Using industrial waste in controlled low-strength material (CLSM) offers many benefits and addresses waste management issues. However, identifying optimal CLSM design while balancing environmental impacts and engineering properties poses a notable challenge. Herein, new alkali-activated CLSM formulations based on urban waste glass and red mud (RM) are proposed for the first time with systematic investigations into the material properties and sustainability. The formulations involve the ternary blends of slag, glass powder (GP) and RM as the binders and crushed glass as aggregate. Our observations show RM has a clay-like texture with complex crystalline phases rich in Al and Fe. Increasing the proportion of RM to GP notably decreases the flowability, extends the setting time and impairs the mechanical properties. Nevertheless, red mud demonstrates a crucial role of reducing bleeding level especially for the mixtures with high proportions of glass aggregate. The standard leachate tests as per toxicity characteristic leaching procedure (TCLP) show the heavy metals leached from the CLSM are below the concentration limits and exhibit low mobility. The microstructural and thermodynamic analyses reveal the precipitation of calcium-(sodium-)aluminosilicate hydrate (C-(N-)A-S-H) as the major binding material. The reactive Fe and Al in red mud most probably contribute to the formation of hydrogarnet phases. Moreover, the suggested CLSM formations typically exhibit reduced carbon emissions and lower costs compared to the traditional CLSM produced with cement and fly ash, underlining their environmental sustainability and cost-effectiveness. This study provides a potential guideline for the flowable fill construction in the regions with bauxite producers.

Auxetic cementitious cellular composites (ACCCs) exhibit desirable mechanical properties (e.g., high fracture resistance and energy dissipation), due to their unique deformation characteristics. In this study, a new type of cementitious auxetic material, referred to as peanut shaped ACCC, has been designed and subsequently architected using additive manufacturing techniques. Two peanut shaped ACCCs specimens with different pseudo-minor axes have been tested under uniaxial compression with Digital Image Correlation (DIC) to assess their compressive behavior, peak strength, Poisson's ratio, and energy dissipation capacity. Additionally, cyclic tests were conducted to investigate their compressive resilience properties, further elucidated through microstructural analysis using a digital optical microscope. The mechanical test results were also compared with those of previously developed elliptical-shaped ACCCs. Furthermore, a numerical model was used to simulate the mechanical behavior of peanut shaped ACCCs under uniaxial compression, and showed a good agreement with the experimental data. The auxetic behavior observed in peanut shaped ACCCs arises from the rotation of sections facilitated by fiber bridging at the ligament of adjacent holes within the cementitious unit cell. In comparison to elliptical-shaped ACCCs, peanut shaped ACCCs can exhibit a slightly more negative Poisson's ratio and mitigate stress concentration. The reduction of stress concentration enables peanut shaped ACCCs to dissipate substantial energy, showcasing enhanced ductility and toughness. In cyclic tests, peanut shaped ACCCs exhibit superior recoverable deformation elasticity, attributed to robust fiber bridging capacity. The exceptional mechanical properties exhibited by peanut shaped ACCCs offer a scalable solution for developing energy-absorbent and multifunctional cementitious materials for smart infrastructure.

This paper presents a state-of-the-art review on the application of additive manufacturing (AM) in self-healing cementitious materials. AM has been utilized in self-healing cementitious materials in three ways: (1) concrete with 3D-printed capsules/vasculatures; (2) 3D concrete printing (3DCP) with fibers or supplementary cementitious materials (SCMs); and (3) a combination of (1) and (2). 3D-printed capsules/vascular systems are the most extensively investigated, which are capable of housing larger volumes of healing agents. However, due to the dimension restraints of printers, most of the printed vasculatures/capsules are in small scale, making them difficult for upscaling. Meanwhile, 3DCP shows great potential to lower the environmental footprint of concrete construction. Incorporation of fibers and SCMs helps improve the autogenous healing performance of 3DCP. Besides, 3D-printed concrete with hollow channels as the vasculature could further improve the autonomous healing and scalability of self-healing cementitious materials. Finally, possible directions for future research are discussed.

