Application of Artificial Neural Networks for Prediction of Mechanical Properties of CNT/CNF Reinforced Concrete

Microcapsule Triggering Mechanics in Cementitious Materials: A Modelling and Machine Learning Approach

Self-healing cementitious materials containing microcapsules filled with healing agents can autonomously seal cracks and restore structural integrity. However, optimising the microcapsule mechanical properties to survive concrete mixing whilst still rupturing at the cracked interface to release the healing agent remains challenging. This study develops an integrated numerical modelling and machine learning approach for tailoring acrylate-based microcapsules for triggering within cementitious matrices. Microfluidics is first utilised to produce microcapsules with systematically varied shell thickness, strength, and cement compatibility. The capsules are characterised and simulated using a continuum damage mechanics model that is able to simulate cracking. A parametric study investigates the key microcapsule and interfacial properties governing shell rupture versus matrix failure. The simulation results are used to train an artificial neural network to rapidly predict the triggering behaviour based on capsule properties. The machine learning model produces design curves relating the microcapsule strength, toughness, and interfacial bond to its propensity for fracture. By combining advanced simulations and data science, the framework connects tailored microcapsule properties to their intended performance in complex cementitious environments for more robust self-healing concrete systems.

Percolation in Carbon Nanotube-Reinforced Polymers for Strain-Sensing Applications: Computational Investigation on Carbon Nanotube Distribution, Curvature, and Aggregation

The present article investigates the possibility of simulating the electrical conductivity of carbon nanotube-reinforced polymer composites by numerical methods. Periodic representative volume elements are generated by randomly distributing perfectly conductive reinforcements in an insulating matrix and are used to assemble an electrical network representative of the nanocomposite, where the nanotube-nanotube contacts are considered equivalent resistors modeled by means of Simmons' equation. A comparison of the results with experimental data from the literature supports the conclusion that a random distribution of reinforcements is not suitable for simulating this class of materials since percolation thresholds and conductivity trends are different, with experimental percolation taking place before the expectations. Including nanotube curvature does not solve the issue, since it hinders percolation even further. In agreement with experimental observations, the investigation suggests that a suitable approach requires the inclusion of aggregation during the volume element generation to reduce the volume fraction required to reach percolation. Some solutions available in the literature to generate properly representative volume elements are thus listed. Concerning strain sensing, the results suggest that representative volume elements generated with random distributions overestimate the strain sensitivity of the actual composites.

Optimisation of Embodied Carbon and Compressive Strength in Low Carbon Concrete

To improve the prediction of compressive strength and embodied carbon of low carbon concrete using a program algorithm developed in MATLAB, 84 datasets of concrete mix raw materials were used. The influence of water, silica fume and ground granular base slag was found to have a significant impact on the extent of low carbon concrete behaviour in terms of compressive strength and embodied carbon. While the concrete compressive strength for normal concrete increases with reducing water content, it is observed that the low carbon concrete using lightweight aggregate material increases in compressive strength with an increase in embodied carbon. From the result of the analysis, a function was developed that was able to predict the associated embodied carbon of a concrete mix for a given water-to-cement ratio. The use of an alkaline solution is observed to increase the compressive strength of low carbon concrete when used in combination with ground granular base slag and silica fume. It is further shown that ground granular base slag contributes significantly to an increase in the compressive strength of Low carbon concrete when compared with pulverised fly ash. The optimised mix design program resulted in a 26% reduction in embodied carbon and an R² value of 0.9 between the measured compressive strength and the optimised compressive strength.

Prediction of Mechanical Properties of Highly Functional Lightweight Fiber-Reinforced Concrete Based on Deep Neural Network and Ensemble Regression Trees Methods

Currently, one of the topical areas of application of artificial intelligence methods in industrial production is neural networks, which allow for predicting the performance properties of products and structures that depend on the characteristics of the initial components and process parameters. The purpose of the study was to develop and train a neural network and an ensemble model to predict the mechanical properties of lightweight fiber-reinforced concrete using the accumulated empirical database and data from construction industry enterprises, and to improve production processes in the construction industry. The study applied deep learning and an ensemble of regression trees. The empirical base is the result of testing a series of experimental compositions of fiber-reinforced concrete. The predicted properties are cubic compressive strength, prismatic compressive strength, flexural tensile strength, and axial tensile strength. The quantitative picture of the accuracy of the applied methods for strength characteristics varies for the deep neural network method from 0.15 to 0.73 (MAE), from 0.17 to 0.89 (RMSE), and from 0.98% to 6.62% (MAPE), and for the ensemble of regression trees, from 0.11 to 0.62 (MAE), from 0.15 to 0.80 (RMSE), and from 1.30% to 3.4% (MAPE). Both methods have shown high efficiency in relation to such a hard-to-predict material as concrete, which is so heterogeneous in structure and depends on many factors. The value of the developed models lies in the possibility of obtaining additional useful information in the process of preparing highly functional lightweight fiber-reinforced concrete without additional experiments.

