To meet the actual demand for feature extraction of vibration signals in mechanical equipment, this paper proposes an improved variational time-domain decomposition method for vibration signals. The core lies in achieving optimization through exponential adaptive weighted moving average. Specifically, this method first takes the exponential weighted moving average as the basic framework and implements adaptive adjustment of the weights. Subsequently, the Hilbert transform is carried out on the vibration signal, and the probability distribution calculation of the envelope signal is completed based on the kernel density estimation, thereby constructing an adaptive adjustment mechanism for the weights. After the signal noise reduction is completed, the variational time-domain decomposition (VTDD) technique is further adopted to decompose the processed signal. To verify the effectiveness of the method, the study conducted tests using both simulation data and the measured data from the diesel engine test bench. The results show that compared with the current mainstream noise reduction methods, the method proposed in this paper can more effectively suppress signal noise interference, and at the same time achieve precise and effective decomposition of vibration and shock signals, providing a more reliable signal basis for the subsequent condition monitoring and fault diagnosis of mechanical equipment.