Investigating the applicability of deep learning and machine learning models in predicting the structural performance of V-shaped RC folded plates

Katlav, MetinErgen, FarukTurk, KazimTurgut, Paki2024-08-042024-08-0420242352-4928https://doi.org/10.1016/j.mtcomm.2024.108141https://hdl.handle.net/11616/101784Reinforced concrete folded plates (RC-FPs), which are a special class of shell structures, have recently become very popular in modern architectural and engineering applications because of the need for lightweight and aesthetic structures to cover large areas. However, it is known that studies on the structural performance of RCFPs are insufficient. Therefore, this article presents a study on the development and comparison of different deep learning (DL) and machine learning (ML) models for the prediction of the structural performance of full-scale Vshaped RC-FPs produced from hybrid fiber-reinforced self-consolidating concrete (HFR-SCC) having different plate thicknesses (50, 60, 70, and 80 mm), fiber volumes (1.00% and 1.25%), and combinations (single, binary, and ternary). While vanilla long short-term memory (VLSTM) and bidirectional long short-term memory (BILSTM) are used as DL models, random forest (RF), extremely randomized trees (ERT), and adaptive boosting (AdaBoost) are preferred for ML models. To construct the models, the structural performance results of a total of 44 full-scale V-shaped RC-FPs subjected to four-point bending loading were adopted as the database. In addition to all these, the AdaBoost model is used to determine the relative feature importance of the input parameters. Based on the results, among the DL models, the BILSTM has the best ability to predict the structural performance values of V-shaped RC-FPs (such as R-squared values for maximum load-bearing capacity, cracking load, toughness, and deflection ductility are 0.934, 0.987, 0.972, and 0.812, respectively), while in ML models, this is valid for the ERT (such as R-square values are 0.917 for maximum load-bearing capacity, 0.936 for cracking load, 0.947 for toughness and 0.825 for deflection ductility). On the other hand, DL models predicted all other structural performance values better than ML models, except for deflection ductility. Besides, the most relative important input features for maximum load-bearing capacity and toughness values is plate thickness, whereas for cracking load and deflection ductility values compressive strength is important. In conclusion, it can be emphasized that the use of DL models can provide significant advantages in engineering applications, such as predicting the structural performance of V-shaped RC-FPs.eninfo:eu-repo/semantics/closedAccessStructural performanceV-shaped RC folded plate Deep learning,Machine learning Hybrid fiber-reinforced SCCShell constructionInvestigating the applicability of deep learning and machine learning models in predicting the structural performance of V-shaped RC folded platesArticle3810.1016/j.mtcomm.2024.1081412-s2.0-85183469058Q2WOS:001170478400001N/A