Contents
Overview
One of the greatest societal challenges of the 21st century is how to meet our societies’ growing demand for buildings, infrastructure, natural resources, and energy while reducing the associated adverse effects on the environment and climate. The research in our group will help address this grand challenge by focusing on the most critical material science problems related to industrial decarbonization, waste encapsulation, waste-to-resources, sustainable and durable infrastructure materials, and digital fabrication in construction, etc. To solve these materials science problems (e.g., uncover the atomic origin governing the reactivity of low-CO2 industrial wastes used in cement and concrete so as to enhance our ability to predict their performance in engineering applications), we combine (i) atomistic modeling (e.g., molecular dynamics simulations and density functional theory calculations) and (ii) advanced characterization (e.g., synchrotron X-ray and neutron-based scattering) with (iii) data-driven modeling.
I. Projects on Data-driven Modeling
Key words: Sustainable concrete mix design, Mineral and glass dissolution, Glass density, Random forest (RF), Artificial neural network (ANN), Gaussian process regression (GPR) and Bayesian optimization.
1.1 ML-empowered development of sustainable concrete design
Cement and concrete are the most widely used building materials in the world, and their production accounts for ~8% of global anthropogenic CO2 emissions, rendering decarbonizing cement and concrete production a critical component of global decarbonization efforts. The appealing strategies for decarbonizing cement and concrete need to not only significantly lower their climate impact but also maintain their performance, economics, and scale of production. This project aims to develop feasible concrete mixtures with lower CO2 emissions (via the use of industrial wastes) by leveraging a large industry data set and artificial intelligence (AI). We first developed various machine learning (ML) models to predict concrete properties (including 28-day strength and the evolution of strength) as a function of concrete mix proportions. We then combined these ML models with a Bayesian optimization scheme, which enables the automatic generation of mix formulations with minimal cost or climate impact, or both, while meeting target strength requirements. The optimized mixes were shown to reduce cost by ~30% and climate impact by up to 60% compared with actual concrete mixes adopted by industry. This study highlights the large potential of AI-driven concrete mix design in improving sustainability and lowering costs compared with the prescriptive and trial-and-error approaches currently adopted by the industry. This project is in collaboration with Prof. Elsa Olivetti at MIT and the IBM machine learning lab.
Ref: O.P. Pfeiffer#, K. Gong#, et al., 2024, (# co-first): paper link (valid before March 06, 2024).
1.2 ML-driven prediction of mineral and glass dissolution
Dissolution of silicate-based minerals and glasses is critical to the study of many natural and engineering processes, including the global geochemical cycle and cement and concrete production. This project aims to develop machine learning (ML) models to predict the dissolution rates of crystalline silica (i.e., quartz) and amorphous silica as a function of different dissolution conditions (e.g., temperature, pressure, pH, concentration of different metal cations, etc.). We show that random forest, artificial neural network, and Gaussian process regression models exhibit high predictive capability, with R2 values of 0.96–0.99 for testing. These prediction errors are much smaller than those of linear regression models and comparable with those achieved in previous studies using reaction models based on a smaller and less complex data set. SHapley Additive exPlanations (SHAP) analysis reveals how individual dissolution conditions impact the dissolution rates of quartz and amorphous silica. These models may enable rapid prediction or optimization of dissolution conditions for specific requirements on dissolution rates in various applications (e.g., minimizing silicate-based aggregate dissolution rates to mitigate the alkali-silica reaction).
Ref: Gong, Aytas, Zhang & Olivetti, Front. Mater., 2022 (paper link); Gong, Aytas, & Olivetti, in preparation.
