Data driven design of ultra high performance concrete prospects and application
Abstract Ultra-high performance concrete (UHPC) is a specialized class of cementitious composites that is increasingly used in various applications, including bridge decks, connections between precast components, piers, columns, overlays, and the repair and strengthening of bridge elements. The mech...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-03-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-94484-2 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Ultra-high performance concrete (UHPC) is a specialized class of cementitious composites that is increasingly used in various applications, including bridge decks, connections between precast components, piers, columns, overlays, and the repair and strengthening of bridge elements. The mechanical and durability properties of UHPC are significantly influenced by factors such as low water-to-binder ratios, the inclusion of supplementary cementitious materials (SCMs), and fiber reinforcement. Machine learning (ML) has been employed to predict the performance of UHPC and optimize its mixture designs by using various raw materials. This study first provides a comprehensive review of ML applications in UHPC, focusing on predicting workability, mechanical, and thermal properties. The use of data crossing, generative AI, physics-guided ML models, and field-applicable software are explored as practical directions for future research. This study also develops ML models to predict the compressive strength of UHPC by using a database containing 1300 data-records. The influence of various input variables is evaluated using SHapley Additive exPlanations (SHAP), revealing that chemical compositions have relatively minor impacts, given the material types used. By excluding insignificant variables, the models enhance both efficiency and accuracy in predicting strength. This advancement facilitates optimized material design and performance prediction while reducing the experimental workload required to inform ML models. Adding more diverse data to the database could further enhance the prediction performance and generalizability of the proposed ML models. |
|---|---|
| ISSN: | 2045-2322 |