In this paper, a denoising method of variational mode decomposition and permutation entropy is proposed, the key parameters and thresholds in permutation entropy are analyzed and set, and then the decomposition layer value of variational mode decomposition is determined by permutation entropy, and the decomposed modes are reconstructed to achieve denoising of vibration signals. The advantages of this method in orthogonality, completeness, signal-to-noise ratio, and efficiency are verified by simulation tests. Finally, the actual vibration signals collected by the system are denoised. The experimental results show that, compared with the existing empirical mode decomposition-correlation coefficient and full empirical mode decomposition-correlation coefficient methods, the proposed method has the best denoising signal-to-noise ratio (the ratio of noise signal to noise reduction) for three kinds of vibration signals (contact, wheel rolling, and rain), which are 32.5358 dB, 30.5546 dB, and 29.3435 dB, respectively, and the time-consuming is also less, which is 1.4432 s, 1.6320 s, 1.2349 s, respectively, and the accuracy of signal pattern recognition is the highest, all above 99%.