ARX Linear Model Set-Up for Fault Diagnosis of Gas Turbine Sensors

Abstract

The diagnosis of gas turbine sensor faults requires models of the system to calculate estimates of the measured output system variables. The model set-up phase is of great importance since the reliability of the diagnostic tool depends on the model accuracy. In the paper two different methodologies of I/O linear model set-up are analyzed and compared to find the more simple and general one. The first methodology consists in obtaining the I/O linear models by directly linearizing the physical laws (system modeling). The second one uses statistical methods (system identification) to calculate model parameters from time series data measured on the machine. The models used are of the ARX (Auto Regressive with eXternal input) type. The number of models and the measured variables correlated by each of them have been determined in order to obtain unambiguous fault signatures for each sensor. The system identification method proves to be more suitable to the system modeling because of its greater simplicity in the fault diagnosis application.

A Method for the Diagnosis of Gas Turbine Sensor Faults in Presence of Measurement Noise

“…With respect to previous Author's papers (Bettochi et al, 1996b;Bettocchi and Spina, 1997) the technique presented in this work is more reliable towards false alarms due to fault signature degradation in presence of model inaccuracy and measurement noise.…”

“…The residual calculation was performed in this paper using the parity equations (Bettocchi and Spina, 1997), whose expression is shown in Appendix. These equations link the residuals to the measured values on the machine by means of a model coefficient matrix.…”

“…The equations which express these differences and, so, which allow the residual calculation are called "parity equations. Their expression in z-domain is the following (Bettocchi and Spina, 1997): e(z) = F(z) x(z) where: e(z) = [e h , e n ] T is the residual vector,…”

Sensor fault diagnosis of gas turbine engines using an integrated scheme based on improved least squares support vector regression

“…17 Majority of modelbased sensor fault diagnosis schemes rely on linear models. 18 For example, Kobayashi and Simon 19 proposed a bank of Kalman filters based on linear models for developing an in-flight fault detection and isolation system for aircraft engine sensors in conjunction with the detection of actuator and component faults. Usually, methods based on linear models are considered to be effective algorithms at steady operating states, but they cannot reflect all characteristics of the engine throughout the flight envelope.…”

Kalman Filtering to Enhance the Gas Turbine Control Sensor Fault Detection

“…Minimum detectable faults by monitoring residual white ness.In order to compare improvements with this fault de tection and isolation technique, the minimal detectable faults obtained in a previous work 4 by using model-based observers and geometrical analysis of residuals are quoted inTable (5).It ensues that the values of the faults obtained by using statistical tests on Kalman filter innovations, col lected in…”