Mesomechanical modelling of asphalt

Asphalt is one of the oldest construction materials known to mankind. Unfortunately, it is also one the most complex materials. Asphalt exhibits elastic, viscous and plastic properties. Furthermore, it damages and shows significant ageing over the course of a few decades. These properties are inherited mostly from the bituminous binder. However, asphalt is actually a random heterogeneous material consisting of mineral aggregate glued together with bitumen. Mixed together, asphalt acquires all the properties of a true composite material: stiffening and strengthening, localisation, scale dependency, and many more. The present work deals with the modelling of asphalt used in road construction. Due to its complexity, asphalt is difficult to model as a bulk material on the macroscale. The rationale is to develop a mesoscale model of asphalt within a computational mechanics multiscale framework. The mechanical field equations are solved using the finite element method. Here, asphalt is separated into two components, namely asphalt mortar and coarse aggregate. This divide and conquer approach leads to simpler submodels, and has the additional benefit of modularity. This cumulative thesis starts off with background information about the history of asphalt, its constituents, mechanical properties and design (section 1).Following the idea of modularity, the first publication (section 2) establishes the multi scale framework in general. A first method to generate geometries of random heterogeneous angular particle systems based on Voronoi tessellation is developed (section 2.3). Tetrahedral finite element meshes are generated (section 2.3.3) and a strain driven first order homogenisation method is employed to compute macroscale material parameters (section 2.5). So far, material modelling resorts to linear elasticity (section 2.4).The second publication (section 3) is specialising almost every aspect of the previous publication w.r.t. road engineering. Material modelling is based on linear viscoelastic data obtained by dynamic shear rheometry. The generalised Maxwell model is used (section 3.3). Geometry generation is extended to represent particle size distributions of several real mixtures (section 3.4). The meshing procedure does now use coarse-graining for volumes where high accuracy is not needed (section 3.5.2). Upscaling from meso- to macroscale is performed in time domain at a temperature of 20°C and a frequency of 1 Hz (section 3.5). At this stage, validation against experimental mixdata becomes possible. The realm of computational mechanics is somewhat left in the third publication (section 4), which concentrates on the analysis of irregular polyhedra instead. Particular, the ambiguous notion of “size” for non-spherical particles is assessed. Theme thodology is based on principal component analysis of the inertia tensor using discrete computational geometry. This is used as a basis to propose estimators for sieve size (section 4.3). Several shape descriptors are introduced and compared between data from X-Ray CT scanning (section 4.3.5) and synthetic data (section 4.3.6). With suitable datasets at hand, the quality of sieve size estimation as a function of sphericity is investigated (section 4.5.2 and section 4.5.3). Additional investigations, which are mostly unpublished yet, are presented in section 5. Unfortunately, some shortcomings have been revealed after publication of the three articles, which are also adressed in this chapter