Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SECOND INTERNATIONAL CONFERENCE ON SOFT COMPUTING TECHNOLOGY IN CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING
Edited by: Y. Tsompanakis, B.H.V. Topping
Bayesian Neural Networks Assessment of Stud Shear Connectors
E.S. Goulart1, P.C.G.S. Vellasco1, R.R. de Araujo2 and M.M.B.R. Vellasco3
1Structural Engineering Department, State University of Rio de Janeiro, UERJ, Brazil
E.S. Goulart, P.C.G.S. Vellasco, R.R. de Araujo, M.M.B.R. Vellasco, "Bayesian Neural Networks Assessment of Stud Shear Connectors", in Y. Tsompanakis, B.H.V. Topping, (Editors), "Proceedings of the Second International Conference on Soft Computing Technology in Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 25, 2011. doi:10.4203/ccp.97.25
Keywords: stud shear connectors, Bayesian neural networks, push-out test.
Shear connectors are used in composite structures to ensure the joint action between the steel profile and the concrete slab. Nowadays the most common types of shear connectors are stud bolts, the Perfobond rib (here investigated) and the "T" connector.
Various push-out tests have been conducted to determine this shear connector structural response. The push out tests consists in a vertically positioned steel profile connected to two adjacent concrete slabs by shear connectors. The use of neural networks was envisaged to create new data after being properly trained and tested with the existing experimental evidence.
An artificial neural network is represented by weighted interconnections between processing elements (PEs). These weights are the parameters that actually define the non-linear function performed by the neural network. The process of determining such parameters is called training or learning, and relies on the presentation of many training patterns.
In this work we decide to use the Bayesian neural networks (BNN), based on a supervised training algorithm. The BNN defines a systematic way to update the synaptic weights of networks, which are composed of an input layer that receives the input values, an output layer, which calculates the neural network output, and one or more intermediary layers, called hidden layers. In all the Bayesian neural network trainings only one hidden layer was adopted where the number of processors varied i.e. 1, 2, 3, 4, 5, ...10. The input network data were related to proprieties of concrete slab, reinforcing steel bars that passes through the Perfobond holes and Perfobond connector. The network output is the ultimate experimental load resisted by the connector at the push out tests. When the initial experimental 58 dataset was used the neural networks did not present a good generalization as a result of the small dataset. Noisy data were then introduced as inputs and outputs and proved to be efficient for improving the BNN performance. The most efficient networks were also associated to a single hidden layer containing three and four processors.
The errors obtained (RMSE and MAPE) proved to be acceptable (considering the safety coefficients used in structural engineering), confirming the possibility of using this method to create new additional data. The generation of this new data is much simpler and more economic than numerous laboratory tests or even finite element non-linear simulations.
purchase the full-text of this paper (price £20)