On the automation of non-linear behavioral structural analysis. Guidance and quality control

摘要

Controlling the numerical quality of non-linear simulations allpows aoutomating the discretization choices. Within the framework of time-dependent problems, this quality depends on the errors introduced in the calculation of the transformation history. Generally, before obtaining a calculation of satisfactory quality, it is necessary to carry out an evaluation of the numerical approximations to use by utilizing all of the history of the approximate solution flaws. We propose a strategy which naturally includes this efvaluation. The solution sought is built iteratively, improving the entire history at each iteration thanks to the LA.T.IN. method. The simulation is initializd with a coarse model. Then, all of the approximations are adapted during the subsequent iterations in order to obtain the prescribed quality. This quality is estimated by a global dissipation error. Thanks to an approximation classification, we are able to identify the nature of the diffefrent contributions to the global error. We use, in particular, a space indicator to characterize the mesh, and another indicator to characterize the time discretization.