Uniaxial tensile ductility behavior of ultrahigh-performance concrete based on the mixture design – Partial dependence approach

Abstract

This study examined how UHPC mixture proportions influence its ductility parameters using random forest models. These parameters were energy absorption capacity (g) and strain at peak stress (εpc) under uniaxial tension. This purpose was achieved using importance analyses of input variables and partial dependence plots (PDPs). The numerical models were developed using the R statistical language and 34 input variables, 600 observation databases, including 50 experimental results. Direct tensile tests were conducted on 46 dogbone specimens to verify the prediction capacity of both the model and the results obtained from the PDP analysis. The tests were carried out in accordance with the JSCE recommendations for eighteen UHPC series resulting from the combination of two cementitious matrices with compressive strengths of 124 and 156 MPa and three-fiber monofiber reinforced systems at three volumetric dosage levels (i.e., 1%,2%, and 3%). The findings indicated that it is more likely to reach exceptional values of the considered responses (εpc ≥0.3% and g ≥50 kJ/m3) when using deformed high-strength steel macro fibers than with another type. Additionally, g rose with the compressive strength value. Nevertheless, the εpc value decreased with the increase of the cementitious matrix compressive strength up to the value of 190 MPa, after which the trend is reversed, and εpc presents exponential growth due to the improvement in the adherence between the matrix and the fiber caused by the enhancement of the matrix microstructure.