eventual aim is to remove the human element altogether and fully automate draft and trim using the neural network.
Helliwell admits that an optimised ship may count for little, or even be suboptimal, if similar initiatives aren’t implemented across the wider supply chain, something that will require increased standardisation and sharing across data platforms.
It’s something Shell is keen to promote as part of its wider commitment to decarbonisation, including the recent publication of its ‘Decarbonising Shipping: All Hands on Deck’ report (jointly published with Deloitte and available at: www.shell.com/DecarbonisingShipping), which sets out a path for emissions reduction.
“One of the key solutions identified by the industry in the report was improvements in operational efficiency. Our work on developing JAWS is an example of the work that Shell are doing on this,” he concludes.
