Nonintrusive monitoring for shipboard fault detection

Abstract

This paper presents a case study applying nonintrusive load monitoring (NILM) for fault detection and isolation (FDI) of automated shipboard systems. A NILM system installed on an engine room subpanel of U.S. Coast Guard (USCG) Cutter SPENCER collected aggregated power consumption data for ten automated systems. A correlation-based transient identifier is used to disaggregate this data, identifying specific automated load events, including on/off events of the gray water disposal pump. A two-parameter model is calculated from these events and used for fault detection. Data collected during two operational periods of the SPENCER demonstrate the effectiveness of this model in identifying a pump sensor fault previously undetected by the crew. Early identification of such malfunctions prevents costly wear on the gray water disposal system pumps and avoids eventual catastrophic failure.