Operational experience with intelligent software agents for shipboard diesel and gas turbine engine health monitoring

Abstract

The power distribution network aboard future navy warships are vital to reliable operations and survivability. Power distribution involves delivering electric power from multiple generation sources to a dynamic set of load devices. Advanced power electronics, intelligent controllers, and a communications infrastructure form a shipboard power distribution network, much like the domestic electric utility power grid. Multiple electric generation and storage devices distributed throughout the ship will eliminate dependence on any single power source through dynamic load management and power grid connectivity. Although new technologies are under development, gas turbine and diesel generators remain as the likely near-term power sources for the future all-electric ship integrated power system (IPS). Health monitoring of these critical IPS power sources are essential to achieving reliability and survivability goals. System complexity, timing constraints, and manning constraints will shift both control and equipment health monitoring functions from human operators to intelligent machines. Drastic manning reductions coupled with a large increase in the number of sensor monitoring points makes automated condition-based maintenance (CBM) a stringent requirement. CBM has traditionally been labor intensive and expensive to implement, relying on human experiential knowledge, interactive data processing, information management, and cognitive processing. The diagnostic robustness and accuracy of these embedded software agents are essential, as false or missed diagnostic calls have severe ramifications within the intelligent, automated control environment. This paper presents some of the first reported results of intelligent diagnostic software agents operating in real-time onboard naval ships with gas turbine and diesel machinery plants. The agents are shown to perform a substantial amount of CBM-related data processing and analysis that would not otherwise be performed by the crew, including real-time, neural network diagnostic inferencing. The agents are designed to diagnose existing system faults and to predict machinery problems at their earliest stage of development. The results reported herein should be of particular interest to those involved with future all-electric ship designs that includes both gas turbine and diesel engines as primary electrical power sources.