Answers are grounded in this paper and cite their sources.
Predicting the compressive strength (CS) of 3D-printed concrete (3DPC) is challenging due to its anisotropic nature, curing processes, and complex material interactions. In this study, a novel hybrid approach combining a Multilayer Perceptron (MLP) neural network with Firefly Algorithm (FA) optimization was used to improve hyperparameter tuning and accurate prediction of the CS of 3DPC. Various databases (299) from previous studies were collected, each containing different features related to material properties and printing parameters. The FA was utilized to fine-tune the MLP’s hyperparameters. The performance of the hybrid MLP-FA model was compared with that of classical MLP, Long Short-Term Memory (LSTM), MLP-LSTM hybrid, and MLP–Genetic Algorithm (MLP-GA) variants using an 80:20 train–test split. Performance indicators showed that MLP-FA outperformed the other models on the test set, achieving an R2 of 0.951, an MSE of 95.168, an RMSE of 9.755, and an MAE of 7.658. The robustness of MLP-FA was confirmed by statistical analysis. SHapley Additive exPlanations (SHAP) analysis and feature importance analysis recognized superplasticizer, fly ash, silica fume, cement, and w/b ratio as the most relevant features. Real-time CS predictions using a Python-based graphical user interface (GUI) have been developed to promote adoption of CS in sustainable construction practices.
No takes yet. Share an insight, caveat, or question.
Haruna et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: