Uncertainty quantification of creep in concrete by Taylor expansion

Abstract

If deterministic creep prediction models are compared with actual measurement data, often significant differences can be observed. These inconsistencies are associated with different causes, i.e. model uncertainty, uncertain input parameters, measurement errors and wrongfully applying creep prediction models outside their limitations. First, the physical mechanism causing creep of concrete is not yet fully understood. Therefore, it is very likely that certain influences on creep of concrete are not considered in these prediction models, resulting in systematic model errors. The model errors can be quantified by comparing prediction results with experimental data. Secondly, the stochastic character of the input parameters form an additional source of uncertainty which can be quantified by the variance of the model response. The coefficient of variation in function of time-duration, i.e. the time since the application of the load, is a useful measure to quantify the level of uncertainty. In the literature, statistical analysis by means of numerical simulations are often used for this matter. However, even for specialized sampling techniques, a large amount of samples is necessary to cover the relevant ranges of various input parameters. The aim of the present study is to provide an approximate uncertainty quantification of the creep prediction models given uncertain input parameters. This approximation is based on a Taylor series approach. This approach has the advantage that is does not require numerical simulations nor does it require the knowledge of the probability density function of the input parameters. This method is evaluated and compared with the statistical analysis for several creep prediction models available in literature and design codes.