Investigation On The Methodologies For Predicting Thermodynamic Properties Of Industrial Fluids Using Molecular Dynamics Simulation

Industrial fluids consisting of molecular liquids,ionic liquids and solutions play significant roles in soft material and chemical engineering.Knowledge of the thermodynamic properties and phase behavior of these fluids is essential for chemical process design and the development of new materials.However,the available thermodynamic data in open literature or databases is rather limited.Experimental measurements at different physical states are often expensive and not environmentally-friendly.In the past decades,considerable efforts have been taken in the development of molecular simulation techniques.Molecular simulation has been broadly applied in various fields like physical biology,drug discovery,materials design,energy and so on.By utilizing statistical mechanics,molecular dynamics simulation is able to predict macroscopic thermodynamic properties of fluid from the microscopic motions of atoms.Qualitative predictions have been reached for limited systems in the literature.However,these studies have focused on a specific class of molecules,owing to the lack of a consistent force field and simulation protocol.The accuracy of prediction for other categories of molecules and properties are not guaranteed.The challenge of qualitative prediction by molecular simulation lies in two aspects:the accuracy of the underlying force field and the applicability of simulation protocol.The molecular force field is a simplified description of the true potential energy surface of the system.Correct capture of the basic physics by the force field ensures the transferability to other molecular species and physical states.The simulation protocol determines the efficiency and convergence of simulations.The purpose of this thesis is to provide basic thermodynamic data for chemical engineering design by molecular simulations.The focus is on how to solve the aforementioned two problems.The main works in this thesis are:Developing force field with enhanced temperature and pressure transferability;Optimizing an general simulation protocol that is applicable and efficient;Developing high-throughput simulation framework based on our force field and simulation protocol,and performing prediction with this framework;Building a web platform to share these predicted data;Developing artificial neural network based on simulation data to enlarge prediction.The accuracy of force fields is key to the successful prediction of the thermodynamic properties of materials.In simulations of organic molecules over large temperature ranges,atomistic force fields that are parameterized at,or near,ambient temperatures are found to systematically underestimate the intermolecular dispersion interactions at elevated temperatures.Analysis of the underestimates using diatomic molecules indicates that a minor part is due to the change in molecular polarizability,while the major part is due to the reduced dielectric constant of the bulk liquid as the density decreases with increasing temperature.By writing the dispersion parameter as a linear function of temperature,we have successfully enhanced the temperature transferability of atomistic force fields.This approach is tested on 66 molecular liquids covering four functional groups-alkane,aromatic,ether,and ketone-aldehyde–over a broad range of temperatures by calculating liquid density,the heat of vaporization,isobaric heat capacity and shear viscosity.The transferability on pressure is equally important.In this work,we test the ability of the force field in predicting high-pressure viscosities of branched alkane.Using four alkane molecules as a test,the force field developed based on equilibrium properties at ambient pressures is found to be inefficient for predicting viscosity at elevated pressures.It is evident that the spherical model used for representing the van der Waals interaction in the force field needs to be corrected for predicting transport properties at high pressures.We make the correction by contracting the C-H bond length to 0.8?.The prediction accuracy of high-pressure viscosity is significantly enhanced without sacrificing thermodynamics properties.The exponent of repulsion in Lennard-Jones potential function also has a significant effect on the transport properties of liquids.Ionic liquids play important roles in industrial liquids and have potential applications in solvation,absorption,catalysis,electrochemistry and so on.The TEAM force field database is extended to cover room-temperature ionic liquids.The training set for parameterization includes five classes of cations and 12 anions.The parameters are primarily derived from quantum mechanical data,with a few non-bonded parameters optimized using experimental data of density and viscosity.The extrapolation of shear viscosity calculated using non-equilibrium periodic perturbation method is examined by analyzing the Newtonian region in the calculated shear viscosities.With the identified extrapolation scheme,the PP method is about five times more efficient than the multi-configuration-ensemble Green-Kubo method to reach similar convergence.The force field is validated on density,viscosity,and isobaric heat capacity.For more complicated electrolyte solution,the applicability of atomistic simulation is limited by the time and space scale.To facilitate the investigation of phenomena related to divalent ions that occur over large length-and time-scales,a coarse-grained force field?CGFF?is developed for MgCl₂and CaCl₂aqueous solutions.The ions are modeled by CG beads with characteristics of hydration shells.To accurately describe the non-ideal behavior of the solutions,osmotic coefficients in a wide range of concentration were used as guidance for parameterization.The resulting force field is applied in the simulation of anionic surfactants in CaCl₂solution.It is found that calcium ion significantly promotes the aggregation of anionic surfactant micelles.Different surfactant prefers different aggregation mechanism.Combined with the above force fields,we have developed a high-throughput force field simulation?HT-FFS?framework to automatically calculate thermodynamic properties for a large number of molecules.This procedure is applied to calculate liquid densities,heats of vaporization,heat capacities,vapor-liquid-equilibrium curves,critical temperatures,critical densities and surface tensions for a wide range of alkanes.The predictions agree well with available experimental data in terms of accuracy and precision,demonstrating that HT-FFS is a valid approach to supplementing experimental measurements.Above works on force field and HT-FFS were extended to over 10000 molecules containing seven elements:C,H,O,N,F,Cl,and Br.The large amounts of data generated should be accessible to the public.Therefore,we developed a database and an online platform.The first function of this platform is to provide computational data.The data can be searched using different queries and downloaded.The second function is to download the molecular model and force field files so that the computation can be repeated and extended.The third function is to collect feedbacks and suggestion which may be comments and suggestions or new computational and experimental data.Furthermore,the large amounts of data generated by HT-FFS lay a foundation for machine learning.We have developed an artificial neural network for alkanes relating the thermodynamic properties to molecular structures.The results demonstrate the feasibility of expanding predictions beyond simulation using a machine learning model.