Adaptive forecasting of stochastic crack growth using empirical mode decomposition: Gaussian process regression for structural health monitoring
Sheng Jiang, Yufei Wang, Zifeng Cheng, Baixi Chen
Abstract
Accurate forecasting of stochastic crack propagation remains a fundamental challenge in structural health monitoring (SHM). Prediction errors not only directly affect maintenance scheduling and structural safety assessments but also indirectly influence reliability-based design and digital twin applications. To address this challenge, this study proposes a framework that integrates empirical mode decomposition (EMD) for noise reduction, Gaussian process regression (GPR) for probabilistic crack growth prediction with uncertainty quantification, and updating strategy for dynamic model refinement. Trained in validated phase-field simulation dataset, the proposed framework achieves superior accuracy and robustness compared to state-of-the-art approaches, particularly in handling noisy signals and quantifying predictive uncertainty, illustrating its potential as a surrogate for real-world crack behaviour. While current validation is limited to simulated datasets, the framework lays the groundwork for future application to experimental and field-collected SHM data, thereby supporting its eventual integration into digital twin platforms that demand real-time and uncertainty-aware predictions.