Search

This article presents an in-depth analysis and evaluation of artificial neural networks (ANNs) when applied to replicate trajectories in molecular dynamics (MD) simulations or other particle methods. This study focuses on several architectures—feedforward neural networks (FNNs), convolutional neural networks (CNNs), recurrent neural networks (RNNs), time convolutions (TCs), self-attention (SA), graph neural networks (GNNs), neural ordinary differential equation (ODENets), and an example of physics-informed machine learning (PIML) model—assessing their effectiveness and limitations in understanding and replicating the underlying physics of particle systems. Through this analysis, this paper introduces a comprehensive set of criteria designed to evaluate the capability of ANNs in this context. These criteria include the minimization of losses, the permutability of particle indices, the ability to predict trajectories recursively, the conservation of particles, the model’s handling of boundary conditions, and its scalability. Each network type is systematically examined to determine its strengths and weaknesses in adhering to these criteria. While, predictably, none of the networks fully meets all criteria, this study extends beyond the simple conclusion that only by integrating physics-based models into ANNs is it possible to fully replicate complex particle trajectories. Instead, it probes and delineates the extent to which various neural networks can “understand” and interpret aspects of the underlying physics, with each criterion targeting a distinct aspect of this understanding.

This micro-article introduces a method for integrating Large Language Models with geometry/mesh generation software and multiphysics solvers, aimed at streamlining physics-based simulations. Users provide simulation descriptions in natural language, which the language model processes for geometry/mesh generation and physical model definition. Initial results demonstrate the feasibility of this approach, suggesting a future where non-experts can conduct advanced multiphysics simulations by simply describing their needs in natural language, while the code autonomously handles complex tasks like geometry building, meshing, and setting boundary conditions.

Deep vein thrombosis is a life-threatening disease that takes millions of people's lives worldwide. Given both technical and ethical issues of using animals in research, it is necessary to develop an appropriate in vitro model that would recapitulate the conditions of venous thrombus development. We present here a novel microfluidics vein-on-a-chip with moving valve leaflets to mimic the hydrodynamics in a vein, and Human Umbilical Vein Endothelial Cell (HUVEC) monolayer. A pulsatile flow pattern, typical for veins, was used in the experiments. Unstimulated human platelets, reconstituted with the whole blood, accumulated at the luminal side of the leaflet tips proportionally to the leaflet flexibility. Platelet activation by thrombin induced robust platelet accrual at the leaflet tips. Inhibition of glycoprotein (GP) IIb-IIIa did not decrease but, paradoxically, slightly increased platelet accumulation. In contrast, blockade of the interaction between platelet GPIbα and A1 domain of von Willebrand factor completely abolished platelet deposition. Stimulation of the endothelium with histamine, a known secretagogue of Weibel-Palade bodies, promoted platelet accrual at the basal side of the leaflets, where human thrombi are usually observed. Thus, platelet deposition depends on the leaflet flexibility, and accumulation of activated platelets at the valve leaflets is mediated by GPIbα-VWF interaction.

Deep vein thrombosis is the clotting of blood in deep veins. Here, a microfluidic device containing flexible valves fabricated in-situ is used to investigate the effects of blood flow conditions and valve elasticity on thrombus formation, revealing the circumstance under which clotting occurs.

The longevity of asphalt pavements is a key focus of road engineering, which closely relates to the self-healing ability of bitumen. Our work aims to establish a CGMD model and matched force field for bitumen and break through the limitations of the research scale to further explore the microscopic mechanism of bitumen self-healing. In this study, a CGMD mapping scheme containing 16 kinds of beads is proposed, and the non-bond potential energy function and bond potential energy function are calculated based on all-atom simulation to construct and validate a coarse-grained model for bitumen. On this basis, a micro-crack model with a width of 36.6nm is simulated, and the variation laws of potential energy, density, diffusion coefficient, relative concentration and temperature in the process of bitumen self-healing are analyzed with the cracking rate parameter proposed to characterize the degree of bitumen crack healing. The results show that the computational size of the coarse-grained simulation is much larger than that of the all-atom, which can explain the self-healing mechanism at the molecular level. In the self-healing process, non-bonded interactions dominate the molecular movement, and differences in the decreased rate of diffusion among the components indicate that saturates and aromatics play a major role in self-healing. Meanwhile, the variations in crack rates reveal that healing time is inversely proportional to temperature. The impact of increasing temperature on reducing healing time is most obvious when the temperature approaches the glass transition temperature (300 K).

In the original publication [...]

