Tracking enemy submarines is critical function for any navy. Since their advent, submarines have been a critical weapon. They are stealthy, underwater hunters and the most advanced types use nuclear reactors, so they rarely have to surface for air. They can travel at up to a few hundred meters below the water surface, and can use underwater obstacles like mountains or infrastructure to effectively mask their presence from other detection technologies, like sonar. And they are a morale killer as much as a legitimate threat.
It requires a significant amount of technological prowess to find andtrack these underwater behemoths. However, a recent paper from researchers at the Air Force Engineering University in Xi'an, China, describes a drone swarming technique that could prove exceptionally good at tracking submarines.
Drones take flight and set sail to discover submarines
Detecting these stealthy giants requires different types of sensors. The most effective arrays combines three: visible light sensors to detect submarines on the surface, and sonar and magnetic sensors to detect them underwater. Visible light and sonar sensors are largely self-explanatory and have been used for decades. Conversely, magnetometers might seem counterintuitive when searching for submarines, however, they are giant steel tubes creating friction in a conductive medium. They act like a Faraday cage and create slight variations in magnetic fields which can be separated from background noise.
The platforms on which each type of sensor is mounted matter as well. Two types of uncrewed vehicles can be used in the ocean: Uncrewed Aerial Vehicles (UAVs) and Uncrewed Surface Vehicles (USVs). UAVs have entered the collective cultural consciousness over the last decade, but USVs are still relatively novel, despite their successful recent use in several operations by the Ukrainian Navy. USVs can be thought of as simple autonomous boats that travel on the ocean's surface. In the overall scheme of submarine tracking, UAVs have the advantage of covering more area than USVs. However, USVs are capable of utilizing underwater sensors that UAVs are not, such as sonar.
According to the paper, the most effective use of the three sensor modalities is to mount visible light sensors on both the UAVs and USVs. At the same time, each UAV would carry a magnetometer, and each USV would carry a sonar. This configuration takes advantage of each platform's unique strengths while ensuring the best coverage of the largest area to try to track their elusive prey.
However, it wouldn't matter what platform each sensor is connected to if they couldn't coordinate their search pattern. That is why a pathfinding algorithm, especially one that coordinates between the two tracking platforms, is so important. Any such pathfinding algorithm should have two goals: complete coverage of a given area and minimizing the travel time to ensure a fast, efficient search.
According to the paper by Dr. Ning Wang and their co-authors, those two goals are best met by utilizing a model call the platform-sensor-environment mapping model based on deconvolution, or PSE-D, model. The PSE-D model uses deconvolution, which is a type of deep-learning algorithm, to map specific waypoints throughout an area for a platform, whether a UAV or USV, to visit. The waypoints are calculated based on the expected range of each sensor modality. The logic then introduces an optimization problem for finding the minimum number of waypoints necessary for full coverage of an area. It sets about solving that optimization problem using a genetic algorithm, which is also commonly used in deep learning scenarios.
Waypoints are only part of the equation, though - an actual UAV or USV must physically go there to cover that area. A second algorithm determines what platform to send to a given location. As input, it takes the number and type of platforms available, their initial positions and the overall area to cover. As an output, it determines the pathing each individual platform will take, with the goal of minimizing the overall travel distance and the search operation's time and energy usage.
To prove their system worked, the authors simulated their algorithms with a variety of real-world situations. They measured success by ensuring that no region was missed and looking at the overall distance traveled to cover the area.
There were some confounding factors, though. In some simulations, a simulated area of debris was introduced, making it impossible for a magnetometer attached to a UAV to detect a submarine due to interference from other nearby metal objects. These areas instead had to be searched by a USV with sonar that could disregard the confounding objects. Additional practical difficulties included the altitude of the UAVs, as they had to fly relatively close to the water in order for their sensors to work accurately and variability in wave height would frustrate that effort.
Actual submarines would add another layer of difficulty because they move. The simulation didn't include accounting for a submarine actively moving through the zone, let alone if it was taking evasive maneuvers, and the likelihood that the search pattern would be able to find it.
High seas, high stakes
In other words, tracking submarines using autonomous drone swarms is still a long way off from practical reality. However, sensors and drone platforms have much higher innovation rates than multi-billion-dollar submarines, which can take decades to design and construct. And they are much cheaper, which means they scale quite nicely.
As we've seen in Ukraine, drones have changed ground reconnaissance and warfare. It might only be a matter of time before navies can successfully locate and track hostile underwater threats with drones. The next round of submarine innovation will consider all of these types threats.
It's an arms race between the hiders and the seekers.
