Scientific Reports published an open-access study on 17 September 2026 that analysed 729 recycled-aggregate concrete mixes containing ground granulated blast furnace slag, waste crumb rubber and recycled coarse aggregate. The researchers combined an artificial neural network with response surface methodology to predict the non-steady-state chloride migration coefficient—a laboratory indicator of resistance to chloride ingress. The work is relevant to GGBFS durability, but it is a model built from existing experimental datasets rather than a universal mix-design rule.

1. GGBFS and curing age were strong durability signals

The study identified curing age as the most influential variable in its dataset, with GGBFS the second most influential. Higher values for both were generally associated with lower chloride migration. The authors link the GGBFS effect to a denser cementitious matrix and reduced permeability. That direction is consistent with the material's established role as a supplementary cementitious material, but actual performance still depends on slag chemistry, fineness, reactivity, cement compatibility, water-to-binder ratio, curing and exposure conditions.

2. Predictive accuracy can improve screening, not replace testing

Within the compiled database, 98 per cent of the neural-network predictions were within plus or minus 6 per cent of measured chloride-migration results, compared with 86 per cent for the response-surface model. This makes data tools useful for narrowing candidate mix ranges and identifying sensitive variables. It does not eliminate laboratory trials, durability testing or specification review. Models inherit the boundaries of their source data, and a result for one GGBFS source, cement system or curing regime cannot automatically be transferred to another.

3. The reported optimum is a study result, not a purchase specification

The durability-focused optimization selected 40 per cent GGBFS by binder mass, alongside specified proportions of waste crumb rubber and recycled coarse aggregate, a 0.35 water-to-binder ratio and 91 days of curing. The authors explicitly caution that the optimization did not simultaneously cover compressive strength, workability, cost, embodied carbon, shrinkage or other durability indicators. Procurement and engineering teams should therefore treat the figure as a research signal. Commercial qualification still needs an agreed specification, representative samples, test methods, acceptance limits and repeatability across delivered lots.

Takeaway: Data-driven durability models can help engineers understand how GGBFS interacts with curing and recycled materials, but dependable use starts with qualified material and ends with project-specific evidence. For international supply, the practical chain remains clear: distinguish GBFS feedstock from finished GGBFS, define the required properties, verify representative material and maintain lot-to-lot control. Source: Scientific Reports, published 17 September 2026.

《Scientific Reports》于 2026 年 9 月 17 日发表了一项开放获取研究，分析了 729 组含矿渣粉、废橡胶颗粒和再生粗骨料的再生骨料混凝土配合比。研究人员结合人工神经网络与响应面法，预测非稳态氯离子迁移系数——这是实验室评估混凝土抗氯离子侵入能力的一项指标。该研究与 GGBFS 耐久性直接相关，但它基于既有实验数据建模，并不是可普遍套用的配合比规则。

1. GGBFS 与养护龄期是重要的耐久性信号

在该研究的数据集中，养护龄期是影响最大的变量，GGBFS 位居第二；两者增加通常都与更低的氯离子迁移系数相关。作者认为，GGBFS 有助于形成更致密的胶凝体系并降低渗透性。这一方向与补充性胶凝材料的既有作用相符，但实际表现仍取决于矿渣化学成分、细度、活性、水泥适配性、水胶比、养护条件和暴露环境。

2. 预测精度可用于筛选，但不能替代检测

在汇总数据库范围内，人工神经网络有 98% 的预测结果与实测氯离子迁移数据之间的误差控制在正负 6% 以内，而响应面模型为 86%。这说明数据工具可以帮助缩小候选配合比范围，并识别敏感变量；但它不能取代实验室试配、耐久性检测或规格审核。模型受到源数据边界约束，某一种 GGBFS 来源、水泥体系或养护制度下的结果，不能自动移植到另一套体系。

3. 研究中的最优值不是采购规格

该耐久性导向优化选择了占胶凝材料质量 40% 的 GGBFS，并配合研究设定的废橡胶颗粒、再生粗骨料比例、0.35 水胶比和 91 天养护龄期。作者明确提醒，这项优化没有同时覆盖抗压强度、和易性、成本、隐含碳、收缩及其他耐久性指标。因此，采购和工程团队应把这一数值视为研究信号。商业认证仍需明确约定规格、代表性样品、检测方法、接受限值，并验证交付批次之间的稳定性。

结论：数据驱动的耐久性模型可以帮助工程师理解 GGBFS 与养护条件、再生材料之间的相互作用，但可靠应用始于合格材料，最终仍要落到项目证据。对国际供应而言，实际链条依然清晰：区分作为原料的 GBFS与成品 GGBFS，明确所需性能，检测代表性材料，并保持批次间质量控制。来源：《Scientific Reports》，发布于 2026 年 9 月 17 日。