Quantifying Uncertainty of Gas Turbine Engine Models Generated Using Inverse Solution Methods

In the current economic and political environment, there is a push for gas turbine operators to achieve higher operating efficiencies, which in turn, reduces emissions and fuel consumption. As these owners and operators seek to increase the efficiency of their machines, they are increasingly turning to physics-based performance modeling. This allows the end user to analyze machine performance, plan for performance upgrades, and evaluate use cases and operating conditions not originally envisioned by the original equipment manufacturers (OEMs). For owners/operators who do not have access to physics-based models provided by the hardware OEM or would like to evaluate modifications to legacy hardware, physics-based models may be developed using measured turbine performance data and high-level knowledge of the turbine architecture. In previous work, a physics-based performance model of an industrial gas turbine engine was created using measured plant operating data and an inverse solution method to allow off-design exploration of its performance. However, this model’s uncertainty was unknown, and knowledge of uncertainty is crucial to understanding a model’s reliability. In the present work, the model’s uncertainty in predicted performance at a particular operating point is investigated using statistical methods. Polynomial regressions of standard deviation are used alongside the performance regressions to describe the uncertainty at various operating points. These regressions are also used to visualize the variation of uncertainty across the performance map. Such knowledge of uncertainty can aid gas turbine operators in decision making with regard to the risks of off-design operation or equipment modifications.