Reliability assessment of steel‐framed industrial buildings

Abstract The classical reliability problem seeking an estimation of the probability of failure is usually undefined due to: i) its scale (size of the design problems), and ii) deficient characterization of the randomness (lack of statistical characterization of the basic random variables). A common challenge in current approaches for the assessment of structural reliability adopted in EN 1990 is the lack of data examples that could support additional research, simulation, and experiments. As the failure probability in Eurocode is about 10 ‐4 , millions of simulations are required, which limits the accuracy of commonly used Monte Carlo simulations (MCS). Nowadays, there are different intelligent data management approaches to augment existing data points by creating synthetic examples that can then be further used in specific behavior prediction. One such intelligent approach is the recently developed Generative Adversarial Networks (GAN), which has found its use mainly in image augmentation, but also in other applications where finding new data examples is challenging, e.g., credit card fraud detection, defect detection, etc. In this paper, a reliability assessment of steel‐framed industrial buildings is presented. For that purpose, a set of portal frames is first selected as a case study and designed using the actual design codes, whereby the critical structural members and failure modes are identified. Subsequently, based on these representative case studies, the database is expanded using GAN, and the current state of reliability for the selected type of buildings is assessed, aiming to provide a statement of the current safety of industrial sheds using the Eurocodes.