1.3 ML-driven prediction of glass density
Density is one of the most commonly measured or estimated materials properties, especially for glasses and melts that are of significant interest to many fields, including metallurgy, geology, materials science, and sustainable cements. Here, an ensemble of machine learning (ML) models (i.e., random forest (RF), artificial neural network (ANN), and Gaussian process regression (GPR)) have been developed to predict the room-temperature density of glasses in the compositional space of CaO-MgO-Al2O3-SiO2-TiO2-FeO-Fe2O3-Na2O-K2O-MnO (CMASTFNKM), based on data mined from the literature. The results show that these data-driven models are much more accurate (R2 = 0.95-0.98) than existing empirical density models based on ionic packing ratio (R2 = 0.14-0.79). Interestingly, glass density is shown to be a reliable reactivity indicator for a range of CaO-Al2O3-SiO2 and volcanic glasses due to its strong correlation (R2 values above ~0.90) with the average metal-oxygen dissociation energy (a structural descriptor) of these glasses. Analysis of the predicted density-composition relationships for selected compositional subspaces suggests that the single-layer ANN model exhibits a certain level of transferability (i.e., the ability to extrapolate to compositional spaces not (or less) covered in the database) and captures known features in glasses, including the mixed alkaline earth effects.
Ref: Gong & Olivetti, JACerS (2023)
2. Projects on Atomistic Modeling
keywords: Amorphous aluminosilicates, Slag, Fly ash, Volcanic ash, Glass, Atomic structure, Reactivity, Structural descriptor, Binding energy, Molecular dynamics (MD) simulation, Density functional theory (DFT) calculation, X-ray total scattering, X-ray pair distribution function (PDF) analysis
2.1 Resolving the atomic structure of blast-furnace slag
Ground granulated blast-furnace slag (GGBFS) is an industrial byproduct from steel manufacturing that has been widely utilized to make sustainable concrete (e.g., blended cements and alkali-activated materials). This project aims to uncover the atomic structure of a GGBFS, which is a predominantly amorphous phase in the compositional space of CaO−MgO−Al2O3−SiO2 (CMAS). The highly disordered nature of the amorphous phase renders conventional lab-based X-ray diffraction experiments powerless for structural determination. In this study, we combined force-field molecular dynamics (MD) simulations and density functional theory (DFT) calculations to generate detailed structural representations for a CMAS glass relevant to GGBFS. The generated structures are not only thermodynamically favorable according to DFT calculations but also agree with our X-ray and neutron total scattering data as well as literature data. Detailed analysis of the final structures allowed us to reconcile the existing discrepancies reported in the literature. It has also revealed important structural information on the CMAS glass, including (i) the preferential role of Mg atoms as network modifiers and Ca atoms as charge compensators, and (ii) the proximity of Mg atoms to free oxygen (FO) sites. Electronic property calculations suggest higher reactivity for Ca atoms as compared with Mg atoms, and at the same time, Mg atoms seem to promote the formation of FO (compared with Ca), which is the most reactive oxygen site in CMAS. This dual effect of Mg (over Ca) may help explain the conflicting observations in the literature regarding the impact of Ca and Mg on GGBFS reactivity. Overall, this information may enhance our mechanistic understanding of the reactivity of GGBFS and related waste materials during the formation of low-CO2 cements. Furthermore, the method presented in this study is relevant to many other fields where the structure and properties of aluminosilicate glasses are of significant interest, such as condensed matter physics, glass science, geology, materials chemistry, medicine, and advanced communication systems.
2.2 Development of novel structural descriptors to predict the reactivity of amorphous aluminosilicates
Establishing the composition-structure-property relationships for amorphous aluminosilicates is critical to many important natural and engineering processes, including alkali-activated materials and blended cements. In this investigation, we employed force field-based molecular dynamics (MD) simulations to generate detailed structural representations for a range of quaternary CaO-MgO-Al2O3-SiO2 (CMAS) and ternary CaO-Al2O3-SiO2 (CAS) glasses, with compositions relevant to slags and fly ashes (i.e., byproducts from coal-fired power plants). Based on the MD-generated structural representations, two novel structural descriptors have been developed, i.e., (i) average metal oxide dissociation energy (AMODE) and (ii) average self-diffusion coefficient (ASDC) of all the atoms at melting. We showed that both structural descriptors can more accurately capture the relative glass reactivity than the degree of depolymerization parameter commonly used in the literature, especially for the eight synthetic CAS glasses that span a wide compositional range (covering slag and fly ash compositions). Hence, these descriptors hold great promise for predicting the reactivity of slag and fly ash (and other waste materials) when used to make low-CO2 cement and concrete, which is important to the optimization (and industrial adoption) of these low-CO2 concretes based on wastes.
