An engine fault diagnosis method based on improved variational mode decomposition (VMD) and dual measure fractal dimension was proposed. The mutual information method was first used to extend the end of cylinder head vibration signal, VMD algorithm was then used to decompose the extended signal into several intrinsic mode functions (IMFs), and the purpose of suppressing the end effect of VMD and improving signal decomposition precision were realized. The orthogonal transform method was further used to orthogonalize each IMF component. The given time scale sequence   was divided into the first and second scale intervals according to the cut point determined by adaptive selection. The fractal dimension of signal was calculated with the orthogonalized IMF components separately in the two scale intervals and the dual measure fractal dimension was hence obtained that describes the detail information and trend information respectively. Finally, the dual measure fractal dimension was used as the input for the classification model of extreme learning machine to realize engine fault diagnosis. The simulation and experimental results show that the proposed method can effectively suppress the end effect of VMD and improve the signal decomposition accuracy. The dual measure fractal dimension has good intraclass aggregation and interclass dispersion, which improves the accuracy of engine fault diagnosis.