Applications of computational intelligence for predictive modeling of properties of blended cement sustainable concrete incorporating various industrial byproducts towards sustainable construction

• Published in: Asian journal of civil engineering. Building and housing 2024-12, Vol.25 (8), p.5939-5954
• Main Authors: Mungle, Niscal P., Mate, Dnyaneshwar M., Mankar, Sham H., Tale, Vithoba T., Vairagade, Vikrant S., Shelare, Sagar D.
• Format: Article
• Language: English
• Subjects: Building Materials; Civil Engineering; Engineering; Sustainable Architecture/Green Buildings
• ISSN: 1563-0854; 2522-011X
• DOI: 10.1007/s42107-024-01155-0

The quest to enhance the strength of concrete, while at the same time reducing the environmental impacts occasioned by its use, has become quite imperative in sustainable construction. Traditional approaches toward supplementary cementitious materials optimization have often fallen short in revealing synergistic interactions that maximize mechanical properties. The current research overcomes these limitations by considering combined effects of different SCMs on concrete strength levels, using advanced artificial intelligence techniques. Current methods often make assumptions with respect to linearity of the models or simple interaction effects that insufficiently represent the multi-level, nonlinear relationships between SCMs and concrete properties. Moreover, integration of microstructural analysis into predictive models is poorly explored. In this paper, a hybrid GBM-CNN methodology is proposed to model complicated interactions within SCM compositions. GBMs are competent in dealing with numerical features, such as SCM proportions, curing time, and temperature, which hold nonlinear relationships in tabular data samples. Meanwhile, CNNs process microstructural images to extract spatial features correlating to mechanical properties. These models will predict the concrete strengths by fusing their outputs using an ensemble method expected to have an R’2 of about 0.85 and an RMSE of about 2 MPa levels. The complexity of the data is managed by using multi-modal data analytics, wherein feature engineering techniques are integrated with Principal Component Analysis, thereby improving the quality of the data while bringing down its dimensionality to retain only the most vital information to explain 95% of data variance. Further, polynomial regression models with regularization—that includes non-linear interaction terms of SCMs, curing conditions, and engineered features—will be built, which highlights the key interaction terms statistically significant with p Value