Optimization method, choice of form and uncertainty quantification of Model B4 using laboratory and multi-decade bridge databases

The preceding article describes a new multi-decade creep and shrinkage prediction model, labeled B4, which extends and improves the previous RILEM recommendation B3, and a separate article presents a new large worldwide database of laboratory data on creep, drying shrinkage and autogenous shrinkage. This article presents the general optimization concepts used to verify and calibrate Model B4. The main objective is multi-decade, even 100-year, prediction, which is what is needed for sustainable design of long-span bridges, tall buildings and other large concrete structures. Since the existing worldwide database is insufficient by far for purely experimental verification and calibration of multi-decade creep, the importance of choosing a model form that is theoretically and physically justified is emphasized. So is the choice of a model form that is able to fit closely individual broad-range creep and shrinkage curves for one and the same concrete, which are free of the huge obfuscating scatter due to differences in concrete composition. The development and calibration of the formulae for predicting the parameters of the creep and shrinkage equations from the composition and strength of concrete is described. An effective method for statistical optimization of the fits of a new database comprising thousands of laboratory test curves is presented. Various types of bias in the database are counteracted by data weighting. To predict and calibrate multi-decade creep, a method to combine the laboratory database with a database of excessive multi-decade deflections of large span bridges is outlined. This leads to a significant increase of the slope of the terminal trend of predicted creep in a logarithmic plot. Finally, the statistical approaches for using the hybrid laboratory-bridge database for multi-objective fit optimization and for Bayesian updating are explained and discussed.