There are few environments more unforgiving than the ocean. Because of its unpredictable weather patterns, and the limitations of communication, large swathes have remained unexplored and cloaked in mystery.
The ocean is a fascinating environment, with many current challenges such as microplastics and algae blooms, coral bleaching and rising temperatures. Wim van Rees at MIT, the ABS Career Development Professor, states that the ocean is fascinating. The ocean also offers many opportunities, including aquaculture, energy harvesting, and exploration of the many sea creatures we have yet to discover.
Engineers working in the oceans and mechanical engineers like van Rees are using scientific computing advances to address their many challenges and capitalize on the opportunities. These scientists are developing technologies to better understand the oceans and how organisms and human-made cars can move within them.
Bio-inspired underwater devices
As fish swim through the water, they perform an intricate dance. Flexible fins move in currents of water and leave a trail of eddies behind.
Fish have intricate internal muscles that allow them to adapt to the shape of their bodies and fins. Van Rees explains that this allows them to propel themselves in many other ways, far beyond what any man-made vehicle is capable of.
Van Rees says that we are now closer than ever to creating flexible and morphing fins for underwater robotics thanks to advances in additive and optimization manufacturing techniques and machine learning. Understanding the impact of soft fins on propulsion is therefore more important.
Van Rees’s team is using numerical simulation to explore the design space of underwater devices that have a greater degree of freedom, such as fish-like deformable fins.
These simulations allow the team to better understand the fluid and structural mechanics involved in fish’s flexible, soft fins moving through fluid flows. They are now able to understand how fin shape changes can affect swimming performance and how they can be improved. Van Rees adds that supercomputers can be used to solve the problem at the interface between flow and structure by using precise numerical techniques and parallel implementations.
Van Rees hopes to create an automated design tool for autonomous underwater devices by combining his simulation algorithms for flexible structures with optimization and machine-learning techniques. This tool could be used by engineers and designers to create underwater vehicles and robotic fins that can intelligently adapt their shapes to meet their immediate operational goals, whether they are swimming faster or more efficiently, or performing maneuvering operations.
Van Rees explains that we can use optimization and AI to do an inverse design within the entire parameter space to create smart, adaptable devices. Or, we can use individual simulations to identify physical principles that determine why a particular shape performs well.
Robotic vehicles can use a variety of algorithms to synchronize their movements
Michael Benjamin, Principal Research scientist, is interested in improving the way vehicles move through water. Benjamin, then a postdoctoral researcher at MIT, launched an open-source project to develop an autonomous helm technology. Multi-objective optimization is a new method that has been used to create the software. This software has been used in a variety of industries, including Sea Machines and Thales UK. Benjamin’s PhD research led to the development of this optimization method. It allows a vehicle to choose its heading, speed, depth, direction, and speed to achieve multiple simultaneous goals.
Benjamin is now developing obstacle-avoidance and swarming algorithms. These algorithms would allow uncrewed vehicles to communicate and explore a specific area of the ocean.
Benjamin is currently focusing on how to disperse autonomous vehicles in oceans.
Let’s say that you want 50 vehicles to be launched in a particular section of the Sea of Japan. We want to find out if dropping all 50 vehicles at the same spot is a good idea or if it would be more efficient to have them dropped off at different points in the area. explains Benjamin.
His team has developed algorithms to answer this question. Swarming technology allows vehicles to periodically communicate their location with other vehicles. Benjamins software allows these vehicles to disperse in the best distribution for the area they are operating in.
The ability to avoid collisions is central to the success and safety of the swarming cars. International maritime rules, known as COLREGS or Collision Regulations, make it difficult to avoid collisions. These rules determine which vehicles are allowed to cross paths. This presents a unique challenge to Benjamins swarming algorithms.
The COLREGS were written with the goal of avoiding another contact. However, Benjamin’s swarming algorithm had been developed to account for multiple unpiloted vehicles trying not to collide with each other.
Benjamin and his team developed a multi-object optimization algorithm, which ranked specific maneuvers on an index ranging from zero to 100. A zero would indicate a collision, while 100 would indicate that the vehicles have avoided collision completely.
Benjamin says that our software is the only marine program where multi-objective optimization provides the core mathematical basis for decision making.
Researchers Benjamin and van Rees employ machine learning and multi-objective optimizing to address the complex nature of vehicles moving through ocean environments. Pierre Lermusiaux (the Nam Pyo Suh Professor) at MIT uses machine learning to better understand ocean environments.
Improved ocean modeling and predictions
Perhaps the most well-known example of a complex dynamical system is the oceans. Fluid dynamics, changing tides, weather patterns and climate change make oceans an unpredictable environment that can change from one moment to another. Forecasting is difficult because of the constant changing nature of the ocean environment.
Researchers have used dynamical systems models to predict ocean environments. However, Lermusiaux explains that these models have their limitations.
Models cannot account for every molecule water in the ocean. The ocean measurements and models’ resolution are limited. If you are looking at climate models for global oceans, you might have a data point about every 10 kilometers. Lermusiaux explains that this can have a significant impact on the accuracy and precision of your predictions.
Abhinav Gupta, a graduate student, and Lermusiaux, a machine-learning framework have been developed to compensate for the low resolution or accuracy of these models. Their algorithm can take a simple model of low resolution and fill in the gaps to create a complex model with high resolution.
Gupta & Lermusiauxs framework learns for the first time and introduces time delays to existing approximate models in order to improve their predictive abilities.
Gupta says that things in nature don’t happen instantly. However, most of the popular models assume that they are happening in real-time. To make an approximate model more precise, the machine learning and data used to create it need to reflect the effects of past states on the future prediction.
The neural closure model of the teams could allow for improved predictions such as a Loop Current Eddy hitting an oil-rig in the Gulf of Mexico or the amount of phytoplankton found in a given area of the ocean.
Researchers can unlock more of the ocean’s mysteries and find solutions to the many problems facing the oceans as computing technologies like Lermusiaux’ neural closure model and Gupta continue to improve.