Metodika použití umělých neuronových sítí pro identifikaci parametrů výpočtových modelů konstrukcí

Methodology of using artificial neural networks for identification of computational model parameters

Paper in proceedings (conference paper)

The paper suggests a new approach of inverse analysis to obtain parameters of FEM computational model in order to obtain best agreement with experimental data. The proposed inverse analysis approach is based on coupling of FEM computational model and the stochastic training of artificial neural network. Identification parameters play the role of basic random variables with a scatter reflecting the physical range of possible values. Novelty of the approach is the utilization of efficient small-sample simulation method Latin Hypercube Sampling (LHS) used for stochastic training of neural network. Once the network is trained it represents an approximation consequently utilized in an opposite way: For given experimental data to provide the best possible set of model parameters. The approach is general and can be applied easily to any inverse problem of engineering mechanics.

The paper suggests a new approach of inverse analysis to obtain parameters of FEM computational model in order to obtain best agreement with experimental data. The proposed inverse analysis approach is based on coupling of FEM computational model and the stochastic training of artificial neural network. Identification parameters play the role of basic random variables with a scatter reflecting the physical range of possible values. Novelty of the approach is the utilization of efficient small-sample simulation method Latin Hypercube Sampling (LHS) used for stochastic training of neural network. Once the network is trained it represents an approximation consequently utilized in an opposite way: For given experimental data to provide the best possible set of model parameters. The approach is general and can be applied easily to any inverse problem of engineering mechanics.

Inverse analysis, computational model, stochastic neural network, small-sample simulation.

NOVÁK, D.; LEHKÝ, D.

11.05.2006

Brno, ČR

80-214-3164-4

Dynamicky namáhané konstrukce - DYNA

115

122

8

http://hdl.handle.net/