Real-time prediction of early concrete compressive strength using AI and hydration monitoring

Abstract

The continuous evolution of construction technologies, particularly in reinforced concrete production, demands advanced, reliable, and efficient methodologies for real-time monitoring and prediction of concrete compressive strength. Traditional laboratory methods for assessing compressive strength are time-intensive and can introduce delays in construction workflows. This study introduces a comprehensive framework for a system designed to predict early-age compressive strength of concrete through continuous monitoring of the cement hydration process using a custom artificial intelligence (AI) model. The system integrates a network of temperature sensors, communication modules, and a centralized database server to collect, transmit, and analyze real-time data during the concrete curing process. The AI model, a deep neural network leverages this data to generate accurate strength predictions. The system architecture emphasizes scalability, robustness, and integration with existing construction management systems. Empirical results indicate that the proposed system achieves high predictive accuracy, with an R² value of 0.996 and RMSE of 0.143 MPa, offering a robust tool for real-time decision-making in construction. This study also critically evaluates the system's performance, identifying key strengths such as predictive accuracy and real-time processing capabilities, and addresses challenges related to wireless communication reliability and sensor power supply. Recommendations are provided for enhancing system precision, improving communication technologies, optimizing power management, and ensuring scalability across diverse construction contexts. The developed system, which is part of the "CONCRESENSE" project and protected under European patent number 245107 (2024), represents a significant advancement in construction technology, with substantial implications for enhancing the safety, efficiency, and quality of reinforced concrete structures.

Introduction

In the rapidly evolving construction industry, the continuous improvement of reinforced concrete production technologies remains crucial. Reinforced concrete is a cornerstone of modern building practices, valued for its versatility, durability, and strength¹. Among its various performance metrics, compressive strength is particularly important, as it not only reflects structural integrity but also determines load-bearing capacity and overall durability². Traditionally, compressive strength has been assessed through standardized laboratory tests, such as those outlined in European Standard EN-206³. While these methods are well-established and accurate, their reliance on time-consuming sample preparation and curing procedures often introduces delays in project timelines, especially when early strength data are needed to inform decisions related to formwork removal, load applications, or subsequent construction phases⁴. Moreover, inconsistencies arising from variations in specimen preparation, curing conditions, and testing protocols can compromise the reliability and representativeness of laboratory-based results⁵. To address these challenges, there is a growing demand for modern, efficient, and reliable methods for real-time monitoring and prediction of concrete strength. Techniques like the maturity method, which correlates concrete temperature history with strength development, have provided more immediate feedback⁶. However, these approaches often require complex calibration and may not fully account for variations in mix design, environmental conditions, or other factors influencing strength gain^7,8. Recent advances in artificial intelligence (AI) and machine learning have shown promise in improving the accuracy of strength predictions^9–11. Combining real-time data acquisition with predictive modeling offers the potential to revolutionize quality control in concrete production, enhancing construction efficiency, reducing downtime, and improving structural safety¹². Despite progress, existing AI- and sensor-based systems still face limitations. These include scalability issues, limited datasets that reduce predictive accuracy, and challenges in integrating with established construction workflows^13,14. Many current solutions struggle to adapt to differing concrete mixes or environmental contexts, thus limiting their widespread applicability¹⁴. The primary objective of this study is to develop a robust, scalable framework for real-time prediction of early-age concrete compressive strength. By integrating continuous temperature monitoring during the hydration process with a custom-trained AI model, our approach aims to deliver immediate, accurate strength forecasts. This data-driven methodology is designed to support more efficient construction scheduling, reduce costs, and enhance structural safety by guiding critical decisions such as formwork removal and load application times. Unlike existing methods, our framework prioritizes scalability, operational efficiency, and enhanced predictive accuracy. Central to this system is a bespoke AI model trained on comprehensive datasets collected under diverse environmental conditions and mix designs. Through the integration of sensor networks, wireless communication modules, and on-site data processing units, the system dynamically responds to real-time changes, thereby improving the reliability and generalizability of strength predictions. This research is conducted as part of the “Prototype System for Predicting Early Compressive Strength of Concrete by Monitoring the Hydration Process, as a Support for the Technological Cycle of Producing Reinforced Concrete Structural Elements (CONCRESENSE) [CONCRESENS_CTWT_II.4.0_1]” project. Funded by the Ministry of Education and Science (MEiN) under the "Innovation Incubator 4.0" program, the resulting device has obtained patent protection in the European Union (Patent No. 245107, 2024). This legal safeguard ensures that the system’s innovative data processing techniques, sensor configurations, and AI-driven solutions remain protected, thereby bolstering its commercial potential and fostering strategic industry partnerships. By delivering accurate, real-time strength predictions, the proposed framework can significantly improve construction efficiency through optimized scheduling and minimized downtime. Enhanced predictive accuracy helps prevent premature loading and ensures that subsequent construction activities proceed only when the necessary strength thresholds are attained, thereby improving safety and structural integrity. From a financial perspective, the system may reduce costs by lowering the need for multiple formwork sets, decreasing delays, and improving workforce allocation. The modular design of our framework facilitates adaptation to a wide spectrum of project scales and complexities—ranging from residential developments to large-scale infrastructure projects such as bridges and high-rise buildings. By addressing current limitations, offering a data-driven approach, and securing intellectual property rights, this work represents a substantial advancement in construction technology. In integrating cutting-edge AI algorithms, patented innovations, and flexible deployment strategies, it paves the way for a new era of efficiency, safety, and quality assurance in the construction industry.

Theoretical foundations of the framework

Cement hydration process

The cement hydration process is the cornerstone of concrete's development, transforming a mixture of cement and water into a hard, durable structure. This chemical reaction between the cement components and water is exothermic, releasing heat and leading to the formation of various hydration products. These products, primarily calcium silicate hydrate (C–S–H) and calcium hydroxide (Ca(OH)₂), are crucial in determining the mechanical properties and durability of the resulting concrete. A deep understanding of this process is essential for optimizing concrete performance and ensuring its long-term stability. Portland cement, the most widely used cement type, consists of four primary clinker phases: alite (C₃S), belite (C₂S), tricalcium aluminate (C₃A), and calcium aluminoferrite (C₄AF), as in Fig. 1A–D respectively. Each of these components reacts with water at different rates and contributes uniquely to the development of concrete's strength and durability.

Fig. 1.

Minerals involved in the cement hydration process: Alite (A), Belite (B), Tricalcium Aluminate (C), Aluminate Ferrite (D).

Alite (C₃S)

Alite, or tricalcium silicate (C₃S), is the most reactive of the cement compounds and is responsible for the early strength of concrete. The hydration of alite is a relatively rapid process, resulting c strength, particularly in the early stages of hydration. The reaction can be represented as follows:

1

where C₃S₂H₃ represents the calcium silicate hydrate gel, and Ca(OH)₂ is the by-product that contributes to the concrete's alkalinity¹⁵.

