Data-Driven Fault Diagnosis of Once-through Benson Boilers

Abstract

Fault diagnosis (FD) of once-through Benson boilers, as a crucial equipment of many thermal power plants, is of paramount importance to guarantee continuous performance. In this study, a new fault diagnosis methodology based on data-driven methods is presented to diagnose faults in once-through Benson boilers. The present study tackles this issue by adopting a combination of data-driven methods to improve the robustness of FD blocks. For this purpose, one-class versions of minimum spanning tree and K-means algorithms are employed to handle the strong interaction between measurements and part load operation and also to reduce computation time and system training error. Furthermore, an adaptive neuro-fuzzy inference system algorithm is adopted to improve accuracy and robustness of the proposed fault diagnosing system by fusion of the output of minimum spanning tree (MST) and K-means algorithms. Performance of the presented scheme against six major faults is then assessed by analyzing several test scenario.