NSWCPD tests AI/ML model to predict compressor failures

Machine Learning


The Naval Surface Warfare Center, District of Philadelphia, has begun testing artificial intelligence and machine learning models to determine whether signs of deterioration in submarine air compressors can be identified early, before failure occurs.

NSWCPD tests AI/ML model to predict subsea compressor failures

Don’t miss the Potomac Officers Club 2026 Digital Transformation Summit Tomorrow, April 22nd. Register now Hear high-level discussions about safe AI deployment.

NSWCPD engineers are evaluating an experimental model that analyzes vibration data from high-pressure air compressors that support critical functions on submarines, Naval Sea Systems Command announced Friday, noting that the goal is not only to detect failures, but to assess their progression and estimate the time to failure. However, calculating the remaining useful life of a component requires large datasets and further model development.

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

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

How does NSWCPD train AI models to detect equipment failures?

To generate usable data, NSWCPD engineers built a controlled test environment and introduced obstacles such as air leaks, intake restrictions, and cooling issues. The team used a series of accelerometers to collect vibration signals and trained a model to distinguish between healthy and faulty conditions.

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

“Lab tests so far have shown real promise,” Dingley said, adding that next steps will focus on extending the model with more diverse datasets and testing its performance in operational conditions.

What other AI initiatives is NSWCPD pursuing?

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



Source link