Fault diagnosis of gas turbines with thermodynamic analysis restraining the interference of boundary conditions based on STN

Abstract

Gas turbines are widely used in natural gas-fired power generation and long-distance natural gas pipelines. The complexity and diversity of boundary conditions usually have a great influence on the measurement parameters, which makes them difficult to reflect the actual health status of gas turbines and leads to lower diagnostic accuracy. In the gas path analysis of gas turbines, traditional mechanism and data-driven methods are difficult to consider the interference of boundary conditions. A novel fault diagnostic method deriving from the spatial transformer network (STN) is proposed to restrain the influences of boundary conditions on fault diagnostic results. It can realize fault detection with higher accuracy by converting different boundary conditions to a designed one.The established STN is then combined with the typical small deviation (SD) theory and a classification network to complete the gas path analyses.The proposed two methods are verified and compared with traditional methodsthrough simulation experiments and field data. Compared with the traditional SD method, the fault detection accuracy is increased by 2.92% in the STN-SD method, and the STN-MLP method improves the diagnostic accuracy by 5.24%. The proposed two methods also have good performance in the field data analysis and realize fast fault detection and early warning.This work provides a reliable way to detect faults subjecting tothe influence of complex boundary conditions and provides reference and support for the fault diagnosis of other thermodynamic systems.