Optimum Design of Fully Composite, Unstiffened, Built-Up, Hybrid Steel Girder Using GRG, NLR, and ANN Techniques

• Published in: Journal of engineering (Cairo, Egypt) Vol. 2022; pp. 1 - 25
• Main Authors: El-Aghoury, Mohamed A., Ebid, Ahmed M., Onyelowe, Kennedy C.
• Format: Journal Article
• Language: English
• Published: Cairo Hindawi 2022; Hindawi Limited; Wiley
• Subjects: Artificial neural networks; Bending moments; Composite beams; Concrete; Costs; Cross-sections; Design optimization; Error analysis; Flanges; Flooring; Genetic algorithms; Linear programming; Low level; Metal fatigue; Multistory buildings; Optimization techniques; Prediction models; Static loads; Steel beams; Steel structures; Webs
• ISSN: 2314-4904; 2314-4912; 2314-4912
• DOI: 10.1155/2022/7439828

Composite steel beams are commonly used element in multistorey steel buildings to enhance floor economy and serviceability and provide more clear height. Due to the low level of stress in the webs of such beams, hybrid sections are used where the flanges have higher strength than the webs. A lot of earlier research was carried out to optimize the design of the hybrid and nonhybrid composite steel beams under both static loading and dynamic behavior. However, there is still a need to develop a more practical optimization method. The aim of this research is to develop simple and practical equations to determine the optimum cross section dimensions for both shored and unshored, simply supported, hybrid and nonhybrid, composite steel beam under static loads. To achieve that goal, a research program of two phases was carried out. The first phase was generating a database of 504 composite beams with different steel grades for flanges and webs, subjected to different values of bending moment. The cross section of each beam in the database was optimized using GRG technique to minimize the cost considering the unit price of each steel grade. In the second phase, the generated database was divided into training and validation subsets and used to develop two predictive models using Nonlinear Regression (NLR) technique and Artificial Neural Network (ANN) technique to predict the optimum cross section dimensions and hence the optimum weight and cost. The accuracies of the developed models were measured in terms of average error percent. NLR and ANN models showed average error percent of 16% and 11%, respectively.