Eigenvalue determination by mixed modular networks

Abstract

Smart systems in general, and specifically neural networks, are expected to be of great assistance in large scale matrix computation. However it was necessary to work with smaller problems at first step. To this end, a method based on using perceptrons and Kohonen networks for determining the first three modal frequencies of frames of up to 20 stories high and two bays wide from their stiffness and mass matrices, was developed. This paper introduces this modular network as a preferable alternative for designing smart systems for learning large scale mapping problems in structural engineering, and in addition, reports the successful application of this modular network to the eigenvalue and modal frequency determination of shear frame structures.