According to Kziazik, the Navy and DIU launched the search two years ago. The University of Maryland reportedly uses AI tool to help sailors spot deadly waves in advance, which still threaten the lives of people and ships.
In their new study, Balakumar Balachandran and Thomas Breunung used neural networks, an AI tool technique. The researchers trained their neural network using a large amount of data collected from floating sensors (buoys) near the coast.
The collection included 14 million measurements, each lasting 30 minutes, recording wave heights from 172 buoys scattered along the coast of the United States and Pacific islands.
The network was carefully trained to distinguish between normal waves and those preceding an abnormal wave. Also, learned to recognize patterns that foreshadow an upcoming rogue wave.
The researchers tested the tool’s capabilities using a second set of 40,000 sea surface height observations from the same buoy.
Filipino fishermen prepare to sail their boat into the West Philippine Sea in Mariveles, Bataan, West Luzon, May 18, 2021, despite harassment by Chinese militia and coast guard in disputed waters.
The test results were impressive. The method accurately predicted the development of 75% of abnormal waves one minute and 73% of irregular waves five minutes in advance, proving the tool’s effectiveness.
In addition, the researchers used the technology on two new buoys that were not part of the training set. It still predicted abnormal waves with great accuracy. This means that the program can even predict them in new ocean areas.
For decades, sailors have worried about rough seas, usually powerful storm waves. The waves are strong enough to collapse or destroy ships.
NOAA defines abnormal waves as more than twice the size of the waves around them. These waves usually appear in the high seas and oceans.
AI Tool sails against the sea.
The study is the latest example of how artificial intelligence is continuously applied in the vast ocean. Recently, the U.S. Navy has been studying AI-driven drones to detect hostile threats automatically, and a series of successful tests have proven that they can detect mines twice as fast.
Due to their stealthy nature, these mines are often difficult to detect, posing a significant threat to naval operations. Using AI in this context has the potential to revolutionize mine detection and significantly improve maritime security.
Tools to predict malicious waves five minutes in advance
Alex Campbell, head of the Pentagon’s Defense Innovation Unit’s maritime service, a California-based unit that aims to integrate commercial technology into the armed forces, said the effort has cut the time required to search the seabed for underwater bombs in half.
Additionally, Campbell also reported that the Navy is announcing other manufacturing contracts to expand the use of the technology in underwater drones and study its potential to detect enemy ships, aircraft, and other threats.
U.S. Marine Corps Major Nick Ksiazek, who manages the project and works in the DIU’s artificial intelligence unit, said the Navy has begun testing machine learning algorithms that use sonar sensors to detect underwater features and navigate on the seafloor.
According to Kziazik, the Navy and DIU launched the search two years ago.
The crew acquires imagery to detect hostile waterways or commercial waters. Campbell said the Navy reduced the number of crew members in the operation by ten, shortening the operation to two days, thanks to the application of artificial intelligence technology.
Rapid training at sea using artificial intelligence
Successful artificial intelligence models often require rapid retraining to adapt to new environments. Ksiacek believes that identifying items on sandy, rocky, and trash-covered seabeds requires separate identification methods. He said underwater drones have been deployed in the Indo-Pacific region and participated in exercises.
In addition, the Navy is upgrading its artificial intelligence models faster to remotely issue commands to drones when the drone deck is on the water without having to move them completely out of the water. Previously, it took six months to implement such models.
What is the progress on the scientific reports?
A study published in Scientific Reports journal offers a new tool for predicting the appearance of unusually large and unexpected waves at sea, known as swells, up to five minutes in advance. The authors suggest that the tool could warn ships and offshore platforms so that people working on them can seek shelter, make emergency stops, or take action to reduce the impact of approaching swells.
The tool Thomas Brunung and Balakumar Balachandran developed consists of a neural network trained to distinguish between waves that will follow a swell and those that will not.
Furthermore, the researchers trained the neural network using a dataset consisting of 14 million 30-minute samples of sea surface height measurements from 172 buoys off the coast of the continental United States and Pacific islands. They used the tool to predict the occurrence of swells with a dataset consisting of 40,000 sea surface height measurements from the same buoys.
The researchers found that their tool correctly predicted 75% of swells one minute in advance and 73% five minutes in advance. The tool could also predict unusual waves near two buoys not included in the training data set one minute in advance with 75% accuracy. It highlights that the tool may be able to predict large waves in new locations.
The authors suggest that instrument forecasts’ accuracy and warning time could be further improved by combining water depth, wind speed, and wave position data. However, future research could also predict the height of upcoming large waves or when they might appear.
