An integrated data processing strategy for pavement modulus prediction using empirical mode decomposition techniques

Data collection for infrastructure health monitoring using embedded sensors is often hindered by noise contamination and inconsistencies in sensor measurements. These challenges are exacerbated by variations in data features across different sensors, complicating the analysis and interpretation process. A comprehensive data processing strategy capable of mitigating noise, harmonizing feature discrepancies, and extracting latent information is essential for enhancing data-based analysis and modeling. This study introduces an integrated data processing strategy combining Empirical Mode Decomposition (EMD) techniques with adaptive Intrinsic Mode Function (IMF) classification to improve the prediction of pavement dynamic modulus. Various EMD methods were applied to decompose signals from wireless embedded sensors, using Maximum Normalized Cross-Correlation (MNCC) and Signal-to-Noise Ratio (SNR) as indices in a K-means clustering process to select effective IMFs. Results show that the ensemble EMD (EEMD) technique effectively captures critical mechanical response information while expanding data dimensionality, leading to enhanced prediction accuracy. Consequently, the integrated EEMD and K-means clustering approach is recommended as a powerful tool for infrastructure data processing and predictive modeling.