Observatorio de I+D+i UPM

Descripción
Automatic differentiation is a very powerful computational-mathematical technique that is capable of differentiating any type of computer function. Among its advantages over other ways of computing derivatives, the most important are accuracy, scalability, efficiency and short development time. This article is about the implementation of automatic differentiation in a recursive penalty formulation for the dynamic simulation of multibody systems. Specifically, the technique is implemented in the most expensive step of the formulation: the differentiation of a set of forces in the form of a Jacobian matrix. Both the details of the formulation and the basics of automatic differentiation appear here thoroughly explained. The numerical examples at the end prove that automatic differentiation overcomes some of the limitations of numerical differentiation, and they offer interesting information about the influence of the problem size in the efficiency of the simulation. In some examples, the speedup of automatic differentiation with respect to numerical differentiation reaches 64%. Summing up, this paper provides an original perspective for the optimization of derivatives in the context of efficient multibody formulations.
Internacional	Si
Nombre congreso	ECCOMAS Multibody Dynamics 2011, J.C. Samin, P. Fisette (eds.), Brussels, July 4-7, 2011.
Tipo de participación	960
Lugar del congreso
Revisores	Si
ISBN o ISSN	978-2-8052-0116-5
DOI
Fecha inicio congreso	04/07/2011
Fecha fin congreso	07/07/2011
Desde la página	1
Hasta la página	10
Título de las actas	ECCOMAS Multibody Dynamics 2011, J.C. Samin, P. Fisette (eds.),

Esta actividad pertenece a memorias de investigación

• Autor: Alfonso Callejo Goena (UPM)
• Autor: Francisco Javier Garcia de Jalon De la Fuente (UPM)

• Creador: Grupo de Investigación: Grupo de Inv. en Seguridad e Impacto Medioambiental de Vehículos y Transportes (GIVET)
• Centro o Instituto I+D+i: Instituto Universitario de Investigación del Automóvil (INSIA)