Abstract: Rolling bearings are important parts of rotating machinery. When the early failure occurs, it is difficult to effectively extract the weak fault features. Aiming at this problem, an early fault diagnosis method of variational mode decomposition (VMD) of optimizing the parameter K value was proposed. First, the instantaneous frequency mean judgment method was used to determine the value of modal number K, and then the fault diagnosis signal was processed by VMD method. By analyzing the intrinsic modal function components obtained by decomposing the fault signal of the bearing, the sensitive components were obtained for the envelope demodulation analysis to judge the fault type and severity of the bearing. Finally, the results obtained by the EMD and VMD algorithm were compared. Results show that the optimized VMD algorithm can successfully extract the early fault features of the bearing and achieve the diagnosis of early bearing failure.