Applicability of Simplified Data-Driven Models in Gas Turbine Diagnostics

Abstract

Abstract Gas turbine engine diagnostic algorithms based on gas path measurements constitute the principal subsystem of an engine monitoring system. These algorithms follow two main approaches. The first of them called a Gas Path Analysis (GPA) relies on a physics-based model (thermodynamic model) to transform gas path measurements into engine components’ health parameters. Within the second approach, diagnosis is made in the space of the deviations of measurements from an engine baseline using the pattern recognition theory. This approach mainly employs data-driven models and does not directly use the thermodynamic model. But the thermodynamic model helps to generate the necessary input data. The thermodynamic model presents a complex software exigent to computer resources. Moreover, the model computation cyclic procedure does not converge sometimes. To simplify diagnostic algorithms and make them reliable, a linear model was introduced in the first GPA studies. Since the linear model is not accurate enough, surrogate nonlinear data-driven models were proposed later. The present paper deals with the thermodynamic model and three simplified data-driven models for the steady-state operation of an industrial aero-derivative gas turbine. The paper focuses on the influence of the model choice on the final diagnostic reliability. Not only the reliability level is studied with these models, but also the influence of diagnostic conditions is investigated. In this way, the applicability of simplified data-driven models is validated. We hope that the paper results will stimulate the design of simple, but reliable diagnostic solutions and general progress in the development of real monitoring systems.