Experimental Examination of the Behavior of Shotcrete-Reinforced Masonry Walls and Xgboost Neural Network Prediction Model

Suzen A. A., Cakiroglu M. A., İNCE G., KABAŞ H.

ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING, vol.46, no.11, pp.10613-10630, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 46 Issue: 11
  • Publication Date: 2021
  • Doi Number: 10.1007/s13369-021-05466-1
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Metadex, Pollution Abstracts, zbMATH, Civil Engineering Abstracts
  • Page Numbers: pp.10613-10630
  • Keywords: Dry-mix shotcrete, Ensemble learning, Masonry building, Out-of-plane loading, Polypropylene fiber, XGBoost
  • Süleyman Demirel University Affiliated: Yes


In this study, the effects of reinforcement using polypropylene fiber dry-mix shotcrete on the behaviors of U-shaped masonry walls were experimentally studied, and predicting modeling was performed with XGBoost. Firstly, five full-scale masonry specimens were constructed. A specimen was left unstrengthened. The other four specimens were coated with dry-mix shotcrete with layer thicknesses of 50 and 100 mm, reinforced with additional amounts of polypropylene fiber. All specimens were tested under reversible and cyclic out-of-plane loads. The results showed that the strengthened specimens had a considerably higher ultimate load-carrying and energy absorption capacities than the bare one. In the second stage of the study, an XGBoost neural network model was developed, predicting the ultimate load capacity and energy absorption capacity with data affecting the result of the model. The ultimate load capacity and energy absorption capacity values obtained as a result of the tests were compared with the results derived from the experiments, and then it was observed that the results of XGBoost modeling were quite close to the results obtained from experimental data. Thus, the thickness of shotcrete required for the desired data can be painlessly predicted using XGBoost model without the need for were experimentally studied.