Neural analysis of vibration problems of real flat buildings and data pre-processing

Abstract

Two vibration problems are discussed in the paper: i) identification of the fundamental (first) period of building vibration T 1, ii) simulation of the maximal displacement of a selected building floor D. The problems are related to prefabricated, five-storey flat buildings. The analysis is based on measurements of paraseismic excitations and dynamic responses of buildings. The identification problem is formulated as the evaluation of a relation between structural and soil basement parameters, and the fundamental period T 1. The simulation problem corresponds to the evaluation of a relation between inputs composed of structural, soil and excitation parameters and the displacement output D. Back-propagation neural networks (BPNNs) were successfully used in [9] to the analysis of identification problem. In the paper this problem is reanalized by BPNNs in which the soil basement parameter C z is changed from the crisp variable into a linguistic variable C l. The compression of discrete response spectrum values is additionally carried out to decrease the BPNN size for inputs in the simulation problem.