Vascular self-healing concrete (SHC) has great potential to mitigate the environmental impact of the construction industry by increasing the durability of structures. Designing concrete with high initial mechanical properties by searching a specific arrangement of vascular structure is of great importance. Herein, an automatic optimization method is proposed to arrange vascular configuration for minimizing the adverse influence of vascular system through a reinforcement learning (RL) approach. A case study is carried out to optimize a concrete beam with 3 pores (representing a vascular network) positioned in the beam midspan within a design space of 40 possibilities. The optimization is performed by the interaction between RL agent and Abaqus simulation environment with the change of target properties as a reward signal. The results illustrates that the RL approach is able to automatically enhance the vascular arrangement of SHC given the fact that the 3-pore structures that have the maximum target mechanical property (i.e., peak load or fracture energy) are accessed for all of the independent runs. The RL optimization method is capable of identifying the structure with high fracture energy in the new optimization task for 4-pore concrete structure.

Direct ink writing of cementitious materials can be an alternative way for creating vascular self-healing concrete by intentionally incorporating hollow channels in the cementitious matrix. In this study, a 3D-printable fibre reinforced mortar was first developed. Three groups of specimens were fabricated using direct ink writing, where the two top and bottom printing layers were printed with different printing directions. The macrostructure of the hardened specimens was studied using CT scanning. Four-point bending tests were carried out to investigate the initial flexural strength and the strength recovery after healing with injected epoxy resin. Furthermore, water permeability test was used to evaluate the healing potential of the samples. The results from CT scanning show that printing direction influences the actual volumes of hollow channels and the volume of small pores which are a consequence of the deposition process. The hollow channels of all samples were squeezed by the upper layers during the printing process, and the longitudinally printed samples were the most affected. When printing direction changes from longitudinal to transverse, the initial flexural strength decreases. Similarly, the average permeability of the cracked samples increases when the printing direction changes from longitudinal to transverse. Although the healing effectiveness regarding flexural strength is remarkable for all specimens, it was only possible to perform a single healing process as hollow channels were then blocked by the epoxy resin. The rough surface of the hollow channels is inferred to make it difficult to extract the epoxy resin out of the specimens.

In this study, an experimental setup to characterize the early-age creep of 3D printable mortar was proposed. The testing protocol comprises quasi-static compressive loading-unloading cycles, with 180-s holding periods in between. An analytical model based on a double power law was used to predict creep compliance with hardening time and loading duration as inputs. Subsequently, this analytical model was validated by comparison to uniaxial compression tests in which loading is increased incrementally, i.e., in steps, showing a good quantitative agreement. Minor differences between the two results were noted, most notably at the beginning of the test. This is because the determination of creep compliance for 3D printable mortar at fresh stage depends on the load level. In the end, the volumetric strain of tested samples from uniaxial compressive test is used to explain why the compressive loading affects the creep deformation.

The current study investigates short-term and long-term crack-healing behaviour of mortars embedded with bacteria-based poly-lactic acid (PLA) capsules under both ideal and realistic environmental conditions. Two sets of specimens were prepared and subjected to different healing regimes, with the first set kept in a mist room for varying short durations (i.e., 1 week, 2 weeks, 3 weeks and 8 weeks) and the second set placed in an unsheltered outdoor environment for a long-term healing process (i.e., 1 year). Alteration of microstructure because of self-healing was characterized by backscattered electron (BSE) imaging and energy dispersive X-ray spectroscopy (EDS) via crack cross-sections. Results show that visible crack healing enabled by bacteria began after 2 weeks in a humid environment. The healing products initially precipitated at crack mouths and gradually moved deeper into cracks, with the precipitated calcium carbonate crystals growing larger over time. After 8 weeks, healing products can be found even a few millimetres deep inside cracks. Observations of crack healing in a realistic environment revealed significant differences compared to healing under controlled conditions. While no healing products can be found at crack mouths, a substantial healing process was observed throughout the entire crack depth. It is likely that the environmental actions such as rainfall and/or freeze and thaw cycles may have worn away the healing products at crack mouths and thus led to a deeper ingress of oxygen into cracks, which promoted the activation of healing agents and associated calcium carbonate precipitation deep inside a crack.

