Intelligent compressive strength prediction of sustainable rubberised concrete using an optimised interpretable deep CNN-LSTM model with attention mechanism

作者：Yang Yu, Roshan Jayathilakage, Yiyang Liu, Ailar Hajimohammadi · 发表于：Applied Soft Computing · 年份：2025 · DOI：10.1016/j.asoc.2025.113993 · 被引用次数：30 · 研究领域：Innovative concrete reinforcement materials、Structural Behavior of Reinforced Concrete、Infrastructure Maintenance and Monitoring

The increasing environmental concerns associated with waste rubber disposal, particularly from used tyres, have led to the exploration of rubberised concrete as a sustainable construction material. Rubberised concrete provides benefits like enhanced flexibility and energy absorption; however, its reduced compressive strength remains a challenge for structural applications. This study puts forward an advanced deep learning model to accurately evaluate compressive strength of rubberised concrete by combining a hybrid convolutional neural network (CNN) and long short-term memory (LSTM) network enhanced with attention mechanism, and optimised using the enhanced firefly algorithm (EFA), featuring chaotic initialisation and nonlinear learning factor for improved convergence, for hyperparameter tuning. The proposed model introduces computing novelties: attention-guided CNN-LSTM feature fusion and chaos-enhanced firefly optimisation. Then, it is trained on an extensive dataset incorporating key mix parameters, including water, cement, supplementary cementitious materials, superplasticiser, coarse and fine aggregates, crumb and chipped rubber content, and concrete age, with validation supported by experimental tests in the laboratory. The proposed model achieves superior prediction accuracy, achieving R² values of 0.967 for training and 0.943 for testing, outperforming conventional machine learning methods. Evaluation metrics showcase the superior performance of model, with root mean ...