Belite (C₂S)

Belite, or dicalcium silicate (C₂S), hydrates much more slowly than alite but plays a crucial role in the long-term strength development of concrete. The hydration of belite also produces C–S–H and Ca(OH)₂, but the rate of reaction is slower, contributing to a gradual increase in strength over time. This slow hydration is beneficial for the durability of concrete, particularly in structures where long-term performance is critical. The reaction of belite can be expressed as:

2

The C–S–H formed by belite is similar to that formed by alite but is typically more crystalline, contributing to the overall density and durability of the concrete matrix¹⁶.

Tricalcium aluminate (C₃A)

Tricalcium aluminate (C₃A) reacts very quickly with water, releasing a significant amount of heat in an exothermic reaction. The rapid hydration of C₃A can led to flash setting, a phenomenon where the concrete sets too quickly, which is undesirable for workability. To mitigate this, gypsum (CaSO₄·2H₂O) is added to the cement, which reacts with C₃A to form ettringite (C₆A₃H₃₂), a stable hydration product that controls the setting time of the concrete:

3

The formation of ettringite is crucial for maintaining workability and preventing flash setting, ensuring that the concrete can be properly placed and finished¹⁷.

Aluminate ferrite (C₄AF)

Aluminate ferrite, or tetracalcium aluminoferrite (C₄AF), hydrates in a manner similar to C₃A but at a slower rate. While its contribution to the early strength of concrete is minimal, C₄AF plays an essential role in the long-term stability and durability of the concrete. The hydration of C₄AF contributes to the color of the cement and provides some resistance to sulfate attack, though its primary function is to act as a flux during the cement manufacturing process:

4

The hydration products of C₄AF are complex and can vary depending on the composition of the cement and the environmental conditions during curing¹⁸.

Phases of the cement hydration process

The hydration process of cement can be divided into several distinct phases, each characterized by different rates of reaction and development of the concrete microstructure¹⁹:

• Induction Phase: Immediately after mixing cement with water, the hydration reactions are relatively slow, leading to a dormant period. This phase, also known as the induction or dormant phase, lasts for several hours and is crucial for the workability of the concrete, allowing sufficient time for transportation, placement, and finishing. During this phase, the initial setting begins as the hydration products slowly start to form.

• Acceleration Phase: Following the induction phase, the rate of hydration reactions rapidly increases. This phase is marked by the accelerated formation of C–S–H and other hydration products, leading to a rapid gain in concrete strength. The microstructure of the concrete undergoes significant development during this phase, with the formation of a dense network of C–S–H and other hydration products that contribute to the early strength of the concrete.

• Deceleration Phase: As the concrete continues to gain strength, the rate of hydration reactions begins to slow down. This deceleration phase can last for several months or even years, depending on the environmental conditions and the composition of the cement. Despite the slower reaction rates, the hydration process continues, contributing to the long-term strength and durability of the concrete. The continued formation of C–S–H and other hydration products helps to fill the pores within the concrete matrix, reducing permeability and enhancing durability.

The sequence of these phases and the overall kinetics of the hydration process are influenced by various factors, including temperature, water-cement ratio, and the presence of admixtures^20–22.

The importance of temperature in the hydration process

Temperature is a critical factor influencing the kinetics of the cement hydration process, impacting both the rate of chemical reactions and the ultimate properties of the hardened concrete. The understanding of temperature effects is vital for the proper management of concrete quality, particularly under varying climatic conditions and during different phases of construction. This section delves into the effects of temperature on hydration, examining its implications on concrete strength, durability, and microstructural development, and provides a review of relevant literature.

Temperature and the rate of hydration

The rate of cement hydration is highly sensitive to temperature, with the reaction rate generally increasing as temperature rises. This phenomenon is can be attributed to the Arrhenius principle, which states that reaction rates increase exponentially with temperature, as demonstrated in multiple studies^23,24. At elevated temperatures (e.g., above 30 °C), the dissolution of cement phases, such as tricalcium silicate (C₃S) and dicalcium silicate (C₂S), is accelerated, leading to a more rapid formation of hydration products, primarily calcium silicate hydrate (C–S–H) and calcium hydroxide (Ca(OH)₂). Consequently, higher temperatures promote faster early strength development, which can be particularly advantageous when early formwork removal or rapid load application is necessary. For instance, in warm climates or when using heated mixing water in cold weather conditions, achieving early strength targets within shorter timeframes becomes more feasible²⁵. Conversely, at lower temperatures (e.g., below 10 °C), the hydration reactions slow considerably, necessitating extended curing times to achieve the same levels of strength and durability that would develop under moderate temperatures (around 20 °C)²⁶. In cold weather concreting, external heating or insulation measures are often employed to counteract the deceleration of hydration and ensure that the desired early strength is attained within the required construction schedule²⁵. While the temperature-driven acceleration of hydration can offer practical benefits, it must be balanced against potential drawbacks. Rapid early strength gain may come at the expense of long-term durability, as the accelerated reactions can produce a more heterogeneous microstructure with increased internal stresses²⁶. Furthermore, the exothermic nature of cement hydration means that higher temperatures, especially in large concrete masses, lead to greater internal heat generation. If not properly managed through measures like using low-heat cements, cooling pipes, or adjusting mix designs, this can result in significant temperature gradients between the interior and exterior of the concrete element. Such gradients can induce tensile stresses that exceed the developing tensile strength of the concrete, ultimately causing thermal cracking²⁷. Beyond compromising structural integrity, these cracks can create ingress pathways for aggressive agents (e.g., chloride ions), reducing the service life of the structure and increasing maintenance costs^27,28.

Temperature and microstructural development

Temperature plays a pivotal role in determining the microstructural characteristics of hydrated cement paste, which in turn influences the mechanical properties and long-term durability of the concrete. At high temperatures, the rapid rate of hydration leads to denser microstructures with lower overall porosity. This densification occurs due to the accelerated precipitation of C–S–H and the quicker consumption of capillary water, often resulting in higher early compressive strength and reduced permeability^29,30. For instance, concretes cured at elevated temperatures may reach a required compressive strength in substantially less time than concretes cured under cooler conditions, which is beneficial in time-sensitive construction projects. These advantages may be offset by certain undesirable effects. The denser microstructure produced at high temperatures can be more brittle, which, combined with faster moisture loss, can increase the propensity for shrinkage cracking. Such microstructural changes have been observed to compromise long-term performance, particularly when the concrete is exposed to aggressive service environments^29,30. Furthermore, higher temperatures can yield larger, less uniform C–S–H gel structures with reduced binding efficiency, leading to heterogeneous microstructures that are detrimental to durability over extended service periods³¹. The formation of C–S–H, the primary driver of strength development in concrete, is notably influenced by temperature. While higher temperatures accelerate C–S–H formation and promote rapid early strength gain, the resultant C–S–H tends to have a coarser and more crystalline morphology, which may be less effective at sustaining strength and durability in the long term^32,33. Similarly, the solubility and mobility of calcium ions increase with temperature, accelerating Ca(OH)2 formation. Although the presence of Ca(OH)2 helps maintain the high pH needed to protect embedded steel reinforcement, excessive Ca(OH)2 can contribute to a more brittle matrix and increase the risk of alkali-silica reaction (ASR) when reactive aggregates are present^34,35. In practice, understanding these temperature-related effects is crucial for tailoring the concrete mixture and curing conditions to meet specific performance criteria. For example, in hot climates or when placing concrete in warm environments, strategies such as employing supplementary cementitious materials (SCMs), using chilled mixing water, or scheduling concrete pours during cooler periods may help mitigate the negative effects of high-temperature hydration. Conversely, in cold climates, insulating formwork and using heating blankets or enclosures can help maintain sufficient temperatures for achieving the desired early strength and long-term durability. These measures, supported by the research findings cited above, demonstrate how temperature management is integral to optimizing both fresh and hardened properties of concrete under varying climatic conditions.

Methods for monitoring the hydration process

The hydration process of cementitious materials plays a crucial role in determining the mechanical properties and durability of concrete. Accurate real-time monitoring of this process is essential for predicting the development of concrete strength and optimizing curing conditions. Over the years, several advanced technologies have been developed to provide detailed insights into the hydration process, each with distinct advantages and limitations. Below is an expanded discussion of these methods, with references to relevant literature. Isothermal calorimetry is one of the most widely used techniques for monitoring the hydration process of cement. This method measures the heat flow associated with the exothermic reactions that occur during hydration. The heat released is directly proportional to the rate of the hydration reaction, making it possible to track the progress of hydration over time. The data obtained from isothermal calorimetry can be used to calculate the degree of hydration, which is a critical parameter for understanding the development of mechanical properties in concrete. Studies have shown that isothermal calorimetry provides highly accurate and reproducible results, making it a reliable method for predicting concrete strength. However, its application is often limited by the need for specialized equipment and the complexity of data interpretation, particularly when dealing with blended cements or supplementary cementitious materials (SCMs)^36,37. Spectroscopic methods, including Nuclear Magnetic Resonance (NMR) spectroscopy, have been increasingly employed to monitor the hydration process at the molecular level. NMR, for instance, allows for the non-destructive analysis of the chemical environment of nuclei in the cement paste, providing detailed information on the formation of hydration products, such as calcium-silicate-hydrate (C–S–H) and ettringite. NMR spectroscopy is particularly valuable in studying the dynamics of water in the cement paste, as it can differentiate between bound and free water, giving insights into the degree of hydration and the formation of microstructure. While NMR offers in-depth molecular-level information, it requires expensive equipment and specialized knowledge for interpretation, limiting its use primarily to research settings^38,39. Electrochemical sensors are an emerging technology for real-time monitoring of the hydration process in concrete. These sensors measure changes in the electrical properties of concrete, such as electrical conductivity and impedance, which are directly influenced by the ongoing hydration reactions. As the hydration progresses, the porosity of the concrete matrix decreases, leading to changes in its electrical conductivity. By monitoring these changes, electrochemical sensors can provide real-time data on the hydration progress and the development of concrete strength. Electrochemical sensing is advantageous due to its relatively low cost and the potential for continuous in-situ monitoring. This method has been shown to correlate well with traditional measures of hydration, such as compressive strength and calorimetry data. However, the accuracy of electrochemical sensors can be affected by environmental factors, such as temperature and moisture content, necessitating careful calibration and interpretation of results^40,41.

Concrete compressive strength

Concrete compressive strength is a fundamental property that is intrinsically linked to the hydration process of cement. During hydration, cement particles react with water to form various hydration products, among which calcium silicate hydrate (C–S–H) is the most significant. C–S–H forms a dense matrix that is largely responsible for the strength and durability of the concrete. Therefore, controlling the hydration process, and consequently predicting the development of concrete strength, is essential for ensuring the quality, safety, and longevity of concrete structures. Compressive strength is defined as the maximum compressive stress that a material can withstand before failure. For concrete, this property is particularly critical as it is a composite material that excels in compression but is relatively weak in tension. The compressive strength of concrete is a primary determinant of its load-bearing capacity and, as such, is a crucial factor in the design and construction of various types of infrastructure, including buildings, bridges, dams, and pavements. In structural engineering, the compressive strength of concrete directly influences the safety and stability of structures. A thorough understanding of this property allows engineers to design structures that are not only safe but also optimized in terms of material use and cost. Accurate assessment and control of compressive strength are therefore fundamental aspects of engineering practice (Fig. 2), particularly in ensuring that structures meet the necessary safety standards and performance requirements^24,42. Several key factors influence the compressive strength of concrete, and these must be carefully controlled to achieve the desired properties in the finished material. Understanding these factors allows for the optimization of concrete mix design and curing conditions, leading to improved strength and durability. The composition of the concrete mix, particularly the proportions of cement, water, aggregates, and admixtures, plays a critical role in determining compressive strength. The water-to-cement (w/c) ratio is particularly important; a lower w/c ratio typically results in higher compressive strength due to reduced porosity in the cement matrix. However, an excessively low w/c ratio can lead to poor workability and incomplete hydration, potentially compromising the concrete's strength and durability. Additionally, the selection of aggregates—both in terms of size and grading—can influence the density and overall strength of the concrete^23,24. Different types of cement exhibit varying chemical and mechanical properties, which in turn affect the hydration process and the development of compressive strength. For instance, Portland cement, the most commonly used type, is known for its relatively fast hydration and early strength gain. In contrast, blended cements, which incorporate supplementary cementitious materials such as fly ash, slag, or silica fume, may hydrate more slowly but often result in higher long-term strength and durability. The choice of cement type depends on the specific requirements of the project, including the desired strength, durability, and environmental conditions^43,44. The inclusion of mineral additives and chemical admixtures in the concrete mix can significantly modify its properties, including compressive strength. Mineral additives, such as fly ash, ground granulated blast furnace slag (GGBFS), and silica fume, react with the calcium hydroxide (Ca(OH)₂) produced during hydration to form additional C–S–H, thereby increasing the overall strength of the concrete. Chemical admixtures, such as superplasticizers, accelerators, and retarders, are used to control the setting time, improve workability, and enhance the development of strength under various curing conditions^45,46. Curing conditions, including temperature, humidity, and time, are crucial factors that influence the hydration process and, consequently, the development of concrete strength. Optimal curing conditions ensure that the hydration process is allowed to proceed to completion, maximizing the degree of hydration and minimizing the risk of early-age cracking. Elevated temperatures, while accelerating hydration and early strength gain, can also lead to thermal cracking if not properly managed. Conversely, inadequate curing, particularly in dry or cold environments, can result in incomplete hydration, leading to reduced compressive strength and increased permeability²⁷.

Fig. 2.

Concrete compressive strength is tested using a compression machine on cube samples (producer: JXSC).

Importance of real-time monitoring and predictive modeling

Given the complex interplay of factors influencing concrete compressive strength, real-time monitoring and predictive modeling have become essential tools in modern construction practices. Traditional methods of strength assessment, such as destructive testing of concrete specimens (e.g., compressive strength tests on cylinders or cubes), while providing accurate results, are inherently time-consuming and often introduce significant delays into construction schedules⁴⁷. These standard tests typically require that concrete specimens be cured under controlled conditions and tested at specified intervals, commonly at 7, 14, or 28 days after casting⁴⁸. This waiting period means that critical construction decisions, such as formwork removal, prestressing, or loading of the structure, must be postponed until sufficient strength is confirmed⁴⁹. The delays caused by traditional testing methods can have several adverse consequences. Extended construction timelines due to waiting for test results can lead to higher labor costs, equipment rental fees, and overhead expenses^4,49. Holding up subsequent construction activities while awaiting strength confirmation can significantly inflate project budgets. The inability to proceed with critical path activities can delay project completion dates, impacting the overall schedule and potentially leading to penalties for late delivery⁴. This can be particularly problematic in projects with tight deadlines or those that are sensitive to time constraints. Proceeding without confirmed strength data may lead to premature loading of concrete elements, risking structural failures or compromising the safety of workers and the public⁵⁰. Conversely, unnecessary delays in formwork removal or load application can lead to inefficiencies and increased risks associated with prolonged use of temporary supports. In contrast, real-time monitoring of the hydration process, combined with advanced predictive models, offers a powerful approach to assess the early-age compressive strength of concrete. By integrating sensors and data analytics, it is possible to continuously track the hydration process and make informed decisions regarding concrete strength development⁵⁰. This approach allows for immediate feedback, enabling construction managers to optimize construction timelines, reduce costs, and enhance the safety and durability of structures⁵¹. The necessity for real-time predictive systems is underscored by the limitations of traditional methods. By providing accurate, timely predictions of concrete strength, real-time monitoring systems can mitigate the negative impacts of delays, ensuring that construction processes are efficient, cost-effective, and safe⁵².

Application of machine learning in predicting concrete strength

In recent years, machine learning (ML) techniques have emerged as powerful tools across various disciplines, including materials science and concrete technology. The adoption of machine learning in predicting concrete strength represents a significant advancement, offering the ability to analyze vast datasets, recognize complex patterns, and make accurate predictions that would be challenging or impossible to achieve through traditional empirical methods. Machine learning, a subset of artificial intelligence, focuses on developing algorithms and statistical models that enable systems to learn from and make decisions based on data without explicit programming for each specific task. The application of machine learning in predicting concrete strength spans various stages of concrete production and curing, from mix design to early-age strength estimation and long-term performance evaluation. This chapter discusses the key steps involved in applying ML to concrete strength prediction, reviews significant contributions in the literature, and outlines the challenges and future directions in this evolving field. Several key studies have demonstrated the potential of machine learning in predicting concrete compressive strength. These studies highlight the effectiveness of various ML algorithms and provide insights into their applications and limitations. Yeh⁵³ applied neural networks to predict concrete compressive strength, achieving high prediction accuracy. The work of Chou and Tsai⁵⁴ focuses on the application of support vector machines (SVM) to predict the mechanical properties of concrete. In more recent studies, such as those conducted by Deng⁵⁵, they analyze the application of deep learning algorithms in the context of modeling concrete properties. In a key study from 2019, Ziolkowski¹⁰ and his team presented an algorithm that accurately predicted concrete compressive strength based on mix composition. However, the algorithm had difficulties accurately predicting high-strength concrete (above 40 MPa) and predicting the properties of mixes containing additives and admixtures. This study highlighted the need for further improvement of models to better handle such challenges. Subsequent research by Ziolkowski⁹ in 2021 focused on adaptive machine learning methods that better handled the diversity of modern concrete mixes. This method utilized deep neural networks and was trained on a comprehensive database of concrete mix recipes, allowing for more precise strength prediction compared to earlier models. In 2023 he studied how computational complexity of deep neural networks affect the predictive capability of the models¹¹. Chen et al.⁵⁶ proposed a convolution-based deep learning approach for predicting the compressive strength of fiber-reinforced concrete (FRC) exposed to elevated temperatures. Their study addressed the limitations of traditional experimental and numerical methods, such as inefficiency, safety concerns, and poor generalizability. By redesigning a Convolutional Neural Network (CNN) to handle arbitrary input feature dimensions—including mix proportions, heating profiles, and fiber properties—they enabled the model to capture complex local feature patterns for enhanced prediction accuracy. The model was validated using an extensive experimental dataset and outperformed conventional machine learning methods and existing standards, such as Eurocode 2, in terms of accuracy and generality. The results demonstrated the model's capability to serve as a reliable tool for mixture design optimization and strength estimation, contributing to safer and more efficient concrete applications under high-temperature conditions. In 2024 Chen et al. introduced Tempnet⁵⁷, a deep learning-based model for assessing fire-damaged concrete by predicting temperature fields. The model combines a convolutional neural network (CNN) for extracting image features and a graph convolutional network (GCN) to capture temperature dependencies across adjacent areas. This innovative approach maps concrete surface color changes—resulting from fire exposure—to temperature fields, effectively estimating the damage level and depth. Tempnet was validated across three application scenarios: concrete surfaces, drilled core samples, and cross-section areas, with corresponding datasets meticulously constructed. Extensive experiments demonstrated its robustness and efficiency, achieving F1 scores above 0.97 in all cases, outperforming traditional methods. Case studies further confirmed Tempnet's reliability and practicality for real-world assessments. By leveraging deep learning, Tempnet accelerates and improves the accuracy of fire-damaged concrete evaluations, providing a scalable tool for structural assessment and post-fire repair planning.

Formwork rotation management

Formwork rotation management is a critical aspect of construction project management, particularly in the context of concrete structures. Formwork, as in Fig. 3, refers to temporary molds or structures used to hold and shape freshly poured concrete until it has gained sufficient strength to support its own weight. The management of formwork rotation is not only essential for maintaining the structural integrity and dimensional accuracy of concrete elements but also has significant implications for the overall efficiency, cost-effectiveness, and safety of construction projects. Formwork is indispensable in the construction of concrete structures, as it ensures that the concrete is held in the desired shape and alignment until it cures and hardens. The quality of the formwork directly impacts the final appearance, durability, and structural performance of the concrete. Inaccuracies in formwork placement or premature removal can lead to defects such as misalignment, surface imperfections, or even structural failure. Therefore, precise and effective management of formwork is crucial to achieving the desired outcomes in concrete construction^49,58.

Fig. 3.

Formwork panels on a construction site.

Benefits of effective formwork rotation management

Effective formwork rotation management offers several significant benefits, contributing to the overall success of a construction project. One of the primary advantages of efficient formwork rotation is the potential for substantial cost savings. Formwork systems represent a significant portion of the overall construction budget, particularly in large-scale projects. By optimizing the rotation and reuse of formwork, construction teams can minimize the number of formwork sets required on-site, thereby reducing the costs associated with renting or purchasing additional formwork. Additionally, efficient formwork management reduces the need for formwork storage and maintenance, further lowering operational costs⁵⁹. Optimizing formwork rotation can also lead to significant reductions in construction time. By ensuring that formwork is efficiently cycled between different sections of a construction site, downtime is minimized, and the overall pace of construction is accelerated. This continuous workflow helps to prevent delays and ensures that each phase of the construction process is completed as quickly as possible. Moreover, faster construction timelines contribute to earlier project completion, which can have financial benefits, particularly in projects where time is a critical factor^60,61. The quality of concrete elements is directly influenced by the management of formwork. Properly managed formwork systems ensure that concrete is poured, shaped, and cured under optimal conditions, leading to higher-quality finishes and more accurate dimensional tolerances. This reduces the likelihood of defects that require costly and time-consuming repairs or adjustments. Furthermore, high-quality formwork systems are designed to be durable and reusable, maintaining their integrity over multiple uses, which further enhances the consistency and quality of the finished concrete elements⁶².

Analysis of cost-effectiveness of implementing a proposed framework

In the competitive landscape of modern construction, efficiently managing processes and resources is crucial for minimizing costs, shortening project timelines, and maximizing return on investment (ROI). This chapter presents a comprehensive analysis of the cost-effectiveness of implementing a real-time system for monitoring and predicting early concrete strength, with particular emphasis on its impact on formwork rotation management. The analysis covers three representative construction project types, an office building, a bridge, and a residential complex, and includes quantitative data on construction time reductions, detailed savings calculations, and clear return-on-investment measures. These insights are supported by specific examples drawn from case studies, illustrating the tangible economic benefits associated with the system’s implementation. These calculations are part of the business analysis of our solution, performed in 2021 by the Technology Transfer Center of the Gdańsk University of Technology, still in the pre-implementation phase. To thoroughly assess the cost-effectiveness of the proposed system, several critical metrics were evaluated across the three project types:

• Total Project Cost: Represents the planned total expenditure for each project, encompassing all phases, from design and planning to completion.

• Cost of System Implementation: Includes the expenses associated with the monitoring system, such as the cost of hardware (sensors, communication modules), software integration, and necessary training for construction personnel. For the cases analyzed, the initial implementation costs ranged from 0.35 million EUR (office building) to 0.66 million EUR (bridge), reflecting differences in project scale and complexity.

• Savings from Formwork Rotation Management: Financial savings achieved through the optimization of formwork rotation, enabled by the accurate and timely predictions of concrete strength provided by the monitoring system. These savings stem from the reduced need for multiple formworks sets and the efficient reuse of existing formwork. For the analyzed cases, the initial implementation costs ranged from 0.35 million EUR (office building) to 0.66 million EUR (bridge), reflecting differences in project scale and complexity.

• Savings from Reduced Construction Time: This metric quantifies the cost reductions that result from shortening the overall project duration. By optimizing construction schedules and ensuring timely progression through various construction stages, the project incurs fewer costs related to labor, equipment rental, and overheads.

• Increased Work Efficiency: The implementation of the monitoring system can enhance work efficiency by enabling workers and machinery to operate more effectively, reducing idle time, and increasing productivity. This metric captures the additional financial benefits from improved labor and resource utilization. For instance, improved scheduling and resource deployment contributed an additional 0.06 million EUR in savings for the office building and up to 0.33 million EUR for the residential complex.

• Total Savings: The cumulative effect of improved formwork management, shorter construction timelines, and enhanced efficiency yielded total savings ranging from 0.30 million EUR for the office building to 1.16 million EUR for the residential complex. These figures reflect the integrated financial benefits realized through better decision-making supported by real-time data.

• Payback Period: The payback period is the time required for the savings to equal the initial investment in the monitoring system. It was calculated by dividing the total cost of system implementation by the annual financial savings. The payback periods ranged from 14 months (office building) down to just 6 months (residential complex), indicating rapid ROI in large-scale projects.

Comparative analysis of construction projects

The analysis compares the economic impact of the proposed monitoring system across three types of construction projects: an office building, a bridge, and a residential complex. The key findings are summarized in Table 1 and illustrated in Fig. 4.

Table 1.

Cost-effectiveness comparison of different construction projects.

Type of construction	Total project cost (million EUR)	Cost of system implementation (million EUR)	Savings from formwork rotation (million EUR)	Savings from reduced construction time (million EUR)	Increased work efficiency (million EUR)	Total savings (million EUR)	Payback period (months)
Office building	11.0	0.35	0.04	0.20	0.06	0.30	14
Bridge	33.0	0.66	0.11	0.55	0.22	0.88	9
Residential complex	44.0	0.55	0.17	0.66	0.33	1.16	6

Fig. 4.

Total savings (in million EUR) and the payback period (in months) for three types of construction projects.

The analysis demonstrates that the implementation of a real-time concrete strength monitoring system is highly cost-effective across all types of construction projects. Larger projects, such as the bridge and residential complex, exhibit particularly strong financial returns, with payback periods as short as six months. The total savings range from 0.30 million EUR for the office building to 1.16 million EUR for the residential complex, highlighting the substantial financial benefits associated with improved formwork rotation management and reduced construction times. The rapid rotation of formwork, facilitated by accurate and timely predictions of concrete strength, brings several key benefits that contribute to the overall cost-effectiveness of construction projects. Effective formwork rotation significantly shortens the time required to complete construction projects. Case studies indicate that optimized formwork management can reduce construction times by 15–30%⁴⁹. This reduction in time translates directly into cost savings, as projects are completed more quickly and resources are used more efficiently. Faster formwork rotation reduces the costs associated with renting or purchasing additional formwork sets. Efficient management of formwork can yield savings of 10–20% of total construction costs, as fewer formwork sets are required, and existing sets are reused more effectively⁴⁹. The ability to quickly move formwork from one section of a project to another allows workers to proceed to the next stages of construction without unnecessary delays. This increased efficiency results in better utilization of labor and equipment, further enhancing productivity and reducing costs^58,63. The analysis clearly indicates that the implementation of a real-time concrete strength monitoring system offers significant economic benefits across various types of construction projects. By enabling more efficient formwork rotation, reducing construction times, and improving work efficiency, the system not only contributes to substantial financial savings but also enhances overall project management, safety, and quality. Looking forward, the continued development and refinement of these monitoring systems, particularly in terms of integrating advanced data analytics and machine learning algorithms, hold the potential to further optimize construction processes. As construction projects become increasingly complex and demand greater precision, the role of real-time monitoring technologies in ensuring cost-effectiveness and operational efficiency will only continue to grow.

Materials and methods

This section presents a comprehensive methodology for developing and implementing the proposed system for real-time prediction of early-age concrete compressive strength. The technical requirements for the device are first outlined, encompassing the selection and configuration of temperature sensors, communication modules, database servers, and computing units. Each component is designed to function seamlessly under harsh construction conditions, thereby ensuring accurate data acquisition and reliable communication. Subsequently, the data generation process is described, detailing the methodologies applied for real-time temperature monitoring within concrete structures. The procedures for data preparation are then explained, including normalization, feature extraction, and labeling, all of which are critical steps for training the predictive AI model. The model training methodology is discussed in depth, covering the selection of the deep neural network architecture, the chosen training parameters, and the evaluation metrics employed to assess predictive performance. These considerations ensure that the model can accurately estimate early compressive strength across various scenarios. Finally, the procedures for device design and manufacturing are detailed, including assembly, installation, and testing protocols. These protocols verify that the system can operate effectively under real-world construction conditions. By adhering to this methodology, the resulting system is rendered robust, accurate, and adaptable to diverse construction environments, ultimately enhancing operational efficiency and safety. Figure 5 provides an overview of the proposed framework that illustrates the structural layout and interactions among the individual system element.

Fig. 5.

Flowchart of proposed framework with major components. CM – communication module with temperature sensors.

Technical requirements and system architecture

The proposed system for real-time monitoring of concrete strength is configured as a sophisticated network of interconnected devices, each contributing to the precise and timely provision of data on the hydration process. This arrangement is essential for improving efficiency and accuracy within construction workflows, particularly for optimizing formwork rotation and managing overall project timelines. The system architecture comprises several key components, including temperature sensors, communication modules, a database server, and a central computing unit. These components have been integrated to ensure reliable data acquisition, transfer, and analysis.

System components

Temperature Sensors: Each temperature sensor in the system is a high-precision device designed to measure the temperature of the concrete during the curing process. Accurate temperature measurement is vital because the rate of concrete hydration, and consequently its strength development, is temperature-dependent. Each sensor is assigned a unique hardware identification number, which enables precise localization within the system. This unique identifier allows for accurate tracking and management of a large number of sensors distributed across different formwork panels and construction zones. The use of uniquely identified sensors ensures that each temperature reading can be traced back to its specific location, which is essential for creating accurate temperature profiles across different sections of the construction project. This granularity of data is critical for detecting any anomalies or inconsistencies in the curing process that could impact the quality and strength of the concrete^52,64.

• Communication Modules: The communication modules serve as the intermediary between the temperature sensors and the central database server. Each communication module is also equipped with a unique physical address, allowing it to be individually identified and managed within the system. The primary function of the communication module is to collect temperature data from the connected sensors, aggregate this data, and transmit it to the database server for storage and analysis. The design of the communication modules ensures robust and reliable data transmission, even in environments with significant electromagnetic interference, which is common on construction sites. By assigning each communication module to a specific set of formwork panels, the system can efficiently manage data flow and minimize the risk of data loss or corruption⁶⁵.

• Database Server: The database server is the central hub for data management within the system. It is responsible for receiving, storing, and organizing the temperature data transmitted by the communication modules. The server architecture is designed to handle large volumes of data, ensuring that information from potentially hundreds of sensors can be processed and stored in real-time. One of the critical decisions in the system design was to eliminate the need for a local system clock within each communication module. Instead, the database server assigns a timestamp to each data packet upon receipt. This approach simplifies the system architecture and avoids potential synchronization issues that could arise from discrepancies between multiple local clocks. Given that the delay from data measurement to database recording is negligible relative to the inertia of the temperature values and the response time of the sensors, this method maintains high accuracy in time-stamping, ensuring the integrity of the data^66,67.

• Computing Unit: The computing unit is tasked with the real-time analysis of the collected data. This unit processes the temperature readings, applies predictive models, and generates insights regarding the hydration process and the developing strength of the concrete. The computing unit must be capable of handling complex computations, including the application of machine learning algorithms, which are used to predict concrete strength based on the temperature data collected. Given the critical role of the computing unit in decision-making, it is designed with redundancy and failover mechanisms to ensure continuous operation even in the event of hardware or software failures. This ensures that the system can provide consistent and reliable data, which is crucial for maintaining the quality and safety of the construction process⁶⁸.

Data flow and system synchronization

The flow of data within the system is designed to be as streamlined and efficient as possible. Temperature sensors continuously monitor the concrete and transmit their readings to the communication modules. The communication modules, in turn, aggregate this data and send it to the database server, where it is time-stamped and stored. The decision to forgo local clocks within the communication modules was driven by the need to maintain synchronization across the entire system. Time-stamping data directly at the server ensures that all data points are synchronized with a single reference time, eliminating potential errors that could arise from discrepancies between multiple clocks. This approach is particularly important in environments where precise timing is crucial for accurate data analysis and decision-making⁶⁹. By centralizing the time-stamping process, the system avoids the complexities and potential errors associated with clock drift and synchronization across a distributed network of devices. The resulting data integrity ensures that the system can reliably monitor the hydration process and provide accurate predictions of concrete strength, ultimately enhancing the quality and efficiency of construction projects.

Device design and manufacturing

The design and manufacturing of a device for monitoring the concrete hydration process necessitate a rigorous approach to ensure that the system meets the high standards required for accuracy, reliability, and durability in the challenging environment of construction sites. This chapter outlines the key steps involved in the design and manufacturing process, including detailed analysis of requirements, component selection, system design, assembly procedures, and testing protocols. The aim is to develop a robust system capable of providing real-time data on concrete hydration, thereby enhancing the efficiency and safety of construction operations.

