Deep learning enhanced framework for multi-objective optimization of cement-slag concrete for the balancing performance, economics, and sustainability

Abstract

This research presents an innovative computational approach that merges artificial intelligence with multi-objective optimization techniques to enhance cement slag concrete design. The proposed framework integrates deep neural networks (DNN), gradient boosting machines (GBM), and extreme learning machines (ELM) with particle swarm optimization (PSO) and multi-objective genetic algorithms (MOGA) to concurrently optimize mechanical properties, cost-effectiveness, and environmental impact. The methodology involved comprehensive data pre-processing, model training, and validation using laboratory tests. Among the models, DNN exhibited the best performance in predicting the uniaxial compressive strength (UCS), achieving an R² of 0.98, and MSE of 0.009, surpassing both the GBM and ELM models. The application of PSO-optimized hyperparameters considerably improved the model accuracy, whereas MOGA identified the optimal mix designs through Pareto front analysis. Grey Relational Analysis determined an ideal cement-to-slag ratio of 85:15, yielding a UCS of 59.8 MPa and the highest grey relational grade (γi = 0.982). The framework achieved a 15% enhancement in the strength-to-cost ratio compared to traditional methods while maintaining environmental advantages through decreased cement usage. This study shows the potential of integrated AI-driven approaches in developing sustainable building materials, offering a solid foundation for future advancements in concrete mix design optimization that balances performance, cost, and environmental factors.