^{1,a}, José Barros Cabezas

^{2}and Juan Chiachío Ruano

^{1,b}

^{1}Structural Mechanics and Hydraulics Engineering Department, University of Granada, Spain.

^{a}mchiachio@ugr.es

^{b}jchiachio@ugr.es

^{2}Civil Engineering Faculty, Catholic University of Santiago de Guayaquil, Ecuador.

Engineering practice commonly requires the calibration of complex numerical models based on experimental data, which is typically carried-out as a trial and error process whose success is highly influenced by human errors. The Bayesian procedure is a robust methodology to solve this problem which also allows quantification of the uncertainties. However, this procedure requires the knowledge of a likelihood function, which sometimes is difficult to evaluate or directly impossible to obtain. For such cases, the Approximate Bayesian Computation (ABC) method is an efficient alternative to address the calibration of a complex numerical probabilistic model based on data. This paper presents the applicability of the ABC method using an efficient algorithm called ABCSubSim for the calibration of a complex non-lineal mechanical model. A set of uncertain model parameters from a reinforced concrete column subjected to lateral cyclic loads, are indirectly inferred with quantified uncertainty with a low computational cost. These parameters are difficult to observe experimentally and are crucial to asses the structural vulnerability and safety under seismic loads. Results show that the proposed methodology reduces the uncertainty about the mechanical parameters and makes them learn from the data, hence making this algorithm useful for uncertainty management and safety assessment. Influence of axial load on reinforced concrete beamcolumn models is also discussed. Finally, the paper discusses further improvements required for the ABC-SubSim based on sensitivity analysis and numerical trials carried-out over the course of this research.