It is difficult to distinguish the faults due to the high similarity of different faults of the cascaded H-bridge multi-level inverter. To cope with such a problem

a novel feature extraction method based on orthogonal globality and locality preserving embedding（OGLPE） was proposed

in which both global variance maximum transformation and local structure preservation are taken into consideration

and it was combined with ensemble empirical mode decomposition（EEMD） to perform multi-scale fault feature extraction on H-bridge multi-level inverter. By taking seven-level inverter IGBT open fault feature extraction as an example

firstly

the original three-phase current signal were collected

and then the current deviation signal could be obtained by combining the original current signal and the standard signal. Secondly

the current deviation signal of each state was decomposed by EEMD

and the intrinsic mode function（IMF） components at different scales were obtained. On that basis

the fault feature set could be constructed by performing OGLPE on each IMF. Finally

the proposed method was compared with the commonly used feature extraction methods. The two-dimensional visualization results show that the proposed method can effectively improve the discrimination between classes of different faults

and the diagnostic results report that the diagnostic accuracy of the proposed method is as high as 98.4%.