The Denoising Method for MEMS Acceleration Signals Based on CEEMDAN-ASA

To address the susceptibility of MEMS inertial sensor acceleration signals to noise and vibration in rail track monitoring, a denoising method is proposed. First, the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is employed to decompose the acceleration signals. Trend Intrinsic Mode Functions (IMFs) are then identified and removed using correlation coefficients. The remaining IMFs are further refined by iteratively optimizing denoising thresholds through the Adaptive Simulated Annealing (ASA) algorithm. Finally, the selected components are reconstructed. Experimental results show that this method significantly outperforms traditional denoising approaches in terms of mean square error and signal-to-noise ratio.