Computational Technology Resources - CCP

Keywords: neural networks, sensitivity analysis, composite beams, deflection, cracking, moment redistribution, design, elastic analysis, inelastic analysis, service load behaviour.

Summary

Composite beams may be subject to cracking near interior supports owing to high moments. This may lead to redistribution of moments and an increase in midspan deflections. The methods available presently for taking into account the effect of cracking of concrete are either iterative or incremental in nature and require significant computational effort.

In the present paper, a simplified technique of application of artificial neural networks (ANNs) has been proposed to drastically reduce the computational effort, without much loss in accuracy. The input parameters for the neural networks have been selected in such a way that they can be easily obtained from the material properties, geometric properties and the elastic analysis results (neglecting cracking). Nine significant input parameters have been identified based on the detailed sensitivity analysis. Sampling points of different parameters have also been identified using the sensitivity analysis to carry out the training. Data sets for training consist of different combinations of input parameters and resulting output parameters have been generated for training and testing of neural networks. An iterative method of analysis that incorporates cracking has been used for sensitivity analysis as well as for generation of data sets.

Multilayered feed forward networks have been chosen with a sigmoid function as the activation function. Back propagation algorithm has been used as training algorithm. The partitioning of data sets into training and testing data sets has been done using the hold out method. The neural networks developed have been validated for a number of example beams and the errors are shown to be small. The weight matrices for the neural networks have also been presented for their use in everyday design.

Two span, three span and nine span beams have been considered for developing the neural networks. For two span beams, one network has been developed for exterior spans (NN1); for three span beams, two networks have been developed one for each exterior span (NN2) and next to the exterior spans (NN3) whereas for nine span beams three networks have been developed one for each exterior span (NN4), next to the exterior spans (NN5) and the interior spans (NN6).

Trained neural networks are validated for a number of beams with a wide variation of input parameters. Example beams have been chosen in such a way that most of the input parameters have not been considered in the training or testing and none of the combinations of input parameters has been used in the training or testing. The maximum error in the prediction of inelastic deflection for any span of the example beams is 5.11%. The percentage errors are therefore small for practical purposes. Also the neural networks developed for a nine span beam are found to be applicable for the beam having more than three spans.

This methodology of application of neural networks for predicting the effects of cracking can be easily extended to large composite structures.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)

	Computational & Technology Resources an online resource for computational, engineering & technology publications
	not logged in - login
Front Page Browse CCP CSETS CTR IJRT Other Authors Search Purchase Guide FAQ Contact us	Civil-Comp Proceedings ISSN 1759-3433 CCP: 92 PROCEEDINGS OF THE FIRST INTERNATIONAL CONFERENCE ON SOFT COMPUTING TECHNOLOGY IN CIVIL, STRUCTURAL AND ENVIRONMENTAL ENGINEERING Edited by: B.H.V. Topping and Y. Tsompanakis Paper 31 Simplified Technique for the Design of Steel Concrete Composite Beams using Artificial Neural Networks S. Chaudhary¹ and A.K. Nagpal² ¹Department of Structural Engineering, Malaviya National Institute of Technology, Jaipur, India ²Department of Civil Engineering, Indian Institute of Technology Delhi, New Delhi, India doi:10.4203/ccp.92.31 purchase the full-text of this paper Full Bibliographic Reference for this paper S. Chaudhary, A.K. Nagpal, "Simplified Technique for the Design of Steel Concrete Composite Beams using Artificial Neural Networks", in B.H.V. Topping, Y. Tsompanakis, (Editors), "Proceedings of the First International Conference on Soft Computing Technology in Civil, Structural and Environmental Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 31, 2009. doi:10.4203/ccp.92.31 Keywords: neural networks, sensitivity analysis, composite beams, deflection, cracking, moment redistribution, design, elastic analysis, inelastic analysis, service load behaviour. Summary Composite beams may be subject to cracking near interior supports owing to high moments. This may lead to redistribution of moments and an increase in midspan deflections. The methods available presently for taking into account the effect of cracking of concrete are either iterative or incremental in nature and require significant computational effort. In the present paper, a simplified technique of application of artificial neural networks (ANNs) has been proposed to drastically reduce the computational effort, without much loss in accuracy. The input parameters for the neural networks have been selected in such a way that they can be easily obtained from the material properties, geometric properties and the elastic analysis results (neglecting cracking). Nine significant input parameters have been identified based on the detailed sensitivity analysis. Sampling points of different parameters have also been identified using the sensitivity analysis to carry out the training. Data sets for training consist of different combinations of input parameters and resulting output parameters have been generated for training and testing of neural networks. An iterative method of analysis that incorporates cracking has been used for sensitivity analysis as well as for generation of data sets. Multilayered feed forward networks have been chosen with a sigmoid function as the activation function. Back propagation algorithm has been used as training algorithm. The partitioning of data sets into training and testing data sets has been done using the hold out method. The neural networks developed have been validated for a number of example beams and the errors are shown to be small. The weight matrices for the neural networks have also been presented for their use in everyday design. Two span, three span and nine span beams have been considered for developing the neural networks. For two span beams, one network has been developed for exterior spans (NN1); for three span beams, two networks have been developed one for each exterior span (NN2) and next to the exterior spans (NN3) whereas for nine span beams three networks have been developed one for each exterior span (NN4), next to the exterior spans (NN5) and the interior spans (NN6). Trained neural networks are validated for a number of beams with a wide variation of input parameters. Example beams have been chosen in such a way that most of the input parameters have not been considered in the training or testing and none of the combinations of input parameters has been used in the training or testing. The maximum error in the prediction of inelastic deflection for any span of the example beams is 5.11%. The percentage errors are therefore small for practical purposes. Also the neural networks developed for a nine span beam are found to be applicable for the beam having more than three spans. This methodology of application of neural networks for predicting the effects of cracking can be easily extended to large composite structures. purchase the full-text of this paper (price £20) go to the previous paper go to the next paper return to the table of contents return to the book description purchase this book (price £78 +P&P)
Back to top	©Civil-Comp Limited 2023 - terms & conditions