Aiming at the difficulty to diagnose the early unbalance fault on the exciting force of a linear vibrating screen, a fault diagnosis method based on variational mode decomposition(VMD) and recurrence quantitative analysis(RQA) was proposed.Firstly, the vibration signal was decomposed by VMD, the fundamental frequency signal of the linear vibrating screen was separated, and the submerged high frequency components were obtained.Then, the dynamic characteristic recurrence plots of different signal components were drawn, and the quantitative indexes of the recurrence plots were calculated to constitute a nonlinear and nonstationary evaluation feature vector of the fault signal.Finally, the high-dimensional eigenvectors were input into a machine learning classifier for diagnosis, and its results were compared with those of traditional feature extraction methods.The results show that, the characteristic parameters extracted by the method has the highest recognition accuracy in the excitation force unbalance fault experiment, and the comprehensive recognition rate is 99.13%.In addition, applying the method to the fault diagnosis of   rotating machinery bearings, the comprehensive recognition rate reaches 99.38%, indicating that the method has certain reliability and engineering application value.