We further extended the method to more complex volcanic glasses in the compositional space of CaO-MgO-Al2O3-SiO2-TiO2-FeO-Fe2O3-Na2O-K2O. Dissolution (or reactivity) of volcanic glasses is important to their use in blended cements and alkali-activated materials as well as the global geochemical cycle. Based on the atomic structural representations generated using MD simulations, we have introduced another novel structural descriptor, i.e., the average metal-oxygen bond strength parameter, based on the classical bond valence models. This new structural descriptor is seen to capture the relative reactivity of ten volcanic glasses reasonably well with R2 values of ~0.80-0.92 for linear regression, which is higher than that achieved using the extent of depolymerization parameter (R2 values of ~0.60) commonly used to describe glass and mineral reactivity. Our results suggest that the structural descriptors derived from MD simulations, especially the average M-O bond strength parameter, are promising structural descriptors for connecting the composition with the dissolution rates of highly complex natural glasses.
Ref: Gong & Olivetti, JACerS.; arXive preprint (PDF)
Building upon the MD simulation results and topological constraint theory, we further developed two physics-based compositional parameters to predict the relative reactivity of CaO-Al2O3-SiO2 and CaO-MgO-Al2O3-SiO2 glasses in alkaline environments: (i) a modified average metal oxide dissociation energy (AMODE) parameter; and (ii) a topology constraint parameter. Both parameters are seen to generally outperform existing compositional parameters in the literature. Importantly, the modified AMODE values are seen to strongly and inversely correlate with the isothermal conduction calorimetry (ICC) cumulative heat from the modified R3 test data (R2 value of 0.99 for linear regression), which has been widely used in the cement and concrete community to gauge the reactivity of industrial wastes (as supplementary cementitious materials (SCM)) and hence their suitability to be used for concrete production. This suggests that the modified AMODE parameter could be developed into a reliable fast-screen method for SCM reactivity. This is potentially significant, as the modified AMODE can be directly calculated from the chemical compositions (i.e., no need for MD simulations and ICC experiments).
Ref: Gong & White, CCR.
2.3 Binding energy calculations using density functional theory
Interactions between positively charged metal cations and negatively charged aluminosilicate species are critical to many important engineering processes and applications, including sustainable cements (e.g., alkali-activated materials) and aluminosilicate glasses. Here, we probe these interactions by calculating the binding energies between an aluminosilicate dimer/trimer and 17 different metal cations Mn+ (Mn+ = Li+, Na+, K+, Cu+, Cu2+, Co2+, Zn2+, Ni2+, Mg2+, Ca2+, Ti2+, Fe2+, Fe3+, Co3+, Cr3+, Ti4+ and Cr6+) using density functional theory (DFT) calculations. We show that the DFT-derived binding energies vary considerably depending on the type of cations (i.e., charge and ionic radii) and aluminosilicate species (i.e., dimer or trimer). Analysis reveals that the calculated binding energies can be used to explain many literature observations regarding the impact of metal cations on materials properties (e.g., ionic transport, glass corrosion, and mineral dissolution). We also show that the binding energy for a given dimer/trimer can well approximated by a 2nd-order polynomial function of the ionic potential or field strengths of the metal cations. Given that we can readily estimate the ionic potential and field strength of any given metal cations using well-tabulated ionic radii available in the literature, these simple polynomial functions would enable rapid calculation of the binding energies for a much wider range of cations, providing guidance on the design and optimization of sustainable cements and aluminosilicate glasses and their associated applications.
3. Projects on Alkali-Activated Materials (AAMs)
Keywords: Reaction kinetics, Formation mechanisms, Sulfate attack mechanisms, In situ X-ray PDF analysis, MD simulation, In situ quasi-elastic neutron scattering (QENS), Isothermal conduction calorimetry (ICC), and Fourier transform infrared spectroscopy (FTIR) analysis
Alkali-activated materials (AAMs) are a class of low-CO2 cementitious materials made from industrial byproducts or calcined clays (slag, fly ash, biomass ashes, waste glasses, metakaolin, etc.). In addition to being used as low-CO2 construction materials, AAMs have also attracted attention for many other applications, including waste encapulation or immobilization, cellular ceramic foam (as a solar receiver), wastewater treatment, etc.
