This paper describes a flexible, widely applicable sensor health monitoring system, as developed in partnership with Northrop Grumman Ship Systems (NGSS) under a small business innovative research contract for the US Navy. Traditional signal processing techniques were employed in conjunction with data driven models and fault identification and classification techniques to provide a robust analysis of sensor health. Key aspects of the system include: • Analysis of both high and low bandwidth data; • Modules that assess a sensor’s performance on an individual basis. These are designed to detect noise, incipient faults, spiking and signal dropout; • Modules that assess a sensor’s performance from an overall system perspective, enabling early identification of sensor drift and calibration issues; • Algorithms for high bandwidth signals designed to detect clipping, abnormal signal mean and range, and signal shape anomalies that enable identification of certain mechanical and electrical failures; • Mode detection algorithms that enable dynamic weighting of calculated health parameters in order to mitigate false alarms and missed detects; • Fusion algorithms that combine and interpret the output from the aforementioned modules, to provide estimates of overall sensor health and failure mode. The system’s capabilities were exercised on 1) laboratory datasets generated in-house with implanted faults, 2) data from tests conducted on the power distribution system driven by a Rolls-Royce MT30 gas turbine slated to power the Navy’s DDG1000 destroyer, and 3) from a low pressure air compressor (LPAC) found on legacy Navy weapons systems. The ability to detect and classify various electrical faults, issues related to calibration, and certain mechanical failures was validated. The system is suitable for offline mining of historical data, embedded on-line monitoring, and for application in distributed computing networks.

This content is only available via PDF.
You do not currently have access to this content.