Advancing Eco-Friendly Construction: Attention-Guided Residual LSTM for High-Accuracy Classification of Sustainable Cementitious Materials

Abstract

The construction sector faces urgent demands to reduce its carbon footprint, driven by the environmental toll of conventional cement production. This study explores the integration of deep learning (DL) to advance sustainable cementitious materials, focusing on the classification and optimization of Ordinary Portland Cement (OPC) and eco-friendly alternatives such as geopolymers. By evaluating the performance of convolutional and recurrent neural networks, including a novel Attention-Guided Residual LSTM (AGResLSTM), the research demonstrates AI’s capacity to streamline material selection while balancing mechanical performance, cost, and environmental impact. Results highlight geopolymersBlast Furnace Slag Geopolymer (BFSGP) and Fly Ash Geopolymer (FAGP) as high-potential substitutes, offering CO₂ reductions of 80–90% compared to OPC without compromising structural integrity. The proposed AGResLSTM model outperforms traditional DL architectures, showcasing superior accuracy in material classification and robustness in handling multi-dimensional data. This work underscores the transformative role of AI in accelerating the adoption of sustainable construction materials, providing a scalable framework to align industry practices with global decarbonization goals. By bridging material science and artificial intelligence, the study advances actionable pathways for reducing the built environment’s ecological impact while maintaining economic and functional viability.