Towards sustainable construction: Machine learning based predictive models for strength and durability characteristics of blended cement concrete

• Published in: Materials today communications 2023-12, Vol.37, p.107428, Article 107428
• Main Authors: Khan, Majid, Javed, Muhammad Faisal
• Format: Article
• Language: English
• Subjects: Blended cement concrete; Machine learning; Predictive models, SHAP analysis; Supplementary cementitious materials
• ISSN: 2352-4928; 2352-4928
• DOI: 10.1016/j.mtcomm.2023.107428

Supplementary cementitious materials (SCMs) are widely utilized in concrete mixtures, either substituting a part of the cement content or replacing a portion of clinker in cement. This commonly practiced approach is highly beneficial for the construction industry, as it generally results in concrete with lower production costs, reduced environmental impact, improved long-term durability, and increased long-term strength. The concrete industry and researchers are actively seeking methods to predict the performance of blended cement concrete (BCC) mixes, to minimize the expenses and time involved in experimental testing of numerous alternatives. Leveraging the potential of machine learning algorithms, which have proven effective in handling extensive datasets and establishing accurate relationships for data. This study utilized two individual models and one ensemble model to predict the characteristics of blended cement concrete. To develop the database for model development, 1287 data points for compressive strength, 361 for carbonation, and 323 for chloride resistivity were collected from experimental studies. Various error metrics were used to assess the efficacy of the developed models. The decision tree (DT) model accurately estimated the compressive strength with correlation coefficient (R) value of 0.99 for both training and validation sets. For both durability characteristics of BCC, the AdaBoost regressor (AR) model demonstrated predictions with an R-value exceeding 0.98. The mean absolute error (MAE) and root mean square error (RMSE) of the AR model remained below 0.5 for carbonation and below 400 for chloride penetration. The SHapley Additive exPlantion (SHAP) technique provided that among the SCMs, silica fume and calcined clay enhance the compressive strength. Moreover, most of the SCMs have an unfavorable influence on the carbonation resistivity of BCC. The substitution of Portland cement with silica fume, calcined clay, lime powder, and ground granulated blast furnace slag leads to a reduction in the chloride penetrability, consequently enhancing the chloride resistivity of the concrete. [Display omitted]