In this work, a combined smoothed particle hydrodynamics and discrete element method (SPH-DEM) model was proposed to model particle agglomeration in a shear flow. The fluid was modeled with the SPH method and the solid particles with DEM. The system was governed by three fundamental dimensionless groups: the Reynolds number Re (1.5~150), which measured the effect of the hydrodynamics; the adhesion number Ad (6 × 10−5~6 × 10−3), which measured the inter-particle attraction; and the solid fraction α, which measured the concentration of particles. Based on these three dimensionless groups, several agglomeration regimes were found. Within these regimes, the aggregates could have different sizes and shapes that went from long thread-like structures to compact spheroids. The effect of the particle–particle interaction model was also investigated. The results were combined into ‘agglomeration maps’ that allowed for a quick determination of the agglomerate type once α, Re, Ad were known.

The performance of solid oral dosage forms targeting the colon is typically evaluated using standardised pharmacopeial dissolution apparatuses. However, these fail to replicate colonic hydrodynamics. This study develops a digital twin of the Dynamic Colon Model; a physiologically representative in vitro model of the human proximal colon. Magnetic resonance imaging of the Dynamic Colon Model verified that the digital twin robustly replicated flow patterns under different physiological conditions (media viscosity, volume, and peristaltic wave speed). During local contractile activity, antegrade flows of 0.06–0.78 cm s−1 and backflows of −2.16–−0.21 cm s−1 were measured. Mean wall shear rates were strongly time and viscosity dependent although peaks were measured between 3.05–10.12 s−1 and 5.11–20.34 s−1 in the Dynamic Colon Model and its digital twin respectively, comparable to previous estimates of the USPII with paddle speeds of 25 and 50 rpm. It is recommended that viscosity and shear rates are considered when designing future dissolution test methodologies for colon-targeted formulations. In the USPII, paddle speeds >50 rpm may not recreate physiologically relevant shear rates. These findings demonstrate how the combination of biorelevant in vitro and in silico models can provide new insights for dissolution testing beyond established pharmacopeial methods.

In this work the first 3D Smoothed Particle Hydrodynamics model of a Rayleigh collapse for an empty cavity is proposed with the aim of improving the hydrodynamic analysis of a non-symmetrical collapse. The hydrodynamics of the model is validated against the solution of the Rayleigh-Plesset equation for a symmetrical collapse. The model is then used to simulate a non-symmetrical collapse of an empty cavity attached to a solid surface with γ=0.6 induced by an external pressure of 50 [MPa]. The results shows that is possible to identify three regions where the hydrodynamics of the collapsing cavity shows different features. For all the stages of the collapse the simulation shows smooth pressure and velocity fields in the liquid and in the solid phase with the formation of a vortex ring in the final phase of the collapse. Finally, the model is compared to a previous 2D model to highlight strong, weak points and the key differences of both approaches in final phase of the collapse.

Stenting is a common method for treating atherosclerosis. A metal or polymer stent is deployed to open the stenosed artery or vein. After the stent is deployed, the blood flow dynamics influence the mechanics by compressing and expanding the structure. If the stent does not respond properly to the resulting stress, vascular wall injury or re-stenosis can occur. In this work, a Discrete Multiphysics modelling approach is used to study the mechanical deformation of the coronary stent and its relationship with the blood flow dynamics. The major parameters responsible for deforming the stent are sorted in terms of dimensionless numbers and a relationship between the elastic forces in the stent and pressure forces in the fluid is established. The blood flow and the stiffness of the stent material contribute significantly to the stent deformation and affect its rate of deformation. The stress distribution in the stent is not uniform with the higher stresses occurring at the nodes of the structure. From the relationship (correlation) between the elastic force and the pressure force, depending on the type of material used for the stent, the model can be used to predict whether the stent is at risk of fracture or not after deployment.

A simplified modelling technique for modelling viscoelastic fluids is proposed from the perspective of Discrete Multiphysics. This approach, based on the concept of linear additive composition of energy potentials, aims to integrate Smooth Particle Hydrodynamics (SPH) with an equivalent elastic potential tailored for fluid flow simulations. The model was implemented using a particle-based software, explored and thoroughly validated with results from numerical experiments on three different flow conditions. The model was able to successfully capture a large extent of viscoelastic responses to external forcing, ranging from pure viscous flows to creep-dominated Bingham type of behaviour. It is concluded that, thanks to the modularity and tunable characteristics of the parameters involved, the proposed modelling approach can be a powerful simulation tool for modelling or mimicking the behaviour of viscoelastic substances.

