Reports: DNI654240-DNI6: Mesoscale Simulation and Machine Learning of Asphaltene Aggregation

Andrew L. Ferguson, PhD, University of Illinois at Urbana-Champaign

Synopsis. We have established mesoscale models of asphaltenes parameterized from all-atom simulations, and used these models to simulate the self-assembly of hundreds asphaltene molecules at 10 nm length scales and microsecond time scales. By combining coarse-grained models parameterized by atomistic data with high performance simulation software running on GPU hardware, we have performed the first molecular dynamics simulations to attain such length and time scales with molecular-level detail, thereby permitting direct simulation of the complete assembly hierarchy posited by the Yen-Mullins model. The morphological trends observed in our mesoscale simulations are in broad agreement with Yen-Mullins, in that we observe hierarchical assembly from monomers to nanoaggregates to clusters to a spanning viscoelastic network. Providing molecular-level insight of the assembly mechanisms, we have uncovered new understanding of asphaltene assembly as a function of concentration, temperature, and chemistry. This work supports the professional development of the graduate researcher by providing him training in molecular simulation, statistical mechanics, and scientific writing, and this year he will attend a national conference (likely APS) to present this work. This work also permits the PI to leverage his training in molecular simulation and statistical mechanics to establish himself as an independent young investigator in a new field with credentials in self-assembly and mesoscale simulation.

Mesoscale model construction. We have constructed mesoscale models of three prototypical asphaltene molecules – A0, A1, and A2 (Fig. 1) – selected to assess the influence of the size of the aromatic core and number of aliphatic side chains on assembly. We conducted all-atom simulations of monomers and dimers of each architecture in heptane solvent using the GROMACS simulations suite employing the GROMOS 54a7 force field. We generated coarse-grained mappings of each molecule under the coarse-grained Martini force field lumping ~4 atoms into each coarse-grained bead to enable accelerated simulations by both removing degrees of freedom and smoothing the potential energy landscape. The distribution of bond lengths, bond angles, torsional angles in each coarse-grained model were adjusted to match those observed in the all-atom model simulations by optimizing the baseline Martini bonded parameters using Boltzmann inversion. The non-bonded interactions in the coarse-grained model were adjusted such that the potential of mean force (PMF) for dimerization matches that computed in all-atom calculations (Fig. 2).

Fig. 1 – Prototypical asphaltene molecules (a) A0, (b) A1, and (c) A2 studied in this work. Each panel shows the all-atom representation of the asphaltene molecule, its coarse grained mapping under the Martini model, and its molecular weight.

Fig. 2 – The standard Martini bonded and non-bonded interaction parameters were adjusted to match the intramolecular distributions of bond lengths, angles, and torsions and the dimerization PMF observed in all-atom simulations. (a) As an illustration of the Boltzmann inversion procedure, the Martini force constant for the harmonic bending of angle #6 in molecule A2 was adjusted by Boltzmann inversion to match the distribution of bond angles observed in an all-atom simulation. (b) As an illustration of the PMF matching procedure, the Martini non-bonded parameters for molecule A1 were optimized such that the dimerization PMF matched that computed in all-atom calculations.

Mesoscale assembly mechanisms. Using our coarse-grained model, we have simulated the assembly of up to 250 asphaltene molecules in heptane solvent over a concentration range of 5–25% mass fraction. In the case of A2, we observe the hierarchical formation of nanoaggregates, clusters, and ultimately a gelation transition to a viscoelastic network as we elevate concentration (Fig. 3). Nanoaggregates form by stacking of the aromatic cores possessing a fractal dimension of ~1.1, whereas clusters of nanoaggregates form by interactions between the asphaltene side chains, forming a connected network with a fractal dimension of ~2.3. The relative scaling of the energetic and entropic contributions to nanoaggregation sets the characteristic number of molecules within a nanoaggregate. The entropic penalty associated with restricting the configurational entropy of the aliphatic side chains within a nanoaggregate compared to a free monomer should be mitigated for asphaltene architectures possessing fewer side chains, leading to an elevated driving force for assembly. Indeed, simulations of asphaltene architecture A1 show there to be no characteristic size of a nanoaggregate, with a downhill thermodynamic driving force for the unbounded kinetic agglomeration of pseudo-1D stacks.


Fig. 3 – Asphaltene assembly hierarchy for molecule A2. (a) At concentrations <5% mass fraction we observe monomers and small oligomers. (b) In the range 5–10% mass fraction, nanoaggregates with a characteristic cluster size of ~8 molecules form. (c) For mass fraction range 10–15% we observe hierarchical aggregation of the nanoaggregates into clusters of nanoaggregates. (d) For mass fraction >15% we observe a gelation transition to form a viscoelastic network forming a giant connected component containing all molecules in the system. (d) The mass-averaged mean cluster size as a function of concentration illustrates the existence of two length scales, hierarchical assembly, and the gelation transition. Defining two molecules to be within a connected cluster if any of their core atoms are within a cutoff of rc = 0.55 nm characterizes the mean size of a nanoaggregate formed by pseudo-1D stacking of the aromatic cores, which is a weak function of concentration. Defining two molecules to be within the same connected cluster if any of their core or side chain atoms are within the cutoff characterizes connectivity within and between nanoaggregates, revealing the formation of clusters fro mass fractions of 5–15%, and a gelation transition at 15%, beyond which all asphaltene molecules exist in single giant connected component.

Future work. We continue to characterize the morphology and mechanisms of hierarchical assembly and compare to published experimental measurements. A publication detailing our work to date is in preparation. We intend to support our mechanistic analyses by performing machine learning of the aggregation pathways using multi-body nonlinear dimensionality reduction techniques developed in our lab. We will also study the thermodynamic stability of the various hierarchical aggregates using alchemical and expanded ensemble methods, and the kinetics of assembly through cluster lifetimes and Markov state models.