Optimierung und Analyse von Fachwerkstrukturen durch neuronale Netze

Problemspecific neural networks, multimembered evolutionary strategy, multicriteria optimization, pruning of neural networks, damage analysis, solution of inhomogeneous linear equation systems using problemspecific neural networks Abstract The integration of neural networks and a multimembered evolutionary strategy leads to a new multicriteria optimization approach for plane truss structures. Conventional as well as problemspecific neural networks describe the appertaining restrictions and criteria of the investigated optimization problem. Advantages of this new approach are the minimal response time required by trained neural networks in the context of a stochastic optimization method as well as a probability greater than zero to find global extrema in multimodal optimization problems. Conventional neural networks are able to assess the impairment of damaged structural I systems. The fully connected starting topologies of these networks are minimized with a combination of pruning algorithms and conventional network training. Statically determi-nate and indeterminate systems are investigated with a main focus on indeterminate truss structures and related problems concerning sufficient data diversification and adequate data preprocessing. As a result neural networks are able to solve specific inverse functional relations, and thereby assess the impairment of a damaged structural system with a minimized network topology. This research work also presents a new methodology for the solution of inhomogeneous linear equation systems using simple problemspecific neural networks. ich für die wertvollen Hinweise nach der Durchsicht der Arbeit.