Advanced gas-path fault diagnostics for stationary gas turbines

The reliabilities of the gas-path components (compressor, burners and turbines) of a gas turbine (GT) are usually high when compared with those of other GT systems such as fuel supply and control. However, in the event of forced outage, downtimes are normally high, giving a relatively low availability. The purpose of condition monitoring and fault diagnostics is to detect, isolate and assess (i.e. estimate quantitatively the magnitude of) the faults within a system, which in this case is the gas turbine. An effective technique would provide a significant improvement in economic performance, reduce operational and maintenance costs, increase availability and improve the level of safety achieved. However, conventional analytical techniques such as gas-path analysis and its variants are limited in their applications to engine diagnostics due to several reasons that include their inability to:- operate effectively in the presence of noisy measurements; distinguish effectively sensor bias from component faults; preserve the nonlinearity in the gas-turbine parameter relationships; and the requirement for more sensors for achieving accurate diagnostics. The novelty of this research stems from its objective of overcoming most of these limitations and much more. In this thesis, we present the approach adopted in developing a diagnostic framework for the detection of faults in the gas-path of a gas turbine. The framework involves a large-scale integration of artificial neural networks (ANNs) designed and trained to detect, isolate and assess the faults in the gas-path components of the engine. Input to the diagnostic framework are engine measurements such as spool speeds, pressures, temperatures and fuel flow while outputs are either levels of changes in sensor(s) for the case of sensor fault(s) or the level of changes in efficiencies and flow capacities for the case of faulty components. The diagnostic framework has the capacity to assess both multiple component and multiple sensor faults over a range of operating points. In the case of component faults, the diagnostic system provides changes in efficiencies and flow capacities from which interpretations can be sought for the nature of the physical problem. The implication of this is that the diagnostic system covers a wide range of problems - both likely and unlikely-. The technique has been applied to several developed test cases, which are not only thermodynamically similar to operational engines, but also covers a range of engine configurations and operating conditions. The results obtained from the developed approach has been compared against those obtained from linear and nonlinear (recursive linear) gas-path analysis, as well as from the use of fuzzy logic. Analysis of the results demonstrates the promise of ANN applied to engine gas-path fault diagnostic activities. Finally, the limitations of this research and direction for future work are presented.