Multichannel signals usually contain more abundant information compared to single channel signal. However, their phases are not always synchronous, resulting in the difficulty of multichannel signals fusion and comprehensive feature extraction. To adequately employ the information contained in multichannel signals, this paper presents a novel multichannel phase synchronization fusion variational mode decomposition (MFVMD) for mechanical fault diagnosis. First, the method involves initially utilizing higher-order singular value decomposition (HOSVD) to synchronously separate the estimated noise of each channel from the original signals, reducing noise interference in subsequent steps. Second, addressing the potential phase desynchronization among multichannel signals, this study achieves phase synchronization compensation for multichannel signals based on multi-component mean phase coherence (MCMPC). Third, analyzing each channel of multichannel signals one by one ignores the joint fault information among channels. Therefore, fusing multichannel signals into one integrated signal for analysis can be an available approach. Due to the feasibility of multichannel signal fusion after phase synchronization, this paper utilizes local estimation of estimated noise cross-covariance to perform distributed estimation fusion of multichannel signals. Next, in cases where multiple fault characteristic information is superimposed and difficult to separate, variational mode decomposition (VMD) is applied to achieve adaptive signal processing and feature extraction. Finally, the various fault characteristic information by MFVMD is refined and separated for fault diagnosis. An engineering case involving a compound fault diagnosis of gearbox for finishing mills is employed to validate the effectiveness of MFVMD, compared with VMD, multivariate variational mode decomposition (MVMD), and successive variational mode decomposition (SVMD). The analyzed results show the significant performance of the proposed method, highlighting its valuable tool for mechanical fault diagnosis in multichannel signals.