Yazar "Ergen F." seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim Yok
    Öğe
    AI-driven design for the compressive strength of ultra-high performance geopolymer concrete (UHPGC): From explainable ensemble models to the graphical user interface
    (Elsevier Ltd, 2024) Katlav M.; Ergen F.; Donmez I.
    Ultra-high performance geopolymer concrete (UHPGC) has become of interest in recent years as a more economical and sustainable alternative while offering similar mechanical performance to ordinary ultra-high performance concrete (UHPC). The lack of an effective mix design methodology has inhibited the widespread use of UHPGC, despite its potential. This paper adopted an artificial intelligence (AI)-based approach to accurately model the compressive strength (CS) of UHPGC, a critical parameter to ensure structural integrity and reliability. Ensemble machine learning (ML) models such as RF, XGBoost, LightGBM and AdaBoost, which have been very popular lately, were selected as AI algorithms. For the establishment of these models, a comprehensive and reliable dataset of 181 test results was used, including 13 input features. Additionally, feature importance and Shapley additive explanations (SHAP) analyses were used to ensure the explainability of the prediction models and tackle the “black box” challenge of ML models. The results obtained revealed that all ensemble models successfully predicted the CS of UHPGC; in particular, the XGBoost model consistently exhibited the best overall performance in terms of higher R2 (0.948) and lower RMSE (6.68), MAE (4.73), MAPE (4 %), and mean error value (1.095), in the test phase. Moreover, feature importance and SHAP analyses revealed that the most influential features on the CS of UHPGC were age, fiber and silica fume, sodium silicate (Na2SiO3), and water content. Lastly, a graphical user interface (GUI) was developed to easily predict the CS of UHPGC in practical applications. © 2024 Elsevier Ltd
  • Küçük Resim Yok
    Öğe
    Machine and deep learning-based prediction of flexural moment capacity of ultra-high performance concrete beams with/out steel fiber
    (Springer Nature, 2024) Ergen F.; Katlav M.
    This paper presents a study to develop the most appropriate machine learning (ML) and deep learning (DL) models for predicting the flexural moment capacity of ultra-high performance concrete beams (UHPC-Bs) with/out steel fibers. As prediction models, five different ML models (support vector regression (SVR), extreme gradient boosting (XGBoost), multilayer perceptron (MLP), random forest (RF), and extremely randomized trees (ERT)) and three different DL models (long short-term memory (LSTM), bidirectional LSTM (BILSTM), and gated recurrent unit (GRU)) are adopted in this study. To train and evaluate these prediction models, a total of 82 experimental results from the literature on rectangular UHPC-Bs with/out steel fibers are used as a database. On the other hand, the Shapley Additive Explanations (SHAP) analysis is used to evaluate the effects of input features on the prediction results. Furthermore, based on the best prediction model obtained in this study, a graphical user interface is designed to facilitate the use of the flexural moment capacity of UHPC-Bs with/out steel fibers in practical applications. The results obtained indicated that both ML and DL models can be effectively used to predict the flexural moment capacity of UHPC-Bs with/out steel fibers. Particularly, ML algorithm such as XGBoost and DL algorithm like BILSTM stand out for their ability to predict the flexural moment capacity with high accuracy. Furthermore, SHAP analysis revealed that the longitudinal tensile reinforcing steel bar area (As), effective depth (d), and fiber-reinforcing index (Vf × lf/df) are the most effective parameters for prediction performance, while the compressive strength (fc) of concrete is less effective. Consequently, the developed prediction models and graphical user interface (GUI) will help structural engineers and designers quickly predict the flexural moment capacity of UHPC-Bs with high accuracy and make the design process more efficient. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024.