In this study, we propose a mesh-free (particle-based) Smoothed Particle Hydrodynamics model for simulating a Rayleigh collapse. Both empty and gas cavities are investigates and the role of heat diffusion is also accounted for. The system behaves very differently according to the ratio between the characteristic time of collapse and the characteristic time of thermal diffusion. This study identifies five different possible behaviours that range from isothermal to adiabatic.

A comprehensive review is carried out on the models and correlations for solid/fluid reactions that result from a complex multi-scale physicochemical process. A simulation of this process with CFD requires various complicated submodels and significant computational time, which often makes it undesirable and impractical in many industrial activities requiring a quick solution within a limited time frame, such as new product/process design, feasibility studies, and the evaluation or optimization of the existing processes, etc. In these circumstances, the existing models and correlations developed in the last few decades are of significant relevance and become a useful simulation tool. However, despite the increasing research interests in this area in the last thirty years, there is no comprehensive review available. This paper is thus motivated to review the models developed so far, as well as provide the selection guidance for model and correlations for the specific application to help engineers and researchers choose the most appropriate model for feasible solutions. Therefore, this review is also of practical relevance to professionals who need to perform engineering design or simulation work. The areas needing further development in solid–fluid reaction modelling are also identified and discussed.

This work proposes a model of particle agglomeration in elastic valves replicating the geometry and the fluid dynamics of a venous valve. The fluid dynamics is simulated with Smooth Particle Hydrodynamics, the elastic leaflets of the valve with the Lattice Spring Model, while agglomeration is modelled with a 4-2 Lennard-Jones potential. All the models are combined together within a single Discrete Multiphysics framework. The results show that particle agglomeration occurs near the leaflets, supporting the hypothesis, proposed in previous experimental work, that clot formation in deep venous thrombosis is driven by the fluid dynamics in the valve.

For colonic drug delivery, the ascending part of the colon is the most favourable site as it offers the most suitable environmental conditions for drug dissolution. Commonly, the performance of a drug formulation is assessed using standardised dissolution apparatus, which does not replicate the hydrodynamics and shear stress evoked by wall motion in the colon. In this work, computer simulations are used to analyse and understand the influence of different biorelevant motility patterns on the disintegration/drug release of a solid dosage form (tablet) under different fluid conditions (viscosities) to mimic the ascending colonic environment. Furthermore, the ability of the motility pattern to distribute the drug in the ascending colon luminal environment is analysed to provide data for a spatiotemporal concentration profile. The motility patterns used are derived from in vivo data representing different motility patterns in the human ascending colon. The applied motility patterns show considerable differences in the drug release rate from the tablet, as well as in the ability to distribute the drug along the colon. The drug dissolution/disintegration process from a solid dosage form is primarily influenced by the hydrodynamic and shear stress it experiences, i.e., a combination of motility pattern and fluid viscosity. Reduced fluid motion leads to a more pronounced influence of diffusion in the tablet dissolution process. The motility pattern that provoked frequent single shear stress peaks seemed to be more effective in achieving a higher drug release rate. The ability to simulate drug release profiles under biorelevant colonic environmental conditions provides valuable feedback to better understand the drug formulation and how this can be optimised to ensure that the drug is present in the desired concentration within the ascending colon.

LAMMPS is a powerful simulator originally developed for molecular dynamics that, today, also accounts for other particle-based algorithms such as DEM, SPH, or Peridynamics. The versatility of this software is further enhanced by the fact that it is open-source and modifiable by users. This property suits particularly well Discrete Multiphysics and hybrid models that combine multiple particle methods in the same simulation. Modifying LAMMPS can be challenging for researchers with little coding experience. The available material explaining how to modify LAMMPS is either too basic or too advanced for the average researcher. In this work, we provide several examples, with increasing level of complexity, suitable for researchers and practitioners in physics and engineering, who are familiar with coding without been experts. For each feature, step by step instructions for implementing them in LAMMPS are shown to allow researchers to easily follow the procedure and compile a new version of the code. The aim is to fill a gap in the literature with particular reference to the scientific community that uses particle methods for (discrete) multiphysics.

In this paper, we present a methodological study of modelling red blood cells (RBCs) in shear-induced flows based on the discrete multiphysics (DMP) approach. The DMP is an alternative approach from traditional multiphysics based on meshless particle-based methods. The proposed technique has been successful in modelling multiphysics and multi-phase problems with large interfacial deformations such as those in biological systems. In this study, we present the proposed method and introduce an accurate geometrical representation of the RBC. The results were validated against available data in the literature. We further illustrate that the proposed method is capable of modelling the rupture of the RBC membrane with minimum computational difficulty.

