Artificial Neural Network-Based Methodology for Optimization of Low-Cost Green UHPFRC Under Ductility Requirements

Several constructions in earthquake-prone areas in developing countries do not meet current seismic codes, mainly because of the rampant informal construction. These circumstances require effective seismic retrofitting interventions through solutions of an acceptable cost that allow the most extensive application possible. This research focuses on developing a low-cost, low-carbon-footprint material with the required ductility parameters for seismic retrofitting applications. First, a plain UHPC is optimized under compressive strength, cost, and carbon footprint criteria. After that, the second stage of this study determines the binary combination of fibers, among those available in the Colombian market, that permit reaching the necessary ductility parameters for the desired application at a lower cost. The ductility parameters considered are the energy capacity absorption (g) and the strain capacity at maximum tensile strength (ε_pc) measured in the direct tensile test. Various statistical and computational tools such as Artificial Neural Networks, Design of Experiments, and Multi-Objective Optimization were utilized to lesser the experimental campaign. The mathematically optimized dosage was experimentally evaluated. Finally, the optimal fiber volume fraction for the necessary UHPFRC ductility parameters for seismic strengthening applications (g ≥ 50 kJ/m³ and ε_pc ≥ 0.3%) was selected at only 1.7%. This optimal fiber combination was composed of 0.34% of smooth high-strength steel (l_f/d_f = 65) fibers, and 1.36% of normal strength hooked end steel fibers (l_f/d_f = 80). It is relevant to highlight that this optimized UHPFRC outperforms the ductility parameters obtained by other authors with successful applications in the seismic strengthening field.