Introduction

The oceans transport 90-95% of goods, supporting world trade and enabling efficient transport. We use our oceans to generate power, provide us with natural resources and for leisure. These vessels and structures will need to be designed, operated, and decommissioned safely and efficiently. Data science and artificial intelligence will play a large role in this future. 

Project aims

We will develop new tools and techniques for marine and maritime, help companies deploy effective data-centric systems and help educate those in the industry on how to make the most of these approaches. The result will be a wider uptake of data science, machine learning and artificial intelligence leading to safer ships and reduced emissions from operations.

Applications

The research will be used across the marine and maritime industries, including shipping, offshore platforms, renewable energies, ports, autonomous systems and leisure industries. The group has been involved in a number of advances in the development and uptake of machine learning and artificial intelligence in the maritime field to date. Three practical examples are:

Voyage Optimisation Software to reduce emissions

Compared with existing optimisation solutions on the market, T-VOS has been proven to increase fuel savings and reduce emissions by 5%, improve time savings by 7%, and deliver Time Charter Equivalent improvements by 8%. The savings have been validated directly by vessel owners and represent average savings over existing solutions.

 

Layout optimisation for leisure yachts

The floor plan for a new yacht designed using a genetic algorithm, developed by the group lead. The design is compared to one generated by a team of designers from Olesinski, on the left. The software reduces the layout design time from two weeks to two days, allowing a focus on the detailed design and reducing the time spent on more repetitive tasks.

olesinksi

 

Machine learning augmented computational fluid dynamics

Every year, the marine industry relies more heavily on computational fluid dynamics (CFD) simulations, but these simulations are still far too slow for real-time operations use, or even open-ended design. By combining a simplified two-dimensional flow model with a deep convolutional neural network (CNN), our group has developed a three-dimensional flow prediction tool with 95-98% generalization accuracy which is 2-3 orders of magnitude faster than standard CFD.

real cnn
 

Organisers

Contact info

Adam Sobey, [email protected]