Diagnosis and Supervision of Industrial Gas Turbines

Monitoring of industrial gas turbines is of vital importance, since it gives valuable information for the customer about maintenance, performance, and process health. The performance of an industrial gas turbine degrades gradually due to factors such as environment air pollution, fuel content, and ageing to mention some of the degradation factors. The compressor in the gas turbine is especially vulnerable against contaminants in the air since these particles are stuck at the rotor and stator surface. The loss in compressor performance, due to fouling, can partially be restored by an on-line/off-line compressor wash. If the actual health state of the gas turbine is known, it is possible to efficiently plan the service and maintenance and thereby reduce the environmental impact and the fuel cost for the customer. A thermodynamic gas turbine modeling package, called GTLib, is developed in the equation-based object-oriented modeling language Modelica. Using the GTLib package, a gas turbine model can be constructed. The gas turbine model can be used for performance calculation and as a base when diagnosis tests are generated. These tests can be used in a diagnosis and supervision system to detect compressor fouling and abrupt sensor faults. One of the benefits with using GTLib is the ability to model a lean stoichiometric combustion at different air/fuel ratio. Using the air/fuel ratio concept, an arbitrary number of gas species in the in-coming air can be considered. The number of equations is reduced if the air/fuel ratio concept is considered instead of modeling each gas species separately. The difference in the number of equations is significant if many gas species are considered. When the gas turbine components deteriorate, a mismatch between the nominal performancemodel and themeasurements increase. To handle this, the gas turbinemodel is augmented with a number of estimation parameters. These estimation parameters are used to detect slow deterioration in the gas turbine components and are estimated with a Constant Gain Extended Kalman Filter (CGEKF). The state estimator is chosen using structural methods before an index reduction of the model is performed. Experimental data is investigated and it is shown that the performance degradation due to compressor fouling can be estimated. After the compressor is washed, the performance of the compressor is partially restored. An abrupt sensor fault of 1 % of the nominal value is introduced in the discharge temperature of the compressor. The sensor fault can be detected using the CUSUM algorithm for change detection. Finally, the overall thesis contribution is the calculation chain from a simulation model used for performance calculation to a number of test quantities used in a diagnosis and supervision system. Since the considered gas turbine model is a large non-linear DAE model that has unobservable state variables, the test construction procedure is automatically performed with developed parsers.