Researchers have developed a novel AI-driven framework that designs eco-friendly concrete mixes by balancing strength, cost, and sustainability in real time.
Study: Optimizing sustainable blended concrete mixes using deep learning and multi-objective optimization. Image Credit: Andrey Suslov/Shutterstock.com
A new study published in Scientific Reports introduces a data-driven framework that combines deep neural networks (DNNs) with multi-objective particle swarm optimization (MOPSO) to design environmentally conscious concrete mixes. The optimized formulations achieved compressive strengths above 50 MPa while reducing cement content by up to 25 %—translating to a total cost savings of approximately 15 % compared to conventional designs.
Background
Traditional concrete mix design often relies on trial-and-error methods, which can be time-intensive and inefficient, especially when balancing strength, cost, and sustainability. With recent advances in computational modeling, more systematic and data-informed approaches have become feasible.
Machine learning, and particularly deep learning, offers the ability to model complex, nonlinear relationships between input variables—such as ingredient proportions and curing time—and performance metrics like compressive strength and durability. At the same time, optimization algorithms are being used to address sustainability concerns by finding optimal trade-offs across competing objectives.
This study presents an integrated approach: a deep learning model predicts concrete compressive strength, while a multi-objective optimization routine aims to maximize strength, minimize cement content, and reduce overall cost. The system uses DNNs to learn from historical data and applies MOPSO to navigate the design space efficiently.
Methods
The proposed methodology involves several stages: data preprocessing, model development, hyperparameter tuning, evaluation and validation, multi-objective optimization, feature importance analysis, and the creation of a graphical user interface (GUI) for real-time use.
The DNN architecture consists of an input layer, multiple hidden layers with ReLU activation functions, and a single output node predicting compressive strength. Bayesian optimization was used to fine-tune the model’s hyperparameters with the goal of minimizing root mean squared error (RMSE).
The model’s accuracy was tested on unseen data using a suite of metrics: RMSE, mean absolute error (MAE), mean absolute percentage error (MAPE), and R2. For optimization, the DNN’s predictions were paired with a cost function based on the weighted sum of ingredient masses, and a third criterion explicitly minimized cement content. A feature-importance analysis using permutation methods identified key performance drivers. Finally, the team developed a MATLAB-based GUI to make the system more accessible for practical use.
Results and Discussion
The trained DNN effectively modeled nonlinear relationships among primary mix components such as cement, fly ash, blast furnace slag, superplasticizer, and water. Strong cross-validation results, highlighted by an average R2 of 0.936 and an RMSE of 5.71 MPa, confirmed the model’s robustness.
Using MOPSO, the researchers explored a range of viable mix designs within realistic constraints (e.g., total volume, water-cement ratio, material limits). The optimized results offered practical trade-offs between cost, strength, and sustainability. Notably, some designs maintained compressive strengths of 50 MPa while reducing cement usage by 25 % and cutting costs by 15 %.
Feature-importance analysis revealed that cement content and curing age were the most influential factors on compressive strength, with other inputs like fly ash, superplasticizer, and slag also contributing significantly.
Overall, the framework serves as a reliable tool for developing concrete mixes that meet performance standards while reducing environmental impact and costs. The inclusion of a user-friendly GUI bridges the gap between theory and on-site application, enabling quicker, data-backed decision-making in construction settings.
Conclusion and Future Directions
This integrated framework demonstrates how DNNs and MOPSO can work together to produce sustainable, high-performance concrete mixes by balancing cost, strength, and environmental considerations.
The authors suggest future work could involve expanding the dataset to account for variations in raw material sources, environmental conditions, and additional performance metrics like workability and long-term durability. They also propose incorporating advanced activation functions, sensitivity analysis, and alternative metaheuristic algorithms to enhance accuracy and optimization scope.
Given that standard MOPSO may struggle with complex or high-dimensional problems, future iterations may include decomposition-based strategies to improve diversity and robustness in the solution set.
In sum, this approach offers a practical, adaptable roadmap for integrating AI-driven optimization into sustainable construction practices.
Journal Reference
Tipu, R. K., Rathi, P., Pandya, K. S., & Panchal, V. R. (2025). Optimizing sustainable blended concrete mixes using deep learning and multi-objective optimization. Scientific Reports, 15(1). DOI: 10.1038/s41598-025-00943-1, https://www.nature.com/articles/s41598-025-00943-1
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.