A novel fault diagnosis method based on compound feature selection and parameter optimization of ELM is presented.
Compound features which consist of time-frequency features, EEMD energy features and EEMD singular features are extracted.
The compound feature set and parameters of ELM are optimized simultaneously by using a hybrid GSA.
Results show that HGSA-ELM achieves high accuracy compared with the original ELM and methods in literatures.