Turbofan engine demonstration of sensor failure detection

In the paper, the results of a full-scale engine demonstration of a sensor failure detection algorithm are presented. The algorithm detects, isolates, and accommodates sensor failures using analytical redundancy. The experimental hardware, including the F100 engine, is described. Demonstration results were obtained over a large portion of a typical flight envelope for the F100 engine. They include both subsonic and supersonic conditions at hoth medium and full, nonafter burning, power. Estimated accuracy, minimum detectable levels of sensor failures, and failure accommodation performance for an F100 turbofan engine control system are discussed. HE objective of the advanced detection, isolation, and accommodation (ADIA) program was to improve demonstrated reliability of digital electronic control systems for turbine engines by detecting, isolating, and accommodat- ing sensor failures using analytical redundancy methods. This paper discusses the results of an engine demonstration of an analytical-redundancy-based algorithm developed as part of the ADIA program. Over the past 45 years, hydromechanical implementations of turbine engine control systems have ma- tured into highly reliable units. However, there is a trend toward increased engine complexity. Engine control has be- come increasingly complex and has evolved from a hydrome- chanical to a full authority digital electronic (FADEC) imple- mentation. These FADEC-type controls have to demonstrate the same or improved levels of reliability as their hydrome- chanical predecessors. Various redundancy management techniques have been ap- plied to both the total control system and to individual compo- nents. Studies1 have shown that the least reliable of the control system components are the engine sensors. Sensor redundancy will be required to achieve adequate control system reliability. One important type of sensor redundancy is analytical redun- dancy2 (AR), which uses a model to generate redundant infor- mation that can be compared to measured information to detect failures. AR-based systems can have cost and weight savings over hardware redundancy. Considerable progress has been made in the application of analytical redundancy to improve turbine engine control sys- tem reliability. However, little has been done to demonstrate AR-based techniques on real engine systems. One exception was a flightiest program3 where selected out-of-range (hard) sensor faults were induced and the resulting actions of the control evaluated. The test program included a ground-thrust stand evaluation and a flight test. The sensors that failed during the flight test included the compressor inlet variable geometry sensor, inlet static pressure, burner pressure, and fan turbine inlet temperature. Most failures were detected and accommodated. However, a recreation of a broken line (hard) burner pressure went undetected. Pilot response to aircraft performance after accommodation was favorable.