Condition monitoring of gas-turbine power units using the Derivative-free nonlinear Kalman Filter

A method is developed for diagnosing faults and cyberattacks in electric power generation units that consist of a gas-turbine and of a synchronous generator. By proving that such a power generation unit is differentially flat its transformation into an input-output linearized form becomes possible. Moreover, by applying the Derivative-free nonlinear Kalman Filter state estimation for the power unit is performed. The latter filtering method, consists of the Kalman Filters recursion on the linearized equivalent model of the power unit, as well as of an inverse transformation providing estimates of the initial nonlinear system. By subtracting the estimated outputs of the Kalman Filter from the measured outputs of the power unit the residuals sequence is generated. The residuals undergo statistical processing. It is shown that the sum of the squares of the residuals vectors, weighted by the inverse of the associated covariance matrix, forms a stochastic variable that follows the chi(2) distribution. By exploiting the statistical properties of this distribution, confidence intervals are defined, which allow for detecting the power units malfunctioning. As long the aforementioned stochastic variable remains within the previous confidence intervals the normal functioning of the power unit is inferred. Otherwise, a fault or cyberattack is detected. It is also shown that by applying the statistical method into subspaces of the system's state-space model, fault or cyberattack isolation can be also performed.