Machine learning technique for the prediction of blended concrete compressive strength
Predicting concrete strength is complex due to the high non-linearity involved in strength development, especially when using supplementary cementitious materials (SCMs) such as fly ash, silica fume, and GGBS. In this paper, an artificial neural network has been used to predict the compressive stren...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Published: |
Korean Society of Civil Engineers
2024
|
| Online Access: | http://psasir.upm.edu.my/id/eprint/105757/ |
| _version_ | 1848864599956258816 |
|---|---|
| author | Jubori, Dawood S. A. Abu B., Nabilah Safiee, Nor A. Alias, Aidi H. Nasir, Noor A. M. |
| author_facet | Jubori, Dawood S. A. Abu B., Nabilah Safiee, Nor A. Alias, Aidi H. Nasir, Noor A. M. |
| author_sort | Jubori, Dawood S. A. |
| building | UPM Institutional Repository |
| collection | Online Access |
| description | Predicting concrete strength is complex due to the high non-linearity involved in strength development, especially when using supplementary cementitious materials (SCMs) such as fly ash, silica fume, and GGBS. In this paper, an artificial neural network has been used to predict the compressive strength of concrete for four cases, namely concrete without cement replacement, and binary, ternary, and quaternary cement concretes corresponding to one, two and three different SCMs in the mix. To predict the strength accurately, a total of 1013 data were collected from 37 literature and trained using two training algorithms namely Levenberg-Marquardt (LM) and Bayesian Regularization (BR). The best predictions were achieved using one hidden layer with 14 and 15 neurons for LM and BR algorithms respectively. A high accuracy has been achieved with a correlation factor of 0.97 and 0.966 using the BR and LM algorithms respectively, with a20-index of 83%. Generally, the BR algorithm gives a better overall performance, while underestimating the compressive strength compared to LM. Sensitivity analysis has also been investigated using linear and quadratic regressions. The findings showed that the highest contributors to concrete strength were cement and water, while the lowest contributor was coarse aggregate. |
| first_indexed | 2025-11-15T13:51:23Z |
| format | Article |
| id | upm-105757 |
| institution | Universiti Putra Malaysia |
| institution_category | Local University |
| last_indexed | 2025-11-15T13:51:23Z |
| publishDate | 2024 |
| publisher | Korean Society of Civil Engineers |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | upm-1057572025-06-03T02:27:29Z http://psasir.upm.edu.my/id/eprint/105757/ Machine learning technique for the prediction of blended concrete compressive strength Jubori, Dawood S. A. Abu B., Nabilah Safiee, Nor A. Alias, Aidi H. Nasir, Noor A. M. Predicting concrete strength is complex due to the high non-linearity involved in strength development, especially when using supplementary cementitious materials (SCMs) such as fly ash, silica fume, and GGBS. In this paper, an artificial neural network has been used to predict the compressive strength of concrete for four cases, namely concrete without cement replacement, and binary, ternary, and quaternary cement concretes corresponding to one, two and three different SCMs in the mix. To predict the strength accurately, a total of 1013 data were collected from 37 literature and trained using two training algorithms namely Levenberg-Marquardt (LM) and Bayesian Regularization (BR). The best predictions were achieved using one hidden layer with 14 and 15 neurons for LM and BR algorithms respectively. A high accuracy has been achieved with a correlation factor of 0.97 and 0.966 using the BR and LM algorithms respectively, with a20-index of 83%. Generally, the BR algorithm gives a better overall performance, while underestimating the compressive strength compared to LM. Sensitivity analysis has also been investigated using linear and quadratic regressions. The findings showed that the highest contributors to concrete strength were cement and water, while the lowest contributor was coarse aggregate. Korean Society of Civil Engineers 2024-01 Article PeerReviewed Jubori, Dawood S. A. and Abu B., Nabilah and Safiee, Nor A. and Alias, Aidi H. and Nasir, Noor A. M. (2024) Machine learning technique for the prediction of blended concrete compressive strength. KSCE Journal of Civil Engineering, 28 (2). pp. 817-835. ISSN 1226-7988; eISSN: 1976-3808 https://linkinghub.elsevier.com/retrieve/pii/S1226798824004306 10.1007/s12205-024-0854-5 |
| spellingShingle | Jubori, Dawood S. A. Abu B., Nabilah Safiee, Nor A. Alias, Aidi H. Nasir, Noor A. M. Machine learning technique for the prediction of blended concrete compressive strength |
| title | Machine learning technique for the prediction of blended concrete compressive strength |
| title_full | Machine learning technique for the prediction of blended concrete compressive strength |
| title_fullStr | Machine learning technique for the prediction of blended concrete compressive strength |
| title_full_unstemmed | Machine learning technique for the prediction of blended concrete compressive strength |
| title_short | Machine learning technique for the prediction of blended concrete compressive strength |
| title_sort | machine learning technique for the prediction of blended concrete compressive strength |
| url | http://psasir.upm.edu.my/id/eprint/105757/ http://psasir.upm.edu.my/id/eprint/105757/ http://psasir.upm.edu.my/id/eprint/105757/ |