The high deformation capacity of auxetic cementitious cellular composites (ACCCs) makes them promising for strain-based energy harvesting applications in infrastructure. In this study, a novel piezoelectric energy harvester (PEH) with ACCCs and surface-mounted PVDF film based on strain-induced piezoelectric mechanisms has been designed, fabricated, and experimentally tested. Furthermore, a numerical model for simulating the energy harvesting of ACCC-PVDF system undergoing repeated mechanical loading has been established and validated against the experimental data. The mechanical behavior of ACCCs was simulated by the concrete damage plasticity model during the preloading stage, which was converted to the second-elasticity model during cyclic loading stage. Based on the mechanical responses, analytical formulas for piezoelectric effects were developed to calculate the output voltage of the PVDF film. The output voltages of the ACCCs-PVDF system under different loading amplitudes and loading frequencies were assessed. The experimental results and models of the ACCCs-PVDF energy harvester lay a solid foundation for utilizing architected cementitious composites in energy harvesting applications to supply self-power electronics in infrastructure.

In this paper, optimization of vascular structure of self-healing concrete is performed with deep neural network (DNN). An input representation method is proposed to effectively represent the concrete beams with 6 round pores in the middle span as well as benefit the optimization process. To investigate the feasibility of using DNN for vascular structure optimization (i.e., optimization of the spatial arrangement of the vascular network), structure optimization improving peak load and toughness is first carried out. Afterwards, a hybrid target is defined and used to optimize vascular structure for self-healing concrete, which needs to be healable without significantly compromising its mechanical properties. Based on the results, we found it feasible to optimize vascular structure by fixing the weights of the DNN model and training inputs with the data representation method. The average peak load, toughness and hybrid target of the ML-recommended concrete structure increase by 17.31%, 34.16% and 9.51%. The largest peak load, toughness and hybrid target of the concrete beam after optimization increase by 0.17%, 14.13%, and 3.45% compared with the original dataset. This work shows that the DNN model has great potential to be used for optimizing the design of vascular system for self-healing concrete.

Additively manufactured vascular networks have great potential for use in autonomous self-healing of cementitious composites as they potentially allow multiple healing events to take place. However, the existence of a vascular tube wall may impede with the healing efficiency if it does not rupture timely to release the healing agent. The issue of vascular material design has therefore been a major topic of research. To overcome this, dissolvable Polyvinyl Alcohol (PVA) filament is adopted in this study to fabricate the vascular networks. Fabricated networks are coated with wax, placed in cementitious mortar and removed upon hardening, thereby leaving a network of hollow channels. Different printing directions were expected to affect the dissolvability of printed structures and were therefore fabricated and tested. Different shapes (i.e., 2D and 3D) of vascular networks were printed and embedded in the cementitious mortar. Four-point bending tests and permeability tests were performed to investigate the healing efficiency. Multiple healing cycles were applied in the cracked specimens. The results show that the vertically printed PVA tubes with wax coating have good dissolution behaviour. As expected, the existence of vascular networks decreases the initial flexural strength of the specimens. In terms of healing efficiency, excellent mechanical and water tightness recovery were achieved when using epoxy resin as the healing agent. The mechanical recovery after the first healing process is higher than the following healing process. The watertightness of the cracked samples keeps decreasing with the increase of healing cycles. Specimens embedded with 3D vascular networks have higher healing potential than those utilizing 2D vascular networks.