The non-symmetrical collapse of an empty cylindrical cavity is modeled using Smoothed Particle Hydrodynamics. The presence of a nearby surface produces an anisotropic pressure field generating a high-velocity jet that hits the surface. The collapse follows a different dynamic based on the initial distance between the center of the cavity and the surface. When the distance is greater than the cavity radius (detached cavity) the surface is hit by traveling shock waves. When the distance is less than the cavity radius (attached cavity) the surface is directly hit by the jet and later by other shock waves generated in the last stages of the of the collapse. The results show that the surface is hit by a stronger shock when distance between the center of the cavity and the surface is zero while showing more complex double peaks behavior for other distances.

This study proposes a 3D particle-based (discrete) multiphysics approach for modelling calcification in the aortic valve. Different stages of calcification (from mild to severe) were simulated, and their effects on the cardiac output were assessed. The cardiac flow rate decreases with the level of calcification. In particular, there is a critical level of calcification below which the flow rate decreases dramatically. Mechanical stress on the membrane is also calculated. The results show that, as calcification progresses, spots of high mechanical stress appear. Firstly, they concentrate in the regions connecting two leaflets; when severe calcification is reached, then they extend to the area at the basis of the valve.

Viscoelastic bonds intended for Discrete Multiphysics (DMP) models are developed to allow the study of viscoelastic particles with arbitrary shape and mechanical inhomogeneity that are relevant to the pharmaceutical sector and that have not been addressed by the Discrete Element Method (DEM). The model is applied to encapsulate particles with a soft outer shell due, for example, to the partial ingress of moisture. This was validated by the simulation of spherical homogeneous linear elastic and viscoelastic particles. The method is based on forming a particle from an assembly of beads connected by springs or springs and dashpots that allow the sub-surface stress fields to be computed, and hence an accurate description of the gross deformation. It is computationally more expensive than DEM, but could be used to define more effective interaction laws.

Positron emission particle tracking (PEPT) was used to investigate the grinding media dynamics in a laboratory-scale attritor mill in the absence of powder. The grinding media motion was analysed as a function of the equipment’s typical operating parameters: impeller speed, impeller clearance and bead fill level. It was observed that the impeller speed had the strongest influence on the media motion. An increase of the impeller speed from 300 rpm to 600 rpm led to a change in the bead recirculation patterns with the increasing formation of well segregated upper and lower recirculation loops that fully developed at the maximum speed of 600 rpm. For a constant impeller speed, an increase of the bead loading did not majorly affect the bead velocity as remarked by minor changes on the flow field. For all the impeller clearance values, the occupancy plots revealed an inefficient dead region at the bottom of the attritor where the beads were moving at very low velocity. In this region the beads were tightly packed under their own weight and, furthermore, there was an absence of direct contact with the impeller arms. The depth of this region increased proportionally to the distance between the bottom of the impeller and the vessel base indicating that a minimum value of clearance should be set to optimise the lower recirculation pattern. For two experimental conditions, the data generated by PEPT measurements were utilised to set-up a friction-adjusted discrete element method (DEM) model. Here, the simulation results were qualitatively and quantitatively compared against the PEPT data by assessing the averaged velocity flow fields and the average velocity profiles at different radial locations inside the vessel. Given the intrinsic uncertainty of the PEPT measurements, the DEM model results were in considerably good agreement with the experimental results. The major discrepancy was observed close to the vessel wall where the simulations overpredicted the velocity by about 10%.

We propose a mesh-free and discrete (particle-based) multi-physics approach for modelling the hydrodynamics in flexible biological valves. In the first part of this study, the method is successfully validated against both traditional modelling techniques and experimental data. In the second part, it is further developed to account for the formation of solid aggregates in the flow and at the membrane surface. Simulations of various types of aggregates highlight the main benefits of discrete multi-physics and indicate the potential of this approach for coupling the hydrodynamics with phenomena such as clotting and calcification in biological valves.

This study proposes a model based on the combination of Smoothed Particle Hydrodynamics, Coarse Grained Molecular Dynamics and the Discrete Element Method for the simulation of dispersed solid-liquid flows. The model can deal with a large variety of particle types (non-spherical, elastic, breakable, melting, solidifying, swelling), flow conditions (confined, free-surface, microscopic), and scales (from microns to meters). Various examples, ranging from biological fluids to lava flows, are simulated and discussed. In all cases, the model captures the most important features of the flow.