Sensor Data Qualification Technique Applied to Gas Turbine Engines

Abstract This paper applies a previously developed sensor data qualification technique to a commercial aircraft engine simulation known as the Commercial Modular Aero-Propulsion System Simulation 40,000 (C-MAPSS40k). The sensor data qualification technique is designed to detect, isolate, and accommodate faulty sensor measurements. It features sensor networks, which group various sensors together and relies on an empirically derived analytical model to relate the sensor measurements. Relationships between all member sensors of the network are analyzed to detect and isolate any faulty sensor within the network. I. Introduction Sensor fault detection and isolation algorithms increase the reliability of systems by enabling the early detection and removal of faulty data from the control system. Undetected sensor faults can impact the performance of closed loop systems that rely on these sensor measurements; therefore the ability to remove faulty sensor data prior to control action is desired to preserve the fidelity of the system. The National Aeronautics and Space Administration (NASA) has previously explored sensor fault detection, isolation, and accommodation algorithms for jet engines. A survey of previous techniques has been provided by Merrill (Ref. 1). Fault detection and isolation schemes rely on redundant information to determine if a sensor measurement is valid. The redundant information can be a measurement obtained from a duplicate sensor, known as physical or direct redundancy, current or successive measurement samples from a single sensor, known as temporal redundancy, or estimated measurement information produced by a reference model which describe the expected behavior, known as analytical redundancy (Ref. 2). Common temporal redundancy checks include rate checks, which ensure that the derivative of the sensor data is less than a predetermined maximum value. Analytical redundancy relies on the use of models, such as linear mapping techniques and/or observers, to produce an estimated value that is used to determine if the data is acceptable. The NASA advanced detection, isolation, and accommodation program focused on improving the overall system reliability of aircraft engines through the use of analytical redundancy methods (Refs. 3 and 4). The applied analytical redundancy method, referred to as an accommodation filter, uses a set of optimized engine estimates to isolate the faulty sensor. This approach can remove the faulty measure-ments from further use, and in some cases can replace the faulty measurements with estimated values as part of the accommodation algorithm. The estimated sensor measurements are created from a Kalman filter that incorporates a simplified engine model. Another proposed approach is to simply select a “safe” measurement when two physically redundant channels disagree (Ref. 5). Consider a fan speed controlled engine that includes two physically redundant fan speed sensors, either of which may fail high or low. The existence of a sensor fault can be detected by comparing the measurements from the two sensors, but without an independent third channel it is not possible to isolate which of the two sensors is faulty. In such a scenario, the safe accommodation action would be to select the high sensor measurement for controlling the engine. If the fan speed sensor fails low, and the lower value is selected, the engine could over speed; resulting in a potentially catastrophic turbo machinery failure. However, if the larger value is selected and the fan speed sensor fails high, the engine will decrease its speed and power, but this is not a catastrophic event. It is noted in Reference 5,