Fault isolation with principal components structured models for a gas turbine

This paper deals with the fault isolation issue in a gas turbine unit of a combined cycle power plant where the residual evaluation is addressed using data driven based structured models. The design of the isolation system is performed considering a) the redundant graphs of the gas turbine obtained from a non-linear complex dynamical model with 37 algebraic and differential equations with 15 known variables; and b) a data driven method to generate the corresponding residuals. The structured residuals subspaces are generated via dynamic principal component analysis with adaptive standardization from sets of nominal historical data. Thus, the proposed integration of data-driven methods with structural analysis to generate the residuals does not require parameterized models which are difficult to obtain for large scale systems, as the gas turbine of a power plant.