Performance prognosis of FRCM-to-concrete bond strength using ANFIS-based fuzzy algorithm
Kumar, Aman ; Arora, Harish Chandra ; Kumar, Krishna LU
and Garg, Harish
(2023)
In Expert Systems with Applications
216.
- Abstract
Nowadays, strengthening of reinforced concrete structures with a new class of sustainable materials is the possible solution to retrofit the aged deteriorated structures. It is difficult to predict the capacity of the strengthened (FRP/FRCM) reinforced concrete elements without considering the bond strength. Therefore, the concrete substrate to Fibre-Reinforced Cementitious Matrix (FRCM) bond is a crucial parameter in the strengthening procedures. As it is known, bond strength is dependent on various parameters, which increases the complexity of the FRCM-to-concrete bond. Analytical models cannot provide a high degree of accuracy, as their predictions are only valid for specific datasets. Machine learning algorithms are the best-suited... (More)
Nowadays, strengthening of reinforced concrete structures with a new class of sustainable materials is the possible solution to retrofit the aged deteriorated structures. It is difficult to predict the capacity of the strengthened (FRP/FRCM) reinforced concrete elements without considering the bond strength. Therefore, the concrete substrate to Fibre-Reinforced Cementitious Matrix (FRCM) bond is a crucial parameter in the strengthening procedures. As it is known, bond strength is dependent on various parameters, which increases the complexity of the FRCM-to-concrete bond. Analytical models cannot provide a high degree of accuracy, as their predictions are only valid for specific datasets. Machine learning algorithms are the best-suited solution to deal with bond strength like complex problems. In this study, curve-fitting, Gaussian Process Regression (GPR) and Adaptive Neuro-Fuzzy Inference System (ANFIS) models have been applied to 336 aggregated datasets. Nine performance matrices have been opted to compare the performance of the developed models. Feature importance analysis has also been used to check the rationality of the model. The parametric analysis has also been done using the 3D-surface plot of the ANFIS model. The R-values of ANFIS, GPR, and curve-fitting models are 0.9895, 0.9882, and 0.9145, respectively. The mean absolute error, root mean square error and Nash-Sutcliffe index of the ANFIS model are 0.9168 kN, 1.4326 kN, and 0.9791, respectively. The mean absolute percentage error of the ANFIS model is 11.19%, which is 8.72% and 76.78% lower than GPR and curve-fitting model, respectively. The error range of the curve-fitting, GPR and ANFIS models are −17.06 kN to 18.04 kN, −4.39 kN to 6.07 kN, and −4.23 kN to 5.19 kN, respectively. Overfitting analysis of the proposed models has been done, and the predicted results show that the curve-fitting model and GPR models are inferior and the ANFIS model is superior based on the selected performance matrices. The overfitting value of ANFIS model is 67.89% and 8.31% lower than curve-fitting and GPR model, respectively. The sensitivity analysis found that the number of layers, the width of the concrete block, and the compressive strength of the concrete had the highest effect on the FRCM-to-concrete bond strength. The findings of the study have the potential to decrease costs and save time by employing an accurate prediction approach instead of expensive and time-consuming testing. The developed model can be easily used by industry experts and FRCM applicators to estimate the bonding strength of FRCM-to-concrete substrate for sustainable designs.
(Less)
- author
- Kumar, Aman
; Arora, Harish Chandra
; Kumar, Krishna
LU
and Garg, Harish
- publishing date
- 2023-04-15
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- ANFIS, Artificial intelligence, Bond strength, Curve-fitting, FRCM, GPR, Machine learning
- in
- Expert Systems with Applications
- volume
- 216
- article number
- 119497
- publisher
- Elsevier
- external identifiers
-
- scopus:85145966560
- ISSN
- 0957-4174
- DOI
- 10.1016/j.eswa.2022.119497
- language
- English
- LU publication?
- no
- additional info
- Publisher Copyright: © 2022 Elsevier Ltd
- id
- 75e91552-e6ff-4ca0-9f1a-42824cb56589
- date added to LUP
- 2024-04-16 09:17:30
- date last changed
- 2025-04-04 15:27:03
@article{75e91552-e6ff-4ca0-9f1a-42824cb56589,
abstract = {{<p>Nowadays, strengthening of reinforced concrete structures with a new class of sustainable materials is the possible solution to retrofit the aged deteriorated structures. It is difficult to predict the capacity of the strengthened (FRP/FRCM) reinforced concrete elements without considering the bond strength. Therefore, the concrete substrate to Fibre-Reinforced Cementitious Matrix (FRCM) bond is a crucial parameter in the strengthening procedures. As it is known, bond strength is dependent on various parameters, which increases the complexity of the FRCM-to-concrete bond. Analytical models cannot provide a high degree of accuracy, as their predictions are only valid for specific datasets. Machine learning algorithms are the best-suited solution to deal with bond strength like complex problems. In this study, curve-fitting, Gaussian Process Regression (GPR) and Adaptive Neuro-Fuzzy Inference System (ANFIS) models have been applied to 336 aggregated datasets. Nine performance matrices have been opted to compare the performance of the developed models. Feature importance analysis has also been used to check the rationality of the model. The parametric analysis has also been done using the 3D-surface plot of the ANFIS model. The R-values of ANFIS, GPR, and curve-fitting models are 0.9895, 0.9882, and 0.9145, respectively. The mean absolute error, root mean square error and Nash-Sutcliffe index of the ANFIS model are 0.9168 kN, 1.4326 kN, and 0.9791, respectively. The mean absolute percentage error of the ANFIS model is 11.19%, which is 8.72% and 76.78% lower than GPR and curve-fitting model, respectively. The error range of the curve-fitting, GPR and ANFIS models are −17.06 kN to 18.04 kN, −4.39 kN to 6.07 kN, and −4.23 kN to 5.19 kN, respectively. Overfitting analysis of the proposed models has been done, and the predicted results show that the curve-fitting model and GPR models are inferior and the ANFIS model is superior based on the selected performance matrices. The overfitting value of ANFIS model is 67.89% and 8.31% lower than curve-fitting and GPR model, respectively. The sensitivity analysis found that the number of layers, the width of the concrete block, and the compressive strength of the concrete had the highest effect on the FRCM-to-concrete bond strength. The findings of the study have the potential to decrease costs and save time by employing an accurate prediction approach instead of expensive and time-consuming testing. The developed model can be easily used by industry experts and FRCM applicators to estimate the bonding strength of FRCM-to-concrete substrate for sustainable designs.</p>}},
author = {{Kumar, Aman and Arora, Harish Chandra and Kumar, Krishna and Garg, Harish}},
issn = {{0957-4174}},
keywords = {{ANFIS; Artificial intelligence; Bond strength; Curve-fitting; FRCM; GPR; Machine learning}},
language = {{eng}},
month = {{04}},
publisher = {{Elsevier}},
series = {{Expert Systems with Applications}},
title = {{Performance prognosis of FRCM-to-concrete bond strength using ANFIS-based fuzzy algorithm}},
url = {{http://dx.doi.org/10.1016/j.eswa.2022.119497}},
doi = {{10.1016/j.eswa.2022.119497}},
volume = {{216}},
year = {{2023}},
}