Cracking is one of the main causes for deterioration of concrete structures. Self-healing concrete with 3D-printed vascular networks has excellent potential for autonomous self-healing. This approach is scarcely investigated: no studies have been devoted to the influence of printing parameters on the properties of vascular based self-healing concrete. In this work, three-dimensional vascular structures with complex geometry were designed and printed with 4 different sets of printing parameters. First, the influence of the four, nominally identical, vascular networks on the initial flexural strength of self-healing concretes was experimentally investigated. In parallel, numerical modeling with a concrete damaged plasticity model (CDPM) in Abaqus software is used to simulate the influence of vascular networks on the mechanical properties of the self-healing composite. After the 4-point bending tests, epoxy resin is injected into the vascular networks as the healing agent to seal the cracks. Then, flexural strength regain and watertightness recovery were also measured. Based on the obtained results, we found that vascular based self-healing concretes have lower initial flexural strengths than the reference sample, as expected. The magnitude of the strength drop is shown to depend strongly on the printing parameters: the specimens with horizontally-printed vascular networks have higher flexural strength than the vertically-printed counterparts. Furthermore, vascular networks with a smaller printing layer-height have less influence on the initial flexural strength of vascular-based self-healing concrete compared to those with the larger printing layer-height. In terms of watertightness recovery, all tested vascular based self-healing samples showed a full (100%) recovery, which means that the printing direction and printing layer-height do not have an obvious effect on the watertightness recovery in this study. Numerical simulations of the mechanical performance of the composites with the CDPM show good agreement with the experiments, although printing quality of the vascular network influences the simulation accuracy. These simulations show great potential of using numerical simulations to design vascular based self-healing concrete in order to minimize a drop in mechanical properties, without compromising the healing efficiency. Overall, the designed 3D-printed vascular self-healing concretes show remarkable strength regain and watertightness recovery and provide a good basis for further research.

Tailoring lattice structures is a commonly used method to develop lattice materials with desired mechanical properties. However, for cementitious lattice materials, besides the macroscopic lattice structure, the multi-phase microstructure of cement paste may have substantial impact on the mechanical responses. Therefore, this work proposes a multi-scale numerical modelling method to simulate the deformation and fracture behavior of cementitious lattice materials, such that the influence of cement paste microstructure can be properly captured. On the microscale, the load–displacement response of cement paste is numerically simulated then experimentally validated. In order to rationally investigate the role of cement paste microstructure, the obtained load–displacement response was then formulated to several types of model inputs reflecting different degree of brittleness. These inputs were then used for simulating the mechanical response of macroscale cementitious lattices. By comparing the simulation to experiment, multi-linear behavior (ML) was found to an appropriate method to include the realistic pre-critical cracking and post-peak softening of cement paste in the model. Compared to ideally brittle behavior, using ML as input, the discrepancy between simulated and experimentally tested fracture energy decreases from 37.4% to 12.4%. In addition, the influence of lattice structure on the strength of cementitious lattices was also accurately captured by the proposed model. Randomized cementitious lattice has 21.6% (22.0% from simulation) lower strength than regular lattice. Moreover, the influence of fracture criterion of the proposed model is discussed and elaborated. Owning to the high simulation accuracy, the proposed multi-scale method in this work could be helpful for tailoring the fracture cementitious lattice materials for future studies.

A deeper insight into SO₃/Al₂O₃ ratio including the contribution of alumina in slag at early age is required to ensure a properly sulfated slag cement. In this paper, to investigate the effect of gypsum and alumina of slag, emphasis was laid on the hydration characteristics of C₃S-gypsum-slag system during the early age, of which slag was synthesized in the laboratory with varying Al₂O₃ contents from 3.69 to 18.19 wt%. The duration of dormant period during the hydration of C₃S depended on Al₂O₃ content of slag significantly; however, the amount of silicate reaction before the onset of aluminate reaction was independent of slag chemistry and gypsum content added. The rate of aluminate reaction was controlled by the availability of reactants, SO₄²⁻ and Al³⁺ ions in particular, which were sourced from gypsum and slag, respectively. Calcium monosulfoaluminate only occurred in mixture when slag contained a high amount of Al₂O₃ (18.19 wt% in this study) at early age, and its formation proceeded continuously at the expense of ettringite. Sulfur rich species incorporated in slag started to participate into aluminate reaction after the main hydration peak of C₃S, and it played a similar role to gypsum.