3.1 Reaction kinetics and formation mechanisms of AAMs
Both the formation process of AAMs and conventional Portland cement (PC) consists of the dissolution of precursor particles and the precipitation of reaction products, which occur concurrently. This formation process not only dictates the early-age properties of AAMs and PC (e.g., workability, setting, and hardening) and subsequent development of strength but also controls their pore structure evolution and long-term durability. However, the exact mechanisms occurring during the formation of AAMs remain somewhat unknown. One challenge that limits our ability to elucidate the exact formation mechanisms is the need for experimental tools that allow us to (i) probe these reactions in situ at a high spatial and temporal resolution and in a non-destructive manner, and (ii) study reactions involving amorphous phases.
In situ quasi-elastic neutron scattering
This project aims to study the reaction mechanisms occurring during the formation of alkali-activated slags (AASs) by probing the evolution of water dynamics in AAS using an in situ quasi-elastic neutron scattering (QENS) technique together with isothermal conduction calorimetry (ICC), FTIR, and NAD. We show that the single ICC reaction peak in the NaOH-activated slag is accompanied by a conversion of free water to bound water (from QENS analysis), indicating the formation of a sodium-containing calcium–alumino-silicate–hydrate (C–(N)–A–S–H) gel. In contrast, we see two distinct ICC reaction peaks for the Na2SiO3-activated slag sample. The first ICC peak is accompanied by the transformation of constrained water to bound and free water while the second ICC peak is associated with the conversion of free water to bound and constrained water (from QENS analysis). The first reaction peak is attributed to the formation of an initial gel that is controlled by the Na+ ions and silicate species in Na2SiO3 solution and the dissolved Ca/Al species from slag, while the second conversion is attributed to the formation of the main reaction product (i.e., C–(N)–A–S–H gel). Hence, this study demonstrates that the in situ QENS technique, when combined with lab-based characterization tools, is powerful in elucidating the evolution of water dynamics and formation mechanisms occurring in complex materials (e.g., alkali-activated slags).
Reference: Gong et al. PCCP (2019)
In situ X-ray total scattering combined with MD simulations
Chemical reactions involving amorphous/disordered phases are ubiquitous in many important natural and engineering processes/applications, including alkali-activated materials (AAMs). However, it’s challenging to quantify reactions involving the formation or transformation of amorphous phases using lab-based tools. Here, we outlined a powerful approach that combines atomistic modeling with in situ pair distribution function (PDF) analysis based on synchrotron X-ray total scattering (with high spatial and temporal resolution) to quantify the amorphous-to-disordered transformation that occurs in hydroxide-activated ground granulated blast-furnace slag (GGBS). We first generate a detailed structural representation for the amorphous GGBS by using force-field molecular dynamics (MD) simulations, which is seen to agree with our X-ray total scattering data. Based on this structural representation for GGBS and those for the reaction products (from the literature), we performed real-space X-ray PDF refinement of the alkali activation of GGBS. This results in the quantification of all phases (including both amorphous and crystalline phases) as a function of reaction time, enabling the reaction kinetics and mechanisms to be studied. This work highlights the power of combining in situ PDF analysis with atomistic modeling to study complex chemical reactions involving amorphous-to-disordered/crystalline transformations.
