A Feature Weighted Kernel Extreme Learning Machine Ensemble Method for Gas Turbine Fault Diagnosis
Abstract Fault diagnosis is a very important section of gas turbine maintenance. Kernel extreme learning machine (KELM), a novel artificial intelligence algorithm, is a potentially effective diagnosis technology. The existing KELMs are all assumed that there is the same influence to the optimal separating hyperplane from all features, which reduces its generalization performance. In this study, a feature weighted kernel extreme learning machine ensemble method (FWKELM-RF) is developed for application in the field of gas turbine fault diagnosis. First, information gain ratio is introduced to assign different weights to the feature space. Furthermore, random forest is used to enhance stable performance of feature weighted KELM. The fault datasets from a gas turbine with three shafts is generated to validate the performance of the developed method, and the results demonstrate that FWKELM-RF can achieve better accuracy and stability for detecting fault in gas turbine.