An optimised method for denoising wheel flat dynamic stress signals based on multi-sensor fusion and ICEEMDAN

Purpose This paper aims to address the limitations of single-sensor feature capture and the interference of environmental noise during the dynamic stress acquisition of built-in axle box bogies for rail vehicles under wheel flat conditions, while avoiding the unintended elimination of periodic impact features by conventional denoising methods. Design/methodology/approach A targeted denoising method is proposed: Multi-Sensor Dynamic Weighted Fusion (MSDWF) fuses multi-sensor data using correlation coefficient-based dynamic weights, wavelet packet-optimized Further ICEEMDAN (F-ICEEMDAN) overcomes the fixed noise intensity constraint of the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Multi-Dimensional Feature Screening (MDFS) screens Intrinsic Mode Functions (IMFs) through entropy-weighted Median Absolute Deviation (MAD) and energy curvature detection, Impact Enhancement (IE) strengthens weak impacts through Hilbert envelope analysis. The effectiveness of the method is validated by simulations (SNR = 5/10/15 dB) and field tests (50–300 km/h), compared with wavelet denoising, Variational Mode Decomposition (VMD), and Singular Spectrum Analysis (SSA). Findings The method outperforms traditional approaches in signal unbiasedness, morphological fidelity, and fatigue damage calculation accuracy (minimal Miner damage deviation (ΔD)), enhancing dynamic stress signal reliability for precise bogie structural fatigue life evaluation. Originality/value Targeting dynamic stress signals of built-in axle box bogies under wheel flat conditions, this study innovatively integrates four modules (MSDWF, F-ICEEMDAN, MDFS, IE). It breaks the contradiction between denoising and impact feature retention in traditional methods via adaptive noise injection, multi-dimensional IMF screening and weak impact enhancement, ensuring efficient retention of key impact features for reliable fatigue evaluation.