Semi-continuous direct-chill (DC) casting of aluminium ingots suffers from a very frequent defect which can lead to a strong decrease in productivity: hot tearing. This phenomenon takes place in the mushy state (i.e., at solid fractions, gs, lower than unity). In order to solve this problem, research moved towards the development of models for the prediction of the hot cracking sensitivity of any kind of alloy under given casting conditions. The defect being of complex nature and revealing some random aspects, the models must describe in the most precise way the interconnected mechanisms intervening in the formation and propagation of the crack. However, an approach supporting the separate study of each mechanism in order to integrate them, thereafter, in a single model, emerged as the way to follow from now on. The objective of this research task is to study the mechanisms leading to hot cracking of aluminium alloys during continuous casting processes, in order to deduce from them the fundamental parameters and mechanisms that will be integrated in the models to predict hot cracking sensitivity. For this purpose, two relevant aspects were targeted: the mechanical behaviour of the material in the semi-solid state as a function of the microstructure, and also the coalescence phenomenon. In order to study these phenomena, specific apparatus were developed in the laboratory. The first one, previously used by Farup et al. [Far01] for in-situ observation of organic alloys (succinonitrile-0.5 wt pct acetone), was improved in order to allow more precise observations under better controlled conditions of the hot cracking phenomenon in columnar grains. Then, in order to study the mechanical behaviour of an Al-4.5 wt pct Cu alloy as a function of three various types of microstructures (equiaxed, E, columnar radial parallel to the shear plane, C//, and columnar longitudinal perpendicular to this plane, Cperp), a torsion test performed on hollow cylindrical specimens in the semi-solid state was built, based on the original idea of Mahjoub et al. [Mah00]. Finally, an apparatus intended to tear a grain-refined Al-1 wt pct Cu alloy in the semi-solid state, avoiding any contact with air during the test, was designed. In order to optimize the tests on the Al-Cu alloys, numerical simulations of the thermal field in the specimens were made. Moreover, as a complement to the experimental results, numerical simulations of the mechanics for the torsion test were carried out so as to study the arrangement influence of the wet grain boundaries within a medium made up of hexagonal grains. From the tests on the organics, two distinct cases depending on the value of gs at which the grain boundary is stressed could be observed: either ''healing'' of the crack due to liquid feeding in the zone under constraints, or the formation of a hot crack. Finally, temperature ranges within which intra- and intergranular coalescence start to take place were established and used in a simple model (Rappaz-Drezet-Gremaud [Rap99]) for the prediction of hot cracking. Temperatures for coherency loss of an Al-4.5 wt pct Cu alloy, as a function of the microstructure, were obtained using the torsion test. Moreover, creep tests made it possible to describe the deformation mode depending on the type of microstructure tested. An evolution of the E and C// samples behaviour towards that of a fully solid material could be highlighted by fitting the curves with a Norton's power law, in the case of an increase of gs from 90 to 92 %. The Cperp samples always showed a higher resistance to deformation compared to that of the other microstructures. The tearing test results highlighted a strong increase in the maximal force needed to break a grain-refined Al-1 wt pct Cu specimen as gs reached approximately 95 %. The latter is attributed to a start of coalescence between the grains. Observations under the microscope of the broken surfaces confirmed this assumption by revealing a transition from a smooth surface with a dendritic aspect to a surface containing important sharp-edged areas. An evolution in the shape of the spikes left by the broken bridges could also be highlighted. The temperature versus solid fraction curves were determined using the thermal analysis method developed at the laboratory by Campanella et al. [Cam01]: the single pan thermal analysis (SPTA). Indeed, this method allows a closer description of the actual solidification path for systems out of equilibrium. Finally, the study by successive transverse sectioning of a DC cast billet containing a hot crack was undertaken. The results point out the crack evolution in the case of a constant increase of the casting speed. The crack starts as a concentration of microporosity in the centre of the billet that quickly changes to an annular shape made of microcracks. Then, it moves off centre taking the shape of a linear crack. Eventually, the crack moves again to the billet centre with a typical "star" crack shape.