Application of artificial neural network to predict buckling load of thin cylindrical shells under axial compression. (1st December 2021)

• Record Type: Journal Article
• Authors: Tahir, Zia ul Rehman; Mandal, Partha; Adil, Muhammad Taimoor; Naz, Farah
• Is Part Of: Engineering structures. Volume 248(2021)
• Publication Date: 2021-12-01
• Subjects: Thin cylindrical shells -- Axial compression -- Classical buckling theory -- Design recommendations -- Artificial neural network; Structural engineering -- Periodicals; Structural analysis (Engineering) -- Periodicals; Construction, Technique de la -- Périodiques; Génie parasismique -- Périodiques; Pression du vent -- Périodiques; Earthquake engineering; Structural engineering; Wind-pressure; Periodicals; 624.105
• DOI: 10.1016/j.engstruct.2021.113221
• Languages: English
• ISSNs: 0141-0296

Abstract

Highlights: ANNs were employed to predict buckling load of shells using experimental data. 450 shell specimens were used for training (90%) and validation (10%) of ANNs. ANN models predict within ±10% of experimental buckling load with few exceptions. ANNs can predict buckling loads with higher accuracy than design recommendations. Abstract: The buckling of thin cylindrical shells under axial compression is long-standing problem due to significant difference between experimental and theoretical buckling load predicted by classical buckling theory. The knockdown factors predicted by present design recommendations are very conservative and predictions are less accurate. Artificial Neural Networks (ANN) are used in this study to accurately predict buckling load using experimental data from 38 previous studies. The buckling load was predicted using nine input parameters. The experimental data was divided into two sets having similar distributions of input parameters: training dataset (90%) and validation dataset (10%). The buckling loads predicted by ANN are in good agreement with experimental buckling loads and predictions are within ±10% with few exceptions. The specimens with parameters falling in the range of input parameters were predicted well by ANN, and accuracy of the prediction depends on number of similar parameters in a specific range used for training. The predictions by five design recommendations including NASA and EC-3 were compared with experimental buckling load, the percentage errors of predictions compared to experimental data for more than 50% specimens were within ±50%. The trained ANN models predict buckling loads with higher accuracy compared to design recommendations and can be used for practical designs.