Applied neural network for navy marine gas turbine stall algorithm development

摘要

In June 2005, Naval Surface Warfare Center (NSWC) Gas Turbine Emerging Technologies conducted testing on a general electric LM2500 gas turbine engine. This engine is the main propulsor for DDG-51 and CG-47 class United States Navy surface ships. The purpose of this testing was to induce compressor stall in order to evaluate existing algorithms for stall prediction and gather data for further algorithm development. In addition to existing sensor data, dynamic pressure sensors, with data rates ranging from 20-1000 KHz, were installed in various compressor stages for additional capability. Utilizing the data collected, in conjunction with a MATLAB-based neural network approach, NSWC has developed algorithms to detect and trend stall margin and related quantities that can eventually be used in an early stall warning system onboard ship. Algorithms can be incorporated into the recently installed full authority digital control, allowing real-time stall detection and prevention. This paper discusses the feasibility of employing a neural network approach to detect and output a compressor stall margin value and associated risk of compressor stall for U.S. Navy LM2500 gas turbine engines.