
Charles River Analytics is set to develop an alternative system to effectively detect submarines for the U.S. Navy, using their magnetic signals to detect stealthier and quieter vessels.
Under the new Small Business Innovation Research (SBIR) Phase II contract, valued at nearly $1 million, Charles River Analytics will improve upon current detection tools which primarily focus on acoustic measurements.
The Magnetometer-based ASW Guidance for Naval Enhancement of Tactical Operations (MAGNETO) tool uses a magnetic anomaly detection (MAD) sensor to detect and track distortions in the earth’s magnetic field caused by the ferromagnetic materials in submarines.
One challenge in this development is that magnetic signals decrease cubically with distance, making them difficult to detect. Interference from other objects also poses a problem, though magnetic signals do not depend on the medium they are traveling through. Artificial intelligence (AI) and machine learning (ML) technologies help solve these challenges by effectively isolating and extracting relevant signals from the surrounding noise.
In addition to detecting the presence of magnetic fields, anti-submarine warfare (ASW) techniques must also differentiate signals from submarines from those of large boats or even sunken shipping containers. To support this, AI can develop a distinct “signature” for submarine detection. Future iterations of MAGNETO will also be capable of analyzing wave noise to determine whether it indicates the presence of submarines.
MAGNETO tracks what is essentially a one-dimensional signal (a vessel’s “magnetic moment”) over time and compares it to previous detections. Charles River’s Vector Intelligence Build Environment (VIBE) workbench provides the 1D signal ML models that form the foundation for MAGNETO.
The tool follows a hierarchical approach, narrowing down the process of identification in successive stages. The first stage detects the presence of a submarine, the second stage determines the type of submarine—such as nuclear missile, diesel—continuing until a precise identification is made. This hierarchical system allows MAGNETO to rapidly process information in real time, ensuring only relevant data passes the initial screening and advances to subsequent stages.
Before reaching the classifier stage, MAGNETO first cleans the signals, removing any noise. In the second stage, a feature extractor selects only critical parts of the signal that may indicate activity. The classifier forms the final layer of analysis.
Dr. Todd Jennings, Research Scientist at Charles River Analytics and Principal Investigator of MAGNETO, commented, “The purpose of this project is to leverage advances in machine learning and signal processing to more reliably extract the signatures of these submarines and distinguish them from the magnetic signatures of other objects. The biggest advantage of MAGNETO is that it allows us to detect ultra-low-noise submarines reliably, quickly, and efficiently while minimizing risk to the warfighter.”
Other use cases for the tool reportedly include detecting airplane wreckage, underground power cables, and dropped shipping containers in the ocean. Lessons from MAGNETO and the VIBE workbench can also translate to other signals, including acoustic, LIDAR, and radio frequency.