Generating a virtual physical model through measurement data and reverse engineering: Applying a performance prediction model for an industrial gas turbine during start-up

Abstract

The gas turbine virtual physical model can be applied to various fields such as performance optimization, diagnostics, and prognostics. For this purpose, it is important to increase the accuracy of the physical model. However, in order to generate such a model, the design information and off-design performance characteristics of its components are required, yet it is difficult to obtain this information because manufacturers do not generally release it. This study proposes a method for the generation of a virtual physical model to predict gas turbine system performance through reverse engineering, based on operational data but without information on component performance or design information; its effectiveness is discussed. There have been many studies to correct an initial gas turbine’s physical model using operational data. In the existing studies, detailed component performance and design information was required, to generate an initial physical model. It is normally difficult to obtain component performances and design information. Therefore, a method is proposed to generate a physical model at a reference point, using operating data and assumed information, and to create a final physical model through component performance map correction. An initial physical model is constructed based on the information and performance characteristics used for general gas turbine components, not design information of the target engine. The component blocks consisting of the gas turbine are simplified by combining them at each region. In addition, a method of estimating the moment of inertia of a gas turbine’s rotating parts, using operating data, was proposed. The target engine is a single-shaft industrial gas turbine consisting of a compressor, combustor, and turbine. A virtual physical model was created using one of the three datasets measured in the start-up operation; the accuracy of the generated model was verified using the other two datasets. Shaft speed, fuel mass flow rate, compressor exit temperature, compressor total pressure ratio, and exhaust gas temperature measurement data were used. As a result, it was confirmed that, for both the data used to generate the model and the data that were not, the model predicted the measured variables well. It is expected that the proposed method can be used to create a virtual physical model for use in operation optimization and diagnostics from the operator's point of view, where it is difficult to obtain component performance and design information.