US Navy logo. NSWCPD has begun testing AI models to detect early signs of submarine compressor failures.
NSWCPD has begun testing AI models to detect early signs of submarine compressor failures.
///

NSWCPD Tests AI/ML Models to Predict Submarine Compressor Failures

2 mins read

Naval Surface Warfare Center, Philadelphia Division, has begun testing artificial intelligence and machine learning models to determine whether they can identify early indicators of degradation in submarine air compressors before failure occurs. 

NSWCPD Tests AI/ML Models to Predict Submarine Compressor Failures

Don’t miss the Potomac Officers Club’s 2026 Digital Transformation Summit tomorrow, April 22. Register now to hear high-level discussions on secure AI adoption.

Engineers at NSWCPD are evaluating an experimental model that analyzes vibration data from high-pressure air compressors, which support critical submarine functions, Naval Sea Systems Command said Friday, noting that the goal is not only to detect faults but also to assess their progression and estimate time to failure. Calculating remaining useful life for components, however, requires large datasets and further model development.

The project aligns with the U.S. Navy’s broader push to advance predictive maintenance through its Condition-Based Maintenance Plus initiative, which integrates traditional maintenance practices with AI-driven diagnostics and prognostics.

“Projects like this help us understand where AI adds value, where it still falls short, and how we can align digital innovation with our core mission of delivering warfighting capability in both acquisition and sustainment to the fleet,” said NSWCPD Technical Director Nigel Thijs.

How Is NSWCPD Training AI Models to Detect Equipment Faults?

To generate usable data, NSWCPD engineers built a controlled test environment and introduced faults such as air leaks, inlet restrictions and cooling issues. Using arrays of accelerometers, the team collected vibration signals to train models to distinguish between normal and faulty conditions.

According to Colin Dingley, machine learning engineer at NSWCPD, early tests show the models can reduce thousands of vibration inputs into a small set of indicators that reliably identify common faults.

“Our lab tests to date show real promise,” Dingley said, noting that the next phase will focus on scaling the models with more diverse datasets and testing performance in operational conditions.

What Other AI Initiatives Is NSWCPD Advancing?

The compressor project is part of a broader portfolio of AI and data-driven efforts at NSWCPD. These include developing machine learning algorithms for power-system health, digital twin models for shipboard systems and enterprise remote monitoring capabilities.