High-Performance Concrete Strength Prediction Based on Machine Learning

High-performance concrete is a new high-tech concrete, produced using conventional materials and processes, with all the mechanical properties required for concrete structures, with high durability, high workability, and high volume stability of the concrete. The compressive strength of high-performance concrete has exceeded 200 MPa. 28-d average strength between 100 to 120 MPa of high-performance concrete has been widely used in engineering. Compressive strength is one of the important parameters of concrete, and carrying out concrete compressive strength prediction is of high reference value for concrete design. Eight variables related to concrete strength are used as the input of the machine learning algorithm, and the compressive strength of HPC is used as the object of study. 60 samples are constructed as the dataset by concrete preparation, and the prediction of compressive strength of HPC is carried out by combining the XGBoost algorithm. In addition, SVR algorithm and RF algorithm are also performed on the same dataset. The results show that the XGBoost model has the highest prediction accuracy among the three machine learning models, and the XGBoost algorithm scores 0.9993 for ${\mathbf{R}}^2$ and 1.372 for RMSE on the test set. The XGBoost algorithm has high prediction accuracy in predicting the compressive strength of HPC, and the choice of model is important for improving the prediction accuracy.

Prediction of Bonding Strength of Externally Bonded SRP Composites Using Artificial Neural Networks

External bonding of fiber reinforced composites is currently the most popular method of strengthening building structures. Debonding performance is critical to the effectiveness of such strengthening. Many models of bond prediction can be found in the literature. Most of them were developed based on laboratory research, therefore, their accuracy with less popular strengthening systems is limited. This manuscript presents the possibility of using a model based on neural networks to analyze and predict the debonding strength of steel-reinforced polymer (SRP) and steel-reinforced grout (SRG) composites to concrete. The model is built on the basis of laboratory testing of 328 samples obtained from the literature. The results are compared with a dozen of the most popular analytical methods for predicting the load capacity. The prediction accuracy in the neural network model is by far the best. The total correlation coefficient reaches a value of 0.913 while, for the best analytical method (Swiss standard SIA 166 model), it is 0.756. The sensitivity analysis confirmed the importance of the modulus of elasticity and the concrete strength for debonding. It is also interesting that the width of the element proved to be very important, which is probably related to the low variability of this parameter in the laboratory tests.

Multiple Analytical Models to Evaluate the Impact of Carbon Nanotubes on the Electrical Resistivity and Compressive Strength of the Cement Paste

Cement paste is the most common construction material being used in the construction industry. Nanomaterials are the hottest topic worldwide, which affect the mechanical properties of construction materials such as cement paste. Cement pastes containing carbon nanotubes (CNTs) are piezoresistive intelligent materials. The electrical resistivity of cementitious composites varies with the stress conditions under static and dynamic loads as carbon nanotubes are added to the cement paste. In cement paste, electrical resistivity is one of the most critical criteria for structural health control. Therefore, it is essential to develop a reliable mathematical model for predicting electrical resistivity. In this study, four different models—including the nonlinear regression model (NLR), linear regression model (LR), multilinear regression model (MLR), and artificial neural network model (ANN)—were proposed to predict the electrical resistivity of cement paste modified with carbon nanotube. Furthermore, the correlation between the compressive strength of cement paste and the electrical resistivity model has also been proposed in this study and compared with models in the literature. In this respect, 116 data points were gathered and examined to develop the models, and 56 data points were collected for the proposed correlation model. Most critical parameters influencing the electrical resistivity of cement paste were considered during the modeling process—i.e., water to cement ratio ranged from 0.2 to 0.485, carbon nanotube percentage varied from 0 to 1.5%, and curing time ranged from 1 to 180 days. The electrical resistivity of cement paste with a very large number ranging from 0.798–1252.23 Ω.m was reported in this study. Furthermore, various statistical assessments such as coefficient of determination (R2), mean absolute error (MAE), root mean square error (RMSE), scatter index (SI), and OBJ were used to investigate the performance of different models. Based on statistical assessments—such as SI, OBJ, and R2—the output results concluded that the artificial neural network ANN model performed better at predicting electrical resistivity for cement paste than the LR, NLR, and MLR models. In addition, the proposed correlation model gives better performance based on R2, RMSE, MAE, and SI for predicting compressive strength as a function of electrical resistivity compared to the models proposed in the literature.