Answers are grounded in this paper and cite their sources.
Acoustic emission (AE) monitoring is widely used for concrete damage identification, but AE descriptors are often interpreted independently from the stress–strain response, which weakens cross-specimen transfer under small-sample destructive testing. To address this issue, this study proposes a curve-conditioned physics-guided learning framework for damage evolution identification in multi-scale fibre-reinforced rubberized concrete under compression. Irregular event-level AE tables are converted into synchronized loading-window sequences, and each window is coupled with curve-derived mechanical variables. A curve-conditioned physics-guided temporal network is further developed with AE tokens, curve tokens, cross-attention fusion, monotonic latent damage evolution, and multi-task supervision. Leave-one-specimen-out experiments show that AE-only models generalize poorly, whereas curve-conditioned learning substantially improves damage-stage recognition. The best tabular curve-conditioned baseline achieves 0.928 accuracy and 0.905 macro-F1, while the full CC-PGT model further improves the results to 0.987 accuracy, 0.977 macro-F1, 0.023 damage RMSE, and 0.017 damage MAE. These results demonstrate that stress–strain curve conditioning provides an effective and mechanically interpretable route for small-sample AE damage assessment.
No takes yet. Share an insight, caveat, or question.
Dong et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: