skip to main content

Title: Predicting Concrete’s Strength by Machine Learning: Balance between Accuracy and Complexity of Algorithms
The properties of concretes are controlled by the rate of reaction of their precursors, the chemical composition of the binding phase(s), and their structure at different scales. However, the complex and multiscale structure of the cementitious hydrates and the dissimilar rates of numerous chemical reactions make it challenging to eluci- date such linkages. In particular, reliable predictions of strength development in concretes remain unavailable. As an alternative route to physics- or chemistry-based models, machine learning (ML) offers a means to develop powerful predictive models for materials using existing data. Here, it is shown that ML models can be used to accurately predict concrete’s compressive strength at 28 days. This approach relies on the analysis of a large data set (>10,000 observations) of measured compressive strengths for industrially produced concretes, based on knowledge of their mixture proportions. It is demonstrated that these models can readily predict the 28-day compressive strength of any concrete based merely on the knowledge of the mixture proportions with an accuracy of approximately ±4.4 MPa (as captured by the root- mean-square error). By comparing the performance of select ML algorithms, the balance between accuracy, simplicity, and inter- pretability in ML approaches is discussed.
; ; ; ; ; ; ;
Award ID(s):
Publication Date:
Journal Name:
ACI Materials Journal
Sponsoring Org:
National Science Foundation
More Like this
  1. Calcium aluminate cement (CAC) has been explored as a sustainable alternative to Portland cement, the most widely used type of cement. However, the hydration reaction and mechanical properties of CAC can be influenced by various factors such as water content, Li2CO3 content, and age. Due to the complex interactions between the precursors in CAC, traditional analytical models have struggled to predict CAC binders’ compressive strength and porosity accurately. To overcome this limitation, this study utilizes machine learning (ML) to predict the properties of CAC. The study begins by using thermodynamic simulations to determine the phase assemblages of CAC at different ages. The XGBoost model is then used to predict the compressive strength, porosity, and hydration products of CAC based on the mixture design and age. The XGBoost model is also used to evaluate the influence of input parameters on the compressive strength and porosity of CAC. Based on the results of this analysis, a closed-form analytical model is developed to predict the compressive strength and porosity of CAC accurately. Overall, the study demonstrates that ML can be effectively used to predict the properties of CAC binders, providing a valuable tool for researchers and practitioners in the field of cement science.
  2. Pervious concrete, which has recently found new applications in buildings, is both energy- and carbon-intensive to manufacture. However, similar to normal concrete, some of the initial CO2 emissions associated with pervious concrete can be sequestered through a process known as carbonation. In this work, the theoretical formulation and application of a mathematical model for estimating the carbon dioxide (CO2) sequestration potential of pervious concrete is presented. Using principles of cement and carbonation chemistry, the model related mixture proportions of pervious concretes to their theoretical in situ CO2 sequestration potential. The model was subsequently employed in a screening life cycle assessment (LCA) to quantify the percentage of recoverable CO2 emissions—namely, the ratio of in situ sequesterable CO2 to initial cradle-to-gate CO2 emissions—for common pervious concrete mixtures. Results suggest that natural carbonation can recover up to 12% of initial CO2 emissions and that CO2 sequestration potential is maximized for pervious concrete mixtures with (i) lower water-to-cement ratios, (ii) higher compressive strengths, (iii) lower porosities, and (iv) lower hydraulic conductivities. However, LCA results elucidate that mixtures with maximum CO2 sequestration potential (i.e., mixtures with high cement contents and CO2 recoverability) emit more CO2 from a net-emissions perspective, despite their enhanced in situ CO2more »sequestration potential.« less
  3. Autonomous chemical process development and optimization methods use algorithms to explore the operating parameter space based on feedback from experimentally determined exit stream compositions. Measuring the compositions of multicomponent streams is challenging, requiring multiple analytical techniques to differentiate between similar chemical components in the mixture and determine their concentration. Herein, we describe a universal analytical methodology based on multitarget regression machine learning (ML) models to rapidly determine chemical mixtures' compositions from Fourier transform infrared (FTIR) absorption spectra. Specifically, we used simulated FTIR spectra for up to 6 components in water and tested seven different ML algorithms to develop the methodology. All algorithms resulted in regression models with mean absolute errors (MAE) between 0–0.27 wt%. We validated the methodology with experimental data obtained on mixtures prepared using a network of programmable pumps in line with an FTIR transmission flow cell. ML models were trained using experimental data and evaluated for mixtures of up to 4-components with similar chemical structures, including alcohols ( i.e. , glycerol, isopropanol, and 1-butanol) and nitriles ( i.e. , acrylonitrile, adiponitrile, and propionitrile). Linear regression models predicted concentrations with coefficients of determination, R 2 , between 0.955 and 0.986, while artificial neural network models showed a slightlymore »lower accuracy, with R 2 between 0.854 and 0.977. These R 2 correspond to MAEs of 0.28–0.52 wt% for mixtures with component concentrations between 4–10 wt%. Thus, we demonstrate that ML models can accurately determine the compositions of multicomponent mixtures of similar species, enhancing spectroscopic chemical quantification for use in autonomous, fast process development and optimization.« less
  4. This study characterized and evaluated the use of reclaimed fly ash (RFA) and reclaimed ground bottom ash (GBA) as alternative sources of supplementary cementitious materials (SCMs) for the production of concrete mixtures. Conventional Class F fly ash (FA) was also evaluated for comparison. The effects of SCM content on fresh and hardened properties of concrete were investigated by replacing 10%, 20%, and 30% of cement by mass. Characterization results showed that all three ashes met ASTM C618 chemical requirements (i.e., sum of SiO 2  + Al 2 O 3  + Fe 2 O 3 , CaO, SO 3 , moisture content, and loss of ignition) and 7- and 28-days strength activity index (SAI) requirements for Class F FA. In addition, RFA exhibited slightly higher SAI at 28 days of curing, followed by GBA and FA. In relation to fresh concrete properties, FA increased the concrete slump compared with the control mixture, whereas RFA and GBA decreased the concrete slump. However, GBA produced more significant slump decrements than RFA, which was attributed to the irregular angular particles of GBA. Generally, all the coal ashes produced decrements in air content compared with the control mixture. Comparatively, among the three ashes, GBA exhibited the highest 28-more »and 90-days compressive strength and surface resistivity (SR) at all cement replacement levels. Furthermore, at 90 days of curing, RFA and GBA concrete mixtures outperformed the FA concrete mixtures in relation to compressive strength and SR. Consequently, both RFA and GBA are promising SCMs for concrete materials.« less
  5. Abstract

