Application of Random Forest and Multi-layer Perceptron ANNS in Estimating the Axial Compression Capacity of Concrete-Filled Steel Tubes

A. Arokiaprakash,S. Senthil Selvan
DOI: https://doi.org/10.1007/s40996-022-00893-y
2022-06-24
Iranian Journal of Science and Technology Transactions of Civil Engineering
Abstract:The aim of the research is to develop a deep learning-based artificial neural network model in order to predict the axial compression capacity of concrete-filled steel tubular (CFST) column. Examining the behaviour of CFST columns, especially the axial compression (ACC) is crucial, where core concrete and steel tubular relationship is restricted and complex in CFST. Further, the association of material properties with geometric and axial compression is estimated as highly nonlinear. Henceforth, to ensure safety in operations of structures in engineering, limitations/restrictions via soft-computing had been adopted by the researchers to predict the 'Nu' (i.e. ultimate-bearing capacity). Three thousand one hundred and three datasets have been taken from thorough examination of 173 literature as sources, and the data have been used for training, testing, evaluating and validating the proposed ANN with random forest and linear regression-based hybrid model. To ascertain the best model of ANN, trial and error techniques were utilized with the lowest mean-square error value (MSE) and the highest correlation coefficient value ( R ). The comparative outcomes of the research revealed that MLP regressor with ANN model is more accurate and stable than that of other ANN models. By utilizing the acquired datasets, estimation of ACC upon circular and rectangular steel tubular columns filled with concrete for circular-concrete-filled steel tube (C-CFST), rectangular-concrete-filled steel tube (R-CFST), circular-concrete-filled steel tube eccentrically loaded (C-CFST-BC) and rectangular-concrete-filled steel tube eccentrically loaded (R-CFST-BC) has been carried-out. Through the relative findings, it could be interpreted that the proposed experimental model achieved higher success than existing models, where MLP regression (MLPR) score was found to be best than linear and random forest regression, with R2 scores of 98.9% in C-CFST, 99% in CFST-BC, 97% in R-CFST and 98.9% in R-CFST-BC. Similarly, MLPR was found effective with best RMSE scores for C-CFST as 452, CFST-BC as 87 and R-CFST as 313, whereas linear regression was best for R-CFST-BC with 394. Thus, research concludes that overall, the MLP regression is best and effective in predicting ACC in concrete-filled steel tube columns.
What problem does this paper attempt to address?