Requirements analysis

The first and most critical step in the design process is conducting a comprehensive analysis of the technical and operational requirements of the system. This analysis forms the foundation for all subsequent decisions regarding component selection and system design. The temperature sensors selected for the system must be capable of accurately measuring the range of temperatures expected during the concrete hydration process. Typically, this range spans from ambient temperatures to the higher temperatures generated during exothermic hydration reactions. Accurate temperature measurement is crucial for predicting the development of concrete strength and ensuring that the concrete cures under optimal conditions⁷⁰. High measurement accuracy is essential to obtain reliable data for analysis. The selected sensors must have a high resolution and low error margin to detect even minor fluctuations in temperature, which can significantly impact the hydration process and, consequently, the strength and durability of the concrete^65,66. The device must be designed to operate reliably under the harsh conditions commonly encountered on construction sites. This includes exposure to humidity, dust, vibrations, and variable temperatures. The selected components must be robust and capable of withstanding these conditions without compromising their performance⁶⁹. Efficient and reliable communication between the sensors, communication modules, and the database server is critical for the system's success. The communication protocols must ensure data integrity, minimize latency, and support the real-time transmission of large volumes of data across the network⁶⁴.

Component selection

Based on the requirements analysis, the next step involves the careful selection of components that meet the identified criteria. The selection of temperature sensors is driven by the need to balance accuracy, durability, and cost. Sensors must be chosen based on their ability to measure the required temperature range with high accuracy, as well as their resistance to environmental factors such as moisture and dust. Thermocouples and RTDs (Resistance Temperature Detectors) are commonly used due to their precision and robustness^71,72. Communication modules are selected for their ability to handle data from multiple sensors and transmit it efficiently to the database server. Wireless communication is often preferred due to its flexibility and ease of deployment in the dynamic environment of a construction site. Modules must be chosen based on their range, data transmission speed, and reliability under site conditions⁷³. The database server must have the capacity to store and manage large datasets generated by the temperature sensors. Scalability is also a critical factor, as the server must be able to accommodate increasing data volumes as the system expands. The server should be equipped with high-performance storage solutions and redundant systems to ensure data availability and integrity⁷⁴. The computing unit is responsible for analyzing the collected data and generating actionable insights. This unit must have sufficient processing power to handle complex algorithms, including predictive modeling and machine learning. Depending on the system's scale, this could range from a high-performance server-class computer to a dedicated computational system with advanced analytical capabilities⁷⁵.

System design and integration

Detailed electrical and communication diagrams are created to define all connections, power supplies, and grounding points. Housing designs are developed to protect sensors and modules against dust, moisture, and temperature fluctuations. The design also ensures ease of access for maintenance. Electrical and communication diagrams map out the connections between sensors, communication modules, the database server, and the computing unit. They provide a blueprint for the assembly of the system, ensuring that all components are correctly interconnected and that data flows efficiently through the system. The diagrams also include provisions for power supply, grounding, and protection against electrical interference⁷⁶. The design of housings and mounts is crucial for protecting the sensors and communication modules from environmental hazards. The housings must be durable, weather-resistant, and easy to install on-site. They should also allow for easy access to the sensors and modules for maintenance purposes. Mounting points must be carefully chosen to ensure that sensors are placed at key locations where temperature variations are most likely to occur⁷⁷. The integration process involves the development of user interface (UI) and system management software that allows for easy monitoring and control of the entire process. The UI should be intuitive, enabling users to view real-time data, generate reports, and configure system settings with minimal training. The system management software must ensure seamless communication between all components, support data synchronization, and provide tools for troubleshooting and diagnostics⁷⁸.

Assembly and installation

Once components are procured, they are assembled in accordance with the design specifications. Sensors are strategically mounted on formwork or embedded within the concrete at representative points. Communication modules are installed to ensure strong, reliable signals from sensors to the server. The database server and computing unit are configured and integrated into the system to manage data flow and perform analyses. Temperature sensors are mounted on the formwork at strategically selected locations where significant temperature variations are expected. The placement of sensors is critical to capturing a representative temperature profile of the concrete, which is essential for accurate strength prediction. Sensors must be securely attached to prevent movement during the curing process, which could affect the accuracy of the readings⁷⁹. Communication modules are installed in locations that optimize signal reception from the sensors and ensure efficient data transmission to the server. Each module is configured with a unique physical address, facilitating its identification and management within the system. The installation process must account for potential obstructions and sources of interference that could impact wireless communication⁸⁰. The database server and computing unit are configured to collect, store, and analyze the data transmitted by the communication modules. The server software is set up to assign timestamps to each dataset and ensure system-wide time synchronization. The computing unit is equipped with the necessary software for data analysis, including machine learning algorithms and predictive models that process the temperature data to estimate concrete strength⁸¹.

Testing and validation

Before full implementation on the construction site, the system should undergo rigorous testing under both laboratory and real-world conditions. These tests are designed to verify the system's operational accuracy, reliability, and robustness. Initial tests are conducted in a controlled environment to ensure that all components function correctly and that the system meets the specified accuracy and reliability standards. These tests include stress testing of the sensors, validation of data transmission integrity, and performance benchmarking of the database and computing units⁸². The system is then tested under actual construction conditions to validate its performance in a real-world environment. This testing phase assesses the system's ability to operate under varying environmental conditions, including temperature extremes, humidity, and dust. The reliability of communication between sensors and the server is also evaluated, as is the accuracy of the data analysis performed by the computing unit. Following successful testing, the system is ready for full deployment on the construction site. Continuous monitoring and regular maintenance are essential to ensure the system's long-term reliability. This includes periodic calibration of sensors, software updates, and inspection of the physical condition of the devices.

Data generation and preparation for model training

The methodology described here outlines the comprehensive workflow for the development of a predictive model for early-age concrete compressive strength using real-time temperature data. The process is systematically divided into four principal stages: data generation achieved through embedded sensor monitoring, preprocessing of the acquired data, development and training of artificial intelligence models, and deployment of the system within a construction environment. Emphasis is placed on methodological rigor, adherence to data handling protocols, and robust model-building practices. Notably, final predictive accuracy or quantitative evaluation outcomes are not disclosed.

Data generation: real-time temperature monitoring

To ensure accurate modeling, temperature data should be generated by monitoring the concrete’s curing environment in real-time. Sensors need to be strategically placed at various depths and locations to capture a comprehensive thermal profile influenced by factors such as ambient conditions, concrete mix design, and structural geometry⁸³. Data collection should focus on the early curing period, typically the first several days, when significant hydration-related temperature changes occur.

Data transmission, storage and preprocessing

Real-time data streaming from sensors to communication modules and subsequently to a centralized database server should be implemented. This minimizes latency and mitigates the risk of data loss⁸⁴. All collected temperature readings must be securely stored on the server, creating a robust dataset for further analysis. Preprocessing steps should involve both data cleaning and normalization. Cleaning the data entails identifying and removing outliers or erroneous readings, which may arise from sensor faults or communication issues.

Normalization involves applying scaling methods, such as min–max normalization, as in Eq. 5, to the raw temperature values. This ensures proportional contributions of features to the model, preventing any single feature from dominating the learning process⁸⁵.