    Carbonaceous (e.g., limestone) and aluminosilicate (e.g., calcined clay) mineral additives are routinely used to partially replace ordinary portland cement in concrete to alleviate its energy impact and carbon footprint. These mineral additives—depending on their physicochemical characteristics—alter the hydration behavior of cement; which, in turn, affects the evolution of microstructure of concrete, as well as the development of its properties (e.g., compressive strength). Numerical, reaction-kinetics models—e.g., phase boundary nucleation-and-growth models; which are based partly on theoretically-derived kinetic mechanisms, and partly on assumptions—are unable to produce a priori prediction of hydration kinetics of cement; especially in multicomponent systems, wherein chemical interactions among cement, water, and mineral additives occur concurrently. This paper introduces a machine learning-based methodology to enable prompt and high-fidelity prediction of time-dependent hydration kinetics of cement, both in plain and multicomponent (e.g., binary; and ternary) systems, using the system’s physicochemical characteristics as inputs. Based on a database comprising hydration kinetics profiles of 235 unique systems—encompassing 7 synthetic cements and three mineral additives with disparate physicochemical attributes—a random forests (RF) model was rigorously trained to establish the underlying composition-reactivity correlations. This training was subsequently leveraged by the RF model: to predict time-dependent hydration kinetics of cement in new, multicomponent systems;more »and to formulate optimal mixture designs that satisfy user-imposed kinetics criteria.

    « less