Digital twin-based non-destructive testing method for ultimate load-carrying capacity prediction

Abstract

Load-carrying experiments are essential for validating the mechanical performance of structural designs. Unpredictable early structure collapses during experiments would lead to severe safety accidents, and thus the real-time prediction of the ultimate load-carrying capacity (ULC) becomes highly necessary. However, due to multi-source experiment deviations, ULC prediction methods based on finite element (FE) analysis struggle to meet the high-accuracy and real-time requirements. To address this issue, this paper proposes a digital twin-based non-destructive testing (DT-NDT) method for ULC prediction. The method employs a deep neural network (DNN) to construct an experiment digital twin (DT) model in the offline phase. In the online phase, the experiment DT model utilizes the current measured strains to predict the ULC at future time steps in real-time, thus achieving non-destructive testing (NDT). First, a series of virtual experiments considering multi-source experiment deviations are carried out by FE analysis to obtain the virtual experiment dataset. The dataset includes strains and collapse loads corresponding to different experiment deviations and is augmented using the data augmentation approach based on structural symmetry. Subsequently, DNN is trained by the virtual experiment dataset to establish the mapping of strain values to collapse load. Finally, the strains are measured in real-time using strain gauges and the strain values are inputted into the DNN, i.e., the experiment DT model, to predict the collapse load. In this paper, the principle and effectiveness of the proposed method are illustrated through an analytical example and two experiment examples. In the experiment examples, the average relative errors of the DT-NDT method are 0.9% for an open-hole plate and -1.45% for a grid-stiffened cylindrical shell. The results illustrate the high prediction accuracy and effectiveness of the DT-NDT method, demonstrating its immense potential in NDT.