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

Abstract

This paper presents a method for the detection and isolation of single gas turbine sensor faults, in presence of model inaccuracy and measurement noise. The method uses a fault matrix with a column-canonical structure (i.e., each matrix column having the same number of zeroes, but in different positions), in order to obtain the unambiguous fault isolation. The fault matrix was obtained by using a number of ARX (Auto Regressive exogenous) MISO (Multi-Input/Single-Output) models equal to the number of measured gas turbine outputs, each model calculating an estimate of one measurable output as a function of other inputs or outputs measured on the machine. Moreover, in order to reduce the threshold of fault detection and, therefore, the minimal detectable faults, digital filters were used, applied to the time series of data measured on the machine and computed by the models. Finally, tests were performed in order to find the minimal sensor faults that can be detected and isolated.