A METHOD FOR GENERATING SYNTHETIC IMPACT ECHO IMAGES IN STRUCTURAL HEALTH MONITORING

Structural Health Monitoring of civil infrastructures often relies on effective and interpretable assessment techniques such as Impact Echo. Traditional Impact Echo methods require substantial manual effort and specialized equipment, limiting their scalability and ease of deployment. In this paper, a novel deep learning pipeline designed to approximate the Impact Echo function starting from strain gauge sensor measurements collected from bridge deck is proposed. By leveraging high-fidelity data provided by The BEAST (Bridge Evaluation and Accelerated Structural Testing) facility, our pipeline transforms these measurements into interpretable Impact Echo images, producing output patches of size 128 × 128 pixels that effectively visualize structural integrity. Addressing the prevalent challenge of limited datasets in Structural Health Monitoring, a custom algorithm for synthetic image generation is introduced to augment the available data, thereby improving the robustness and performance of our deep learning model. The experimental evaluation demonstrates that our pipeline accurately approximates the impact echo signals, providing an automated and interpretable approach for structural health monitoring.