Modern ship power systems aren’t just generators and breakers anymore—they’re networks of distributed controllers constantly exchanging data to keep voltage stable and share load. That connectivity enables smarter, more resilient control, but it also creates an uncomfortable truth: if an attacker tampers with the signals between controllers, they can push the system into unsafe behavior without ever touching a physical component.

A Winter 2025 NEJ technical paper proposes a cyber “tripwire” for distributed DC ship power systems that can detect—and even help pinpoint—attacks quickly without requiring every controller’s raw data to be centralized. The approach is easy to picture: each controller gets a “digital twin,” a learned model of what its signals should look like during normal operation. If the controller’s real measurements start deviating from what the twin predicts, that mismatch becomes a continuous red flag.

What makes the framework especially practical is how the digital twins are trained. Instead of pulling all operational data into a central server, the authors use federated learning: each controller trains locally on its own historical data and sends only model updates—not raw data—to a coordinating server. The server aggregates those updates into a shared model and distributes improvements back to the fleet of controllers, reducing bandwidth load and limiting exposure of sensitive system data.

The method is designed to handle two attack types that keep engineers up at night: false data injection (subtle corruption of measurements) and controller hijacking (replacing correct signals with malicious inputs). Critically, the goal is not just “something’s wrong,” but actionable localization—helping identify which controller is compromised so operators can respond faster.

The authors also address a real-world complication: controllers don’t all behave identically, and not every participant can be assumed trustworthy. Their aggregation approach uses an attention-based method that gives more weight to reliable updates and downplays outliers, improving robustness in messy, heterogeneous shipboard conditions.

Read the Winter 2025 NEJ issue / get access to the paper

ASNE members receive NEJ access as a core benefit. If you’re working ship defense, combat systems integration, autonomy, power/cooling, or next-gen survivability, this is the kind of analysis NEJ exists to publish.