A Bayesian network framework for statistical characterisation of model parameters from accelerated tests: application to chloride ingress into concrete. Issue 5 (4th May 2018)

• Record Type: Journal Article
• Authors: Tran, Thanh-Binh; Bastidas-Arteaga, Emilio; Schoefs, Franck; Bonnet, Stéphanie
• Is Part Of: Structure and infrastructure engineering. Volume 14:Issue 5(2018)
• Page Start: 580
• Page End: 593
• Publication Date: 2018-05-04
• Subjects: Accelerated test -- reinforced concrete -- corrosion -- Bayesian network -- statistical characterisation -- random variable; Structural analysis (Engineering) -- Periodicals; Structural engineering -- Periodicals; Buildings -- Performance -- Periodicals; 620.005
• DOI: 10.1080/15732479.2017.1377737
• Languages: English
• ISSNs: 1573-2479

Abstract

Abstract: This paper addresses the topic of long-term characterisation and probabilistic modelling of chloride ingress into reinforced concrete (RC) structures. Since the corrosion initiation stage may cover various decades, normal tests which simulate chloride penetration into concrete in laboratory conditions as the same as natural conditions, will require significant experimental times. Hence, long-term lifetime assessment of RC structures under chloride attack remains still a challenge. In practice, this problem is solved through the use of accelerated tests which speed up the chloride ingress rate and provide valuable mid- and long-term information on the chloride penetration process. Nevertheless, this information cannot be directly used for parameter statistical characterisation if the equivalent times required in natural conditions to reach the same chloride concentrations in the accelerated tests are unknown. Consequently, this study proposes a novel iterative approach based on Bayesian network updating to estimate chloride ingress model parameters from the data obtained under accelerated laboratory conditions. The Bayesian network structure and iterative approach are first tested with numerical evidences. Thereafter, the complete proposed methodology is verified with results from real experimental measurements. The results indicate that combining data from normal and accelerated tests significantly reduces the statistical characterisation error of model parameters.