A hybrid machine learning approach in prediction and uncertainty quantification of ultimate compressive strength of RCFST columns

This article introduces a machine learning-based approach to estimate the ultimate compressive strength of rectangular concrete-filled steel tube (RCFST) columns, and to perform an uncertainty quantification of such values. A hybrid model, namely PANN, is built based on artificial neural networks (ANNs) optimized by particle swarm optimization (PSO). The proposed model is based on an experimental data set of 622 cases. The PANN estimation accuracy is benchmarked against four design codes, including EC4, ACI, AISC, and AS. Benchmarked results show that hybrid models are the perfect alternative models to predict the compressive strength of the RCFST column. The RCFST compressive strength depends mainly on geometry, material, and loading configurations which have stochastic features in nature. This study proposes the Monte Carlo technique to investigate stochastic compressive strength behaviors of RCFST columns considering complex interactions of variables with an exposed to inevitable source-uncertainties. Night cases of individual and compound randomness are examined. Convergence studies and sensitivity analyses are performed before uncertainty quantifications on each random parameter are conducted. Systematic consideration of uncertainties in compressive strength may allow for more trust in model results and, if columns are analyzed, designed, it improves the construction's trustworthiness.