Scholarlink

<p>Traditional pervious concrete design is based on deterministic optimization techniques that are ill-suited to the challenges posed by material variability, trade-off exploration, and sustainability requirements (e.g. the use of RAP and/or geopolymer binders). In this work, a high-dimensional (540 specimens in 180 mix combinations) experimental database was generated in which 7 mix design and processing parameters were systematically varied (RAP content (0–75%), FA/GGBS mass ratio (1:0, 2:1, 1:1, 1:2, 0:1), W/B (0.30–0.45), target porosity (15–25%), alkaline activator molarity (8–14&#xa0;M), fine aggregate (0–15%)). This data was used to construct a BN model (32-nodes, 67-edges) for the representation of complex non-linear and multi-causal dependencies between composition/processing parameters and material performance objectives (compressive strength, permeability, environmental impact, and cost). Multi-objective Bayesian optimization (Gaussian Process surrogate models) was then used to globally and simultaneously explore the trade-off between these material performance objectives. The optimized design (45% RAP, 1:1 FA:GGBS, 20% porosity) achieved a compressive strength of 24.3 ± 1.8&#xa0;MPa (21% higher than paving requirements), a permeability of 8.7 ± 1.2&#xa0;mm/s (6.7 × higher than the ACI minimum requirement), and a mix-level embodied carbon (cradle-to-gate) reduction of 54.5 ± 3.5% with respect to OPC-based systems (full system-level LCA reduction: 68%, inclusive of maintenance avoidance and end-of-life valorisation credits). Model predictive capability was validated using 80/20 training/validation data splits, with resulting R2 values of 0.92, 0.89, and 0.94 for strength, permeability, and environmental impact, respectively. To the best of the authors’ knowledge, this work represents the first integrated BN–GP–NSGA-II framework for pervious geopolymer concrete incorporating RAP, enabling simultaneous probabilistic multi-objective optimisation across structural, hydraulic, environmental, and economic objectives with full uncertainty quantification. Prior probabilistic and ML-based studies in concrete have addressed either single-objective prediction or deterministic multi-objective optimisation, but have not combined Bayesian Network causal modelling with Gaussian Process surrogates and NSGA-II evolutionary search within a unified reliability-based design framework for this material system.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Probabilistic Multi-Objective Optimization of Pervious Geopolymer Concrete with Reclaimed Asphalt Pavement Using Bayesian Networks

  • 期刊论文

  • 发表日期：

    2026年6月1日

查看全文
  • Nagaraj M. Tattien,
  • M. Abhay,
  • B. Manjunatha,
  • H. T. Avinash,
  • Kalappa Sutar,
  • S. Krishna Rao

摘要

Traditional pervious concrete design is based on deterministic optimization techniques that are ill-suited to the challenges posed by material variability, trade-off exploration, and sustainability requirements (e.g. the use of RAP and/or geopolymer binders). In this work, a high-dimensional (540 specimens in 180 mix combinations) experimental database was generated in which 7 mix design and processing parameters were systematically varied (RAP content (0–75%), FA/GGBS mass ratio (1:0, 2:1, 1:1, 1:2, 0:1), W/B (0.30–0.45), target porosity (15–25%), alkaline activator molarity (8–14 M), fine aggregate (0–15%)). This data was used to construct a BN model (32-nodes, 67-edges) for the representation of complex non-linear and multi-causal dependencies between composition/processing parameters and material performance objectives (compressive strength, permeability, environmental impact, and cost). Multi-objective Bayesian optimization (Gaussian Process surrogate models) was then used to globally and simultaneously explore the trade-off between these material performance objectives. The optimized design (45% RAP, 1:1 FA:GGBS, 20% porosity) achieved a compressive strength of 24.3 ± 1.8 MPa (21% higher than paving requirements), a permeability of 8.7 ± 1.2 mm/s (6.7 × higher than the ACI minimum requirement), and a mix-level embodied carbon (cradle-to-gate) reduction of 54.5 ± 3.5% with respect to OPC-based systems (full system-level LCA reduction: 68%, inclusive of maintenance avoidance and end-of-life valorisation credits). Model predictive capability was validated using 80/20 training/validation data splits, with resulting R2 values of 0.92, 0.89, and 0.94 for strength, permeability, and environmental impact, respectively. To the best of the authors’ knowledge, this work represents the first integrated BN–GP–NSGA-II framework for pervious geopolymer concrete incorporating RAP, enabling simultaneous probabilistic multi-objective optimisation across structural, hydraulic, environmental, and economic objectives with full uncertainty quantification. Prior probabilistic and ML-based studies in concrete have addressed either single-objective prediction or deterministic multi-objective optimisation, but have not combined Bayesian Network causal modelling with Gaussian Process surrogates and NSGA-II evolutionary search within a unified reliability-based design framework for this material system.