In order to make full use of synchronously sampled vibration signals for the fault diagnosis of mechanical equipment, a method for building time-frequency coherence network of vibration signals and extracting the multifractal features of the network is proposed. It takes each vibration signal as a node. According to the physical issues of concern, the nodes are connected to form a network with proper structure. Then, cross wavelet transform is performed for each pair of adjacent nodes in the network, and the time-frequency coherence spectra are obtained. Wavelet leaders are used to estimate the multifractal spectra of the time-frequency coherence spectra. The morphological features of the multifractal spectra are extracted by curve fitting. Finally, using feature fusion and dimension reduction methods, all the features obtained are fused with reduced dimension, and the final features of the whole network are obtained. This method proposes a framework for the extraction of time-frequency coherence multifractal features of multiple vibration signals. It has been successfully applied to the crack fault diagnosis for the automatic mechanism of an antiaircraft machine gun, and has a wide scope of applications.