Presentation Type
Poster
Student
Yes
Abstract
Stiffened cylindrical shell buckling strength mainly depends on the geometric and stiffness properties. A detailed parametric study was conducted to investigate the influence of these properties on the stiffened aluminium cylindrical shell buckling strength. The proposed framework involves an integration of finite element method and various machine learning techniques. The dataset obtained from the eigenvalue buckling analysis of 350 numerical simulations using ANSYS workbench 2022; however, the FE simulation of ten ring-stiffened cylindrical specimens was initially substantiated by experimental work conducted in the literature. 350 sample specimens were categorized into seven groups based on the no. of stiffeners varying from 3 to 17 while their optimum sizes were obtained from optimization study. Each group consists of 50 samples with ten distinct values of length to diameter ratio and five distinct values of wall thickness. Dataset were trained (80%) and tested (20%) with various simple to complex machine learning algorithms such as kNN regression, linear regression, polynomial regression, random forest, and artificial neural networks. The predicted buckling strength obtained from each ML technique was compared to the numerical buckling strength. R2 and Mean Square Error (MSE) were considered as cost functions to evaluate the performance of each ML algorithm. A comparison between the proposed algorithms revealed that the artificial neural networks (ANN) performed excellent for both train and test data achieved highest accuracy with R2 of 0.996 and 0.993 while the associated MSE values are 0.47 and 0.97 respectively. Random forest and polynomial regression gave better R2 values but have slightly higher MSE values compared to the ANN. kNN and linear regression models are the least performing models, gave lower R2 and higher MSE values compared to the other ML techniques for the present dataset.
Keywords: Stiffened Cylinders, External Pressure, Buckling, Linear Analysis, ANSYS workbench, Optimization, Machine Learning, Artificial Neural network.
Start Date
2-6-2024 1:00 PM
End Date
2-6-2024 2:00 PM
Included in
Civil Engineering Commons, Electrical and Computer Engineering Commons, Structural Engineering Commons
Buckling Behavior of Thin Wall Stiffened Cylindrical Shells Through ML Techniques
Volstorff A
Stiffened cylindrical shell buckling strength mainly depends on the geometric and stiffness properties. A detailed parametric study was conducted to investigate the influence of these properties on the stiffened aluminium cylindrical shell buckling strength. The proposed framework involves an integration of finite element method and various machine learning techniques. The dataset obtained from the eigenvalue buckling analysis of 350 numerical simulations using ANSYS workbench 2022; however, the FE simulation of ten ring-stiffened cylindrical specimens was initially substantiated by experimental work conducted in the literature. 350 sample specimens were categorized into seven groups based on the no. of stiffeners varying from 3 to 17 while their optimum sizes were obtained from optimization study. Each group consists of 50 samples with ten distinct values of length to diameter ratio and five distinct values of wall thickness. Dataset were trained (80%) and tested (20%) with various simple to complex machine learning algorithms such as kNN regression, linear regression, polynomial regression, random forest, and artificial neural networks. The predicted buckling strength obtained from each ML technique was compared to the numerical buckling strength. R2 and Mean Square Error (MSE) were considered as cost functions to evaluate the performance of each ML algorithm. A comparison between the proposed algorithms revealed that the artificial neural networks (ANN) performed excellent for both train and test data achieved highest accuracy with R2 of 0.996 and 0.993 while the associated MSE values are 0.47 and 0.97 respectively. Random forest and polynomial regression gave better R2 values but have slightly higher MSE values compared to the ANN. kNN and linear regression models are the least performing models, gave lower R2 and higher MSE values compared to the other ML techniques for the present dataset.
Keywords: Stiffened Cylinders, External Pressure, Buckling, Linear Analysis, ANSYS workbench, Optimization, Machine Learning, Artificial Neural network.