Error assessment and adaptivity for structural transient dynamics

• Autores: Francesc Verdugo Rojano
• Directores de la Tesis: Núria Parés Marine (dir. tes.), Pedro Díez Mejía (dir. tes.)
• Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2013
• Idioma: inglés
• Tribunal Calificador de la Tesis: Antonio Huerta Cerezuela (presid.), Marino Arroyo Balaguer (secret.), Antonio Javier Gil Ruiz (voc.), Nicolas Moes (voc.), Wolfgang Wall (voc.)
• Materias:
  • Matemáticas
    • Análisis numérico
      • Análisis de errores
  • Ciencias tecnológicas
    • Tecnología de la construcción
      • Resistencia de estructuras
• Enlaces
  • Tesis en acceso abierto en: TDX
• Resumen

  • The finite element method is a valuable tool for simulating complex physical phenomena. However, any finite element based simulation has an intrinsic amount of error with respect to the exact solution of the selected physical model. Being aware of this error is of notorious importance if sensitive engineering decisions are taken on the basis of the numerical results. Assessing the error in elliptic problems (as structural statics) is a well known problem. However, assessing the error in structural transient dynamics is still ongoing research. The present thesis aims at contributing on error assessment techniques for structural transient dynamics. First, a new approach is introduced to compute bounds of the error measured in some quantity of interest. The proposed methodology yields error bounds with better quality than the already available approaches. Second, an efficient methodology to compute approximations of the error in the quantity of interest is introduced. The proposed technique uses modal analysis to compute the solution of the adjoint problem associated with the selected quantity of interest. The resulting error estimate is very well suited for time-dependent problems, because the cost of computing the estimate at each time step is very low. Third, a space-time adaptive strategy is proposed. The local error indicators driving the adaptive process are computed using the previously mentioned modal-based error estimate. The resulting adapted approximations are more accurate than the ones obtained with an straightforward uniform mesh refinement. That is, the adapted computations lead to lower errors in the quantity of interest than the non-adapted ones for the same number of space-time elements. Fourth, a new type of quantities of interest are introduced for error assessment in time-dependent problems. These quantities (referred as timeline-dependent quantities of interest) are scalar time-dependent outputs of the transient solution and are better suited to time-dependent problems than the standard scalar ones. The error in timeline-dependent quantities is efficiently assessed using the modal-based description of the adjoint solution.