Bridging Simulation and Experiment: A Self-Supervised Domain Adaptation Framework for Concrete Damage Classification

TL;DR

This study tackles concrete damage classification from coda wave signals under a sim-to-real domain shift by combining a physics-based simulation platform with a self-supervised domain adaptation framework. The method integrates domain-adversarial learning (DANN) and MCC losses with BYOL self-supervision to transfer knowledge from synthetic data to unlabeled experimental data. Empirical results show pronounced improvements over a plain 1D CNN and six DA baselines, achieving an accuracy of 0.776 and macro F1 of 0.771, with strong robustness across runs. The approach is end-to-end and computationally efficient, underscoring its potential for practical SHM applications in concrete structures. Limitations include focus on compressive damage and the need to extend to tensile damage and in-situ monitoring scenarios.

Abstract

Reliable assessment of concrete degradation is critical for ensuring structural safety and longevity of engineering structures. This study proposes a self-supervised domain adaptation framework for robust concrete damage classification using coda wave signals. To support this framework, an advanced virtual testing platform is developed, combining multiscale modeling of concrete degradation with ultrasonic wave propagation simulations. This setup enables the generation of large-scale labeled synthetic data under controlled conditions, reducing the dependency on costly and time-consuming experimental labeling. However, neural networks trained solely on synthetic data often suffer from degraded performance when applied to experimental data due to domain shifts. To bridge this domain gap, the proposed framework integrates domain adversarial training, minimum class confusion loss, and the Bootstrap Your Own Latent (BYOL) strategy. These components work jointly to facilitate effective knowledge transfer from the labeled simulation domain to the unlabeled experimental domain, achieving accurate and reliable damage classification in concrete. Extensive experiments demonstrate that the proposed method achieves notable performance improvements, reaching an accuracy of 0.7762 and a macro F1 score of 0.7713, outperforming both the plain 1D CNN baseline and six representative domain adaptation techniques. Moreover, the method exhibits high robustness across training runs and introduces only minimal additional computational cost. These findings highlight the practical potential of the proposed simulation-driven and label-efficient framework for real-world applications in structural health monitoring.

Figures (17)

• Figure 1: Workflow of the virtual testing platform for damage assessment in concrete structures.
• Figure 2: Schematic of the proposed self-supervised domain adaptation framework for concrete damage classification using coda wave signals. The architecture integrates supervised learning on simulated data with domain adversarial training, minimum class confusion (MCC), and self-supervised BYOL strategy on experimental data. These components work synergistically to align feature distributions and improve generalization across domains.
• Figure 3: Architecture of a 1D CNN model consisting of a convolutional feature extractor $F$ and an MLP classifier $C$.
• Figure 4: Detailed architecture of the domain adversarial training module within the proposed framework, including a feature extractor $F(\cdot; w_F)$, a classifier $C(\cdot; w_C)$, and a domain discriminator $D(\cdot; w_D)$. Forward propagation paths are shown in black, while colored arrows indicate component-specific gradient flows during backpropagation. Detailed layer-wise configurations for convolutional (Conv1d) and fully connected (FC) layers are provided in Table \ref{['tab:nnparams']}.
• Figure 5: Overview of the Minimum Class Confusion (MCC) loss computation.