To improve the noise reduction accuracy of reciprocating friction vibration signals of piston ring friction pairs in marine diesel engines, this paper proposes a collaborative noise reduction method that integrates complementary ensemble empirical mode decomposition (CEEMD) and an adaptive correlation coefficient screening mechanism. Simulated vibration signals were obtained using the BRUKER UMT friction and wear testing machine to construct an experimental dataset. The original signals were subjected to multi-scale decomposition using CEEMD, and effective intrinsic modal components were screened using an adaptive correlation coefficient threshold. The dominant noise component was removed to achieve signal reconstruction. MATLAB software was applied to implement the noise reduction processing. Three types of indicators—signal-to-noise ratio (SNR), normalized cross-correlation coefficient (NCC), and mean square error (MSE)—were used for quantitative evaluation. Multi-scale permutation entropy (MPE) theory was also innovatively introduced to verify the dynamic characteristics of the denoised signal. The experimental results show that the CEEMD adaptive correlation coefficient screening method has significantly improved noise reduction performance compared to other methods. The correlation coefficients between the multi-scale permutation entropy image of the stripped noise signal and the overall laboratory noise signal image are all above 0.8, thus proving the accuracy of effective signal denoising.