Reference: Gong & White, CCR (2022), arXive preprint (PDF); Gong et al. CCR (2016)
3.2 Durability of AAMs
Nanoscale chemical degradation mechanisms of sulfate attack in alkali-activated slag (AAS)
Chemically induced material degradation constitutes a major durability issue for many technologically important material systems. Sulfate attack is a major durability issue for concrete materials and structures, especially in coastal areas where the seawater is abundant with MgSO4 and Na2SO4. Both sulfate-bearing chemicals can cause severe degradation in cement and concrete. Furthermore, the microbial-induced formation of H2SO4 in sewer pipelines is known to cause severe degradation of the concrete pipes. This project aims to uncover the nanoscale chemical degradation mechanisms for a NaOH-activated slag paste exposed to different types of sulfate-bearing solutions (i.e., Na2SO4, MgSO4, and H2SO4), by combining synchrotron-based X-ray diffraction (XRD), X-ray pair distribution function (PDF) analysis, and Fourier transform infrared (FTIR) spectroscopy. Analysis of the chemistry and structure shows that the AAS paste is essentially immune to Na2SO4 attack, which is destructive to conventional Portland cement systems. However, exposure to 5–10 wt % MgSO4 and H2SO4 completely disintegrates the main strength-giving binder gel (i.e., sodium-containing calcium–(alumino)–silicate–hydrate). These differences are seen to be correlated with the ability of the ions (i.e., Na+, Mg2+, H+) in the sulfate-bearing chemicals to regulate the pH of the pore solution in AAS. This study has provided important mechanistic insight on the chemical degradation mechanisms in AAS under sulfate attack.
Reference: Gong et al. JPCC (2018)
Tailoring slag chemistry to achieve superior resistance to sulfate attack
As AAS is prone to MgSO4 attack, similar to OPC-based materials, the goal of this study is to explore the possibility of fundamentally enhancing the resistance of AAS to MgSO4 attack by tailoring the chemistry of AASs. We employed synchrotron-based X-ray diffraction and pair distribution function analysis to evaluate the impact of MgSO4 exposure on the chemistry and local atomic structure of different AASs. We discovered that, by switching from Ca-rich slags to an Fe-rich slag, the resulting AAS is essentially immune to MgSO4 attack. These results clearly demonstrate the immunity certain AASs possess to sulfate attack-induced (i.e., Na2SO4 and MgSO4) durability issues that plague numerous concrete structures around the world.
Ref: unpublished data (Poster).
4. 2-D materials and cementitious nanocomposites
Keywords: CO2 capture, Portlandene, Graphene, Graphene oxide reinforced cement, Rheological properties
Highly Surface-Active Ca(OH)2 Monolayer as a CO2 Capture Material
Designing novel materials that can efficiently capture CO2 can play a crucial role in the global decarbonization efforts. Here, we theoretically demonstrated that dimensional reduction of bulk crystalline portlandite yields a stable monolayer substance (called portlandene) that is highly effective at capturing CO2. We evaluated the robustness and stability of this single-layer phase by using force-field molecular dynamics simulations and ab initio quantum mechanical calculations. We further showed that portlandene’s chemical activity can be increase by introducing surface flaws (i.e., vacancy sites). Despite being inert to water vapor, portlandene with defects can separate CO and CO2 from a syngas (CO/CO2/H2) stream. In order to understand this selective behavior and the underlying mechanisms, we have studied portlandene’s electrical structure, local charge distribution, and bonding orbitals. Moreover, the regeneration of CO2 captured by portlandene can be released by application of a mild external electric field, which is different from other CO2 capturing technologies that often require high temperature heat treatment.
Graphene oxide (GO) offers great potential as nanoscale reinforcement for cementitious material. In this project, we investigated the reinforcing effects of graphene oxide (GO) on Portland cement paste. We found that the addition of 0.03 wt.% GO sheets into the cement paste can increase the compressive strength and tensile strength of the cement composite by more than 40%. This enhancement of strength is attributed to (i) the seeding effect of GO which leads to finer pore structure and (ii) nano-scale reinforcing effect of GO fibers. However, the addition of 0.03 wt.% GO result in significant reduction in the workability of the GO-cement composite. The rheological properties (i.e., viscosity and yield strength) of cement paste are greatly increased, and amount of increment is seen to depend on the size of the GO sheet. The overall results indicate that GO could be a promising nanofillers for reinforcing the engineering properties of portland cement paste.
Refs: Gong et al, J. Mater. Civil Eng. (2015); Gong et al., ACUN6 proceedings (2012)
