Answers are grounded in this paper and cite their sources.
The long-term prediction of concrete deterioration, such as the depassivation and pitting corrosion of embedded rebars induced by chloride ingress, requires accurate numerical simulation of spatial field evolution within the concrete matrix. Existing data-driven approaches, notably classic Dynamic Mode Decomposition, often discard spatial dimensionality information during vectorization, leading to an increased computational complexity and potential loss of prediction accuracy. Therefore, a novel modeling framework is proposed based on Tensor-Train Dynamic Mode Decomposition to significantly enhance computational performance while retaining the physical interpretability of the classic Dynamic Mode Decomposition in the form of spatial modes and temporal frequencies. By dimension-wise singular value decomposition, the proposed framework approximates high-dimensional snapshot tensors into a sequence of low-rank interconnected cores, thereby shifting the prediction-related computation into a compressed latent space. The method was numerically implemented and validated against Finite Element analysis results across three diverse chloride ingress scenarios, under the assumption of a homogeneous and uncracked concrete domain. Within the investigated modeling cases, the proposed method achieves a fundamental shift from polynomial-to-linear scaling of computational complexity. For these cases, the framework reduces the degrees of freedom to within 0.4% of the original data dimensionality. While maintaining a relative L1 error under 0.12, this method provides a computational speedup of up to 4.12 times compared to classic Dynamic Mode Decomposition. This advantage becomes increasingly prominent with growing model scales; for a 3D grid with a side length of 2000 units, the method reduces complexity by over eight orders of magnitude. The compatibility of the proposed framework with data represented in a polar coordinate system is also demonstrated. Furthermore, its robustness to noise is confirmed by its ability to maintain stable predictive accuracy even under severe measurement noise, as the prescription of fixed truncation ranks for the tensor cores effectively acts as an implicit low-pass filter. To support reproducible research, an open-access interactive online solver has been deployed on the Hugging Face platform, enabling the engineering community to perform real-time deterioration forecasting.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: