Author(s)

P.D. Panagiotopoulos & E.S. Mistakidis

Abstract

Neural networks for computing contact problems P.D. Panagiotopoulos"'\ E.S. Mistakidis" "Institute of Steel Structures, Aristotle University, GR-54006 Thessaloniki, Greece ^Faculty of Math, and Pyhsics, RWTH, D-5100 Aachen, Germany ABSTRACT A neural network model is proposed and studied for the treatment of struc- tural analysis contact problems. Moreover new results generalizing the re- sults of Hopfield and Tank are obtained. Numerical applications illustrate the theory and show clearly the advantages of the neural network approach to inequality problems. Finally the parameter identification problem is for- mulated and solved as a "supervised learning" problem. 1. INTRODUCTION Neural network models are very efficient in computing where many assump- tions have to be satisfied in parallel, e.g. in image recognition problems. This is achieved, in contrast to the classical sequential computers, by us- ing networks of a high degree of interconnection of analog neurons with non

Keywords