Despite its importance ductility remains one of the most difficult parameters to predict and there is a need to model it more precisely. However, there is also a lack of quantitative experimental results to assess the validity of the models due to the stochastic nature of the ductile fracture process. In this thesis, model materials have been fabricated and tested in situ either in a scanning electron microscope or by x-ray tomography. The first model material was a metal matrix composite with a special core/shell design which prevented the composite from breaking in a stochastic manner. The second consisted of a single metallic sheet containing laser drilled holes (2D approach) which could be diffusion bonded to sheets without holes to obtain a regular array of holes in the bulk of the material (3D approach). The laser micromachining of metals has been investigated in detail to verify that it is suitable for the fabrication of the model materials.

The results show that the whole ductile fracture process which consists of the nucleation, growth and coalescence of voids can be visualized in detail. For the composite material, the strong effect of the stress triaxiality on the void nucleation, growth and coalescence has been observed. In order to be able to predict the decrease in coalescence strains with increasing stress triaxiality, the Brown and Embury model for void coalescence has been successfully modified. The results from the laser drilled materials showed that depending on the hole configuration, different failure modes and failure strains are observed. The importance of the void spacing and of constraining effects on coalescence were demonstrated. Also, depending on the material, there is a competition between ductile fracture and shear localization. In the 2D approach it is shown that the Rice and Tracey model for void growth and the Thomason model for void coalescence are not appropriate to predict the behavior of this model material. However, the Brown and Embury model gives excellent predictions of void coalescence. In the 3D approach the void growth predictions from the Rice and Tracey model are in good agreement with the experimental results. The Brown and Embury model cannot however capture the much higher deformations prior to coalescence found experimentally. This is due to constraining effects which delay coalescence. The Thomason model gives excellent predictions in 3D for the copper samples containing holes coalescing normal to the tensile axis. However, it overestimates the failure strains in the case of the Glidcop samples because of strain localization and secondary void nucleation which precipitate failure. Finally, a finite element model has been built which captures well void growth and coalescence. Damage is introduced through an element deletion technique using a nonlocal version of the Gunawardena damage indicator with a characteristic length equals to the void diameter. It is shown that a nonlocal approach is required in order to predict both failure path and failure strains.