Concrete 3D printing proposes an off-site industrial process allowing to deposit material only where required. However, most mixture design methods struggle to perform, which is why a majority of 3D printing materials display high clinker contents. This study proposes a reproducible methodology for tailor-made 3D printing materials. Applied to a low-clinker quaternary blend, an iterative optimization process leads to a significant reduction of labor in material tuning. It involves life cycle assessment and artificial neural networks as objective functions in the Pareto selection of best-performing solutions. Following the constitution of an 18-mixture database with 6 independent variables and 5 objective functions, printable mortars of different strength classes are designed within 2 to 4 active learning runs. Consequently, this optimum-driven technique allows to rapidly converge toward low-carbon solutions for 3D printing, using local materials and custom characterization procedures.

中文翻译：

---

最大限度地减少 3D 打印混凝土的碳足迹：通过多目标优化利用参数化 LCA 和神经网络


混凝土 3D 打印提出了一种场外工业流程，允许仅在需要时沉积材料。然而，大多数混合物设计方法都难以执行，这就是为什么大多数 3D 打印材料都显示出高熟料含量的原因。本研究提出了一种用于定制 3D 打印材料的可重复方法。应用于低熟料四元混合物，迭代优化过程可显著减少材料调整的劳动力。它涉及生命周期评估和人工神经网络，作为帕累托选择最佳解决方案的目标函数。在构建一个具有 6 个自变量和 5 个目标函数的 18 种混合物数据库之后，在 2 到 4 次主动学习运行中设计出不同强度等级的可打印砂浆。因此，这种最佳驱动的技术允许使用本地材料和定制表征程序快速向 3D 打印的低碳解决方案收敛。