A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Gusset Plate Compression Capacity Prediction Using Ensemble Machine Learning Models
Gusset plates have a vital role in aspect of stability and behaviour of bracing system and truss bridges, especially under compression, where the strength of the gusset plate connections is highly depended on boundary conditions and plate geometry. Even with an extensive amount of existing study and analysis, due to the complexity of the stress distribution mechanism in the connection area a high amount of uncertainty still remains in compressive strength design methods. With the growing interest of machine learning (ML) approach in performance evaluation of a structure, this study aims to develop a gusset plate compressive strength prediction model using the regression and ensemble machine learning techniques named support vector regression (SVR), decision tree regression (DT), XGBoost (XB), CATBoost (CB) and AdaBoost (AB), while avoiding the crude assumptions and complex calculations. These ensemble models are performed on a database of 68 experimental and 184 numerical datasets. In terms of performance evaluation, XB and CB have outperformed compared to the conventional design practice and empirical solutions, with an accuracy of 96% and 95%, respectively. To explain the models’ feature importance and performance evaluation, Shapley Additive exPlanations (SHAP) is used. After analysis with the existing empirical equations—Thornton (Eng J 21(3):139–148, [1]) and Chou et al. (Earthq Eng Struct D., 41(7):1137–1156, [2]), it is found that CB has shown the best performance in predicting the compressive strength of gusset plate with the lowest mean absolute error of 50.76 kN and Nash–Sutcliffe model efficiency coefficient of 0.984 compared with the existing methods, where the Modified Thornton method has achieved the highest accurate prediction among the existing methods with mean absolute error and Nash–Sutcliffe model efficiency coefficient of 227.22 kN and 0.872, respectively.
Gusset Plate Compression Capacity Prediction Using Ensemble Machine Learning Models
Gusset plates have a vital role in aspect of stability and behaviour of bracing system and truss bridges, especially under compression, where the strength of the gusset plate connections is highly depended on boundary conditions and plate geometry. Even with an extensive amount of existing study and analysis, due to the complexity of the stress distribution mechanism in the connection area a high amount of uncertainty still remains in compressive strength design methods. With the growing interest of machine learning (ML) approach in performance evaluation of a structure, this study aims to develop a gusset plate compressive strength prediction model using the regression and ensemble machine learning techniques named support vector regression (SVR), decision tree regression (DT), XGBoost (XB), CATBoost (CB) and AdaBoost (AB), while avoiding the crude assumptions and complex calculations. These ensemble models are performed on a database of 68 experimental and 184 numerical datasets. In terms of performance evaluation, XB and CB have outperformed compared to the conventional design practice and empirical solutions, with an accuracy of 96% and 95%, respectively. To explain the models’ feature importance and performance evaluation, Shapley Additive exPlanations (SHAP) is used. After analysis with the existing empirical equations—Thornton (Eng J 21(3):139–148, [1]) and Chou et al. (Earthq Eng Struct D., 41(7):1137–1156, [2]), it is found that CB has shown the best performance in predicting the compressive strength of gusset plate with the lowest mean absolute error of 50.76 kN and Nash–Sutcliffe model efficiency coefficient of 0.984 compared with the existing methods, where the Modified Thornton method has achieved the highest accurate prediction among the existing methods with mean absolute error and Nash–Sutcliffe model efficiency coefficient of 227.22 kN and 0.872, respectively.
Gusset Plate Compression Capacity Prediction Using Ensemble Machine Learning Models
Lecture Notes in Civil Engineering
Gupta, Rishi (editor) / Sun, Min (editor) / Brzev, Svetlana (editor) / Alam, M. Shahria (editor) / Ng, Kelvin Tsun Wai (editor) / Li, Jianbing (editor) / El Damatty, Ashraf (editor) / Lim, Clark (editor) / Arafin, Palisa (author) / Billah, A. H. M. Muntasir (author)
Canadian Society of Civil Engineering Annual Conference ; 2022 ; Whistler, BC, BC, Canada
Proceedings of the Canadian Society of Civil Engineering Annual Conference 2022 ; Chapter: 69 ; 1017-1032
2023-08-06
16 pages
Article/Chapter (Book)
Electronic Resource
English
Behavior and design of gusset plate connections in compression
| Online Contents | 2002