Point-wise attention digital twin modeling for full-field stress prediction in delivery testing of complex structures

Structural integrity assessment of complex structures, including rocket thrust frames, during delivery testing requires comprehensive full-field stress monitoring to ensure product quality and safety. Traditional testing approaches rely on discrete strain gauge measurements, which provide limited spatial coverage and may fail to detect critical stress concentrations in unmonitored regions, potentially compromising structural reliability. To address these limitations, this paper presents a point-wise attention surrogate modeling framework for predicting full-field stress distributions across complex structures under real-time varying loads during delivery testing processes. The methodology introduces a point-based data segmentation strategy to generate structure-level datasets for training the stress prediction network. A multi-head attention surrogate model is developed that combines physics-informed baseline stress fields from finite element analysis with real-time sensor measurements to predict evolving stress distributions under quasi-static testing conditions. The point-wise field outputs are concatenated and visualized as comprehensive stress distribution maps for structural health assessment. The effectiveness of the proposed framework is validated through comparisons with existing field prediction algorithms based on rocket thrust frame static tests, demonstrating improved accuracy and reliability in full-field stress prediction for delivery testing applications.