ANALYSIS OF INFILL FRAMES USING NEURAL NETWORKS

Abstract

High raised buildings are vulnerable to the lateral loads caused during natural calamities like earthquakes and cyclones. Lateral loads always affect the stability of structures. The shear walls, which are provided to resist these lateral forces, are often considered undesirable from utility point of view as they come in the way of vehicular movement in the parking area. Methods to evaluate the lateral strength from infill wall panels are studied here. In this paper, the contribution offered from infill walls to the lateral strength of the building is evaluated using a combination of FEM approach and Neural Networks. Since the integral action of the wall panels with the bounding frame is considered in this approach, the frame is called as infill frame. In the study the infill wall contribution is replaced using two diagonal struts joining the opposite corners of the frame. The Finite Element package MSC Nastran is used to evaluate the infill contribution. Frames with different aspect ratios of span to height, with and without wall openings are analysed. The strain energy of the infill frame is equated to the strain energy of frame with diagonal struts, from which the width of the equivalent struts is obtained. The equivalent diagonal struts are provided joining the opposite corners of the frame. In the present study the non linear behaviour of concrete has also been taken into consideration. To make the method more widely applicable, a large data bank with varying parameters of span/depth, beam/column sections, with and without infill openings and different grades of concrete is generated which shall be used for training a BOPN based neural net paradigm. The trained neural net is expected to furnish the width of equivalent strut, which can readily be used in the structural analysis of framed buildings. The utility of the study enhances as more training patterns are added to the neural net.