In composite processing, a compressible dry fiber preform is in many cases impregnated by a molten polymeric matrix. The impregnated part is sometimes used as produced, but most often it requires a reheating step before stamping, flow molding or even painting of the part. This latter step may lead to the release of locked-in stresses, a phenomenon often called lofting or deconsolidation. The aim of the present thesis is to analyze the mechanisms and provide tools for predicting the kinetics of consolidation and deconsolidation in order to produce cost-effective and sound composite materials. To achieve this, the two fundamental stages of composite processing, consolidation and deconsolidation, have been studied. The influence of both processing parameters and materials systems have been quantified. A Glass Mat reinforced Thermoplastic (GMT) was selected and characterized for the case study. The first part of this work concerns the consolidation stage, emphasizing the influence of processing parameters such as applied pressure or processing temperature on the final microstructure of the composite products. Experimental impregnation processes were carried out and were completed by parallel theoretical simulations. A model accounting for the saturated infiltration of compressible preform was used to predict the residual fiber content gradient in a part, and its influence on the mechanical properties. It was observed that a gradient of the reinforcement in a GMT part, if controlled by impregnation times, can improve the bending modulus by 50%. Furthermore, a new infiltration model accounting for multi-phase flow in compressible preforms showing a dual-scale porosity was elaborated. The aim was to take into account the porosity by considering not only the solid and liquid phases, but also air as a third phase. The local fiber and void content can then be precisely predicted as a function of the infiltration time and position. The radial micro-impregnation of the fiber bundles was also taken into account by adding a sink term of micro-impregnation to the macro-impregnation. The predictions have shown that the time required to provide full micro-infiltration is about four times longer than the time to provide full macro-infiltration. Special care therefore has to be taken when infiltrating a dual-scale porosity system: residual voids may remain even if time to provide macro-infiltration is attained. Once the consolidation state of the part was well defined in terms of void content, fiber content profile and stress state, a study of the post-processing steps was carried out. To achieve this, model experiments of compression and unloading of a stack of mats embedded in a model matrix were carried out. This springback experiment enabled the unloading behavior of the fiber preform to be studied; for instance, the influence of the matrix viscosity on the kinetics of the process. Deconsolidation experiments were also carried out for GMT parts: the consolidated samples were reheated at processing temperatures for different times. Comparing these different experimental results with theory has shown that the deconsolidation phenomenon is mainly governed by the stress/strain behavior of the preform upon unloading. However, the kinetics of deconsolidation are also influenced by the void formation and growth that appear during reheating. Moreover, it was found that the springback effect of the preform leads to tensile forces in the matrix, which enhance void growth. A correlation between the elastic behavior of the fiber and the growth of air bubbles in the matrix was then demonstrated. Since it was observed that the mechanical behavior of the preform plays a major role in the deconsolidation phenomenon, the study on the springback effect was completed by a comparison between two different glass fiber and polypropylene systems. The classic GMT studied previously was compared with a new type of commingled glass and polypropylene fibers. Compressive tests on both dry preforms, as well as consolidation and deconsolidation experiments showed that the commingled system presents a higher void content after deconsolidation than the classic GMT. A porosity of respectively 72% and 55% was measured. It was demonstrated that differences in glass fiber arrangement are at the origin of the result. Finally, the deconsolidation of a GMT part in the solid sate was investigated. The role of the matrix which often exhibits a viscoelastic behavior and which may lead to distorsion of the part, was emphasized. After, for instance, 3 months at 100°C, the part showed no significant void content increase whereas, it was observed that thermal effects may affect the adhesion between the matrix and the fibers, thereby leading to decohesion of the part. This study demonstrated that a combination of numerical approach and experimental observations leads to a deeper understanding of the physical mechanisms governing consolidation and deconsolidation and then opens the path for an optimization of processing parameters and choice of materials.