Estimation of accidental eccentricities for multi-storey buildings using artificial neural networks

Abstract

Most of the research work on the torsional effect of the accidental eccentricity has been based on single-storey building models. Although the accidental eccentricity parameter is expressed for each individual floor, but the overall torsional behaviour of a multi-storey building can be significantly different from that of a single-storey model. In this context, a model of artificial neural networks (ANN) is elaborated to predict the accidental eccentricity in each floor of symmetrical and asymmetrical multi-storey buildings using acceleration or displacement records at the base and on the structure as well as the measured natural vibration frequencies of the structure. The database used for learning is constituted by varying randomly the live loads, the stiffness and the mass of the structural elements. The developed model is applied to eight standard multi-storey buildings. The calculated accidental eccentricities were in good agreement with the corresponding target accidental eccentricities. This generalized model for multi-storey buildings is an efficient tool to predict the real accidental eccentricities of such structures under different level of seismic loadings.