This study aims to provide an efficient and accurate machine learning (ML) approach for predicting the creep behavior of concrete. Three ensemble machine learning (EML) models are selected in this study: Random Forest (RF), Extreme Gradient Boosting Machine (XGBoost) and Light Gradient Boosting Machine (LGBM). Firstly, the creep data in Northwestern University (NU) database is preprocessed by a prebuilt XGBoost model and then split into a training set and a testing set. Then, by Bayesian Optimization and 5-fold cross validation, the 3 EML models are tuned to achieve high accuracy (R² = 0.953, 0.947 and 0.946 for LGBM, XGBoost and RF, respectively). In the testing set, the EML models show significantly higher accuracy than the equation proposed by the fib Model Code 2010 (R² = 0.377). Finally, the SHapley Additive exPlanations (SHAP), based on the cooperative game theories, are calculated to interpretate the predictions of the EML model. Five most influential parameters for concrete creep compliance are identified by the SHAP values of EML models as follows: time since loading, compressive strength, age when loads are applied, relative humidity during the test and temperature during the test. The patterns captured by the three EML models are consistent with theoretical understanding of factors that influence concrete creep, which proves that the proposed EML models show reasonable predictions.

This study aims to provide an efficient alternative for predicting creep modulus of cement paste based on Deep Convolutional Neural Network (DCNN). First, a microscale lattice model for short-term creep is adopted to build a database that contains 18,920 samples. Then, 3 DCNNs with different consecutive convolutional layers are built to learn from the database. Finally, the performance of DCNNs is tested on unseen testing samples. The results show that the DCNNs can achieve high accuracy in the testing set, with the R² all higher than 0.96. The distribution of creep modulus predicted by the DCNNs coincides with that of the original data. Furthermore, through analyzing the feature maps, it is found that the DCNNs can correctly capture the local importance of different microstructural phases. The DCNN allows therefore prediction of the creep modulus based on microstructural input, which saves computational resources of segmentation procedure and multiple incremental FEM calculations.

Compressive strength is the most significant metric to evaluate the mechanical properties of concrete. Machine learning (ML) methods have shown promising results for predicting compres-sive strength of concrete. However, at present, no in-depth studies have been devoted to the influence of dimensionality reduction on the performance of different ML models for this application. In this work, four representative ML models, i.e., Linear Regression (LR), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Artificial Neural Network (ANN), are trained and used to predict the compressive strength of concrete based on its mixture composition and cur-ing age. For each ML model, three kinds of features are used as input: the eight original features, six Principal Component Analysis (PCA)-selected features, and six manually selected features. The performance as well as the training speed of those four ML models with three different kinds of features is assessed and compared. Based on the obtained results, it is possible to make a relatively accurate prediction of concrete compressive strength using SVR, XGBoost, and ANN with an R-square of over 0.9. When using different features, the highest R-square of the test set occurs in the XGBoost model with manually selected features as inputs (R-square = 0.9339). The prediction accuracy of the SVR model with manually selected features (R-square = 0.9080) or PCA-selected features (R-square = 0.9134) is better than the model with original features (R-square = 0.9003) without dramatic running time change, indicating that dimensionality reduction has a positive influence on SVR model. For XGBoost, the model with PCA-selected features shows poorer performance (R-square = 0.8787) than XGBoost model with original features or manually selected features. A possible reason for this is that the PCA-selected features are not as distinguishable as the manually selected features in this study. In addition, the running time of XGBoost model with PCA-selected features is longer than XGBoost model with original features or manually selected features. In other words, dimensionality reduction by PCA seems to have an adverse effect both on the performance and the running time of XGBoost model. Dimensionality reduction has an adverse effect on the performance of LR model and ANN model because the R-squares on test set of those two models with manually selected features or PCA-selected features are lower than models with original features. Although the running time of ANN is much longer than the other three ML models (less than 1s) in three scenarios, dimensionality reduction has an obviously positive influence on running time without losing much prediction accuracy for ANN model.