5

Feature extraction, labeling and data splitting

Key features must be extracted from the raw temperature data to characterize the concrete hydration process. These features should include average temperature over time intervals, temperature gradients, and cumulative thermal measures, where:

• Average Temperature (T_avg): Mean temperature over specified time intervals.

• Maximum Temperature (T_max) and Minimum Temperature (T_min): To capture temperature extremes.

• Temperature Gradient (∆T/∆t): Rate of temperature change over time.

• Cumulative Degree Hours (CDH): Integral of temperature over time, calculated as in Eq. 6

  6

• Standard Deviation of Temperature (σ_T): To measure temperature variability.

In parallel, concrete samples cured under similar conditions should undergo laboratory testing to determine their compressive strength at early-age intervals (e.g., 1, 3, and 7 days). The extracted features should then be paired with the corresponding strength test results to create a labeled dataset suitable for supervised machine learning. The dataset should be divided into training, selection and testing subsets, commonly following an 60/20/20 ratio. This ensures that the model is trained on a substantial portion of the data and evaluated on unseen data, facilitating an assessment of its generalization capability beyond the conditions it was trained on⁸⁶.

Model training methodology and deployment

The core of the predictive system is the development and training of an AI model capable of accurately estimating the early compressive strength of concrete based on pre-processed temperature data. This process involves several well-defined stages, each contributing to the creation of a robust and reliable predictive framework.

Model architecture and training procedure

The development of the predictive system should begin with the selection of model architecture⁸⁷ as the core model for estimating concrete strength. The chosen architecture must be capable of capturing the complex relationships between temperature data and concrete strength. Based on experience from previous research^9–11 and analyses, it was decided that a deep neural network (DNN) would be best for this task. This network should consist of an input layer representing the extracted features, one or more hidden layers utilizing nonlinear activation functions to capture complex relationships within the data, and an output layer designed to provide continuous predictions of compressive strength. Hyperparameter tuning and optimization should be carried out to refine the model architecture. Parameters such as the learning rate, number of neurons in each layer, number of training epochs, and batch size should be systematically optimized using techniques like grid search or cross-validation. This process ensures a balance between achieving high predictive accuracy and maintaining generalization capability. The training procedure should involve iterative adjustments of the model’s parameters to minimize prediction errors⁸⁸. The network must be trained on a designated training dataset, with techniques such as early stopping employed to prevent overfitting by halting training when validation performance no longer improves. Throughout the process, loss metrics and other performance indicators should be monitored to ensure convergence to a stable and reliable solution. Once the model has been trained and its parameters optimized, its performance should be validated using a separate validation dataset^87,89. Finally, the model must be tested on a held-out testing dataset to evaluate its predictive accuracy and stability. This step verifies the model’s capability to reliably estimate early concrete strength under conditions similar to those encountered in real-world construction scenarios. Metrics such as mean absolute error, root mean square error, and R² should be used to confirm the robustness and generalization of the predictive system⁸⁹. This framework provides a structured and scientific methodology for developing a reliable AI-based model for predicting concrete compressive strength.

Framework for deployment in construction

The deployment of the predictive system should begin with its integration into a real-time monitoring framework. After thorough validation, the trained AI model must be embedded into a dedicated computing unit that interfaces seamlessly with the real-time temperature monitoring system. The continuous flow of temperature data from sensors should be fed directly into the model, enabling it to generate ongoing predictions of concrete strength. These predictions must be presented through an intuitive user interface to provide actionable insights for engineers and construction managers. This interface should support data-driven decision-making processes, such as determining optimal formwork removal timing, load application schedules, and adjustments to project timelines. To ensure the system's long-term reliability, periodic maintenance procedures must be established. This includes regular sensor recalibration to maintain data accuracy, software updates to incorporate improvements and address any system vulnerabilities, and routine checks to verify system integrity under varying site conditions. Additionally, the deployment framework should support the continuous refinement of the predictive model by incorporating newly collected data. This process enables the model to adapt to evolving conditions, including variations in concrete mixes, environmental factors, and construction practices, thereby enhancing its accuracy and versatility over time. This structured framework ensures that the deployment of the AI-based predictive system is both effective and sustainable, fostering improved decision-making and operational efficiency in construction environments.

Results and discussion

Limited laboratory testing of the prototype device

Laboratory tests were conducted on a small-scale prototype device to validate its basic functionality and performance prior to field deployment. The results demonstrated the system's ability to accurately measure temperature, reliably transmit data, and maintain stable data storage. Additionally, the preliminary performance of the machine learning predictions was verified using a controlled sample of concrete specimens, confirming the prototype's readiness for further testing in real-world conditions. Objectives of laboratory testing:

• Sensor Accuracy and Reliability: Validate that the temperature sensors provide stable, accurate temperature readings under controlled conditions representative of early-age concrete curing.

• Data Transmission and Synchronization: Confirm that the communication modules reliably transmit data to the database server without significant data loss or transmission delays.

• Data Integrity and Storage: Ensure proper timestamping and storage of incoming data in the database, with consistent synchronization and no observable data corruption.

• Preliminary Predictive Modeling Feasibility: Assess the basic functionality of the AI model by comparing model predictions to measured compressive strength of small-scale concrete samples at early ages, verifying that the framework is capable of producing plausible estimates.

The laboratory tests were conducted in a controlled environment, designed specifically to evaluate the functionality of the prototype device and validate its predictions. A small batch of concrete cubes (15 × 15 × 15 cm) was prepared, consisting of a total of eight specimens. The laboratory setup provided stable ambient temperature control and sufficient space for the mixing, casting, and curing of concrete samples. For the concrete specimens, a standardized mix design with known properties and expected compressive strength gain curves was used. The prototype device was set up as follows. Sensors were embedded in the four designated specimens immediately after casting. To ensure accurate readings, each sensor was carefully placed near the geometric center of the cube, maintaining direct contact with the fresh concrete. A communication module was installed within the laboratory to receive data from the sensors. For wireless transmission, the module was placed within range, and for wired connections, the sensors were properly linked to the module. Each sensor and communication module were configured with a unique identifier to facilitate reliable data collection. A database server and computing unit were set up using a dedicated laptop connected to the communication module. The system included a local SQL database for data storage and a pre-configured environment for the AI model. The server was programmed to assign accurate timestamps to each incoming data packet, ensuring the integrity and traceability of the collected data throughout the testing process. Figures 6, 7, 8, 9 and 10 present visual documentation of the prototype device, its laboratory setup, and the integration with the dedicated web application used for real-time monitoring and analysis.

Fig. 6.

The prototype device (communication module with sensors).

Fig. 7.

The prototype device (interior of the communication module).

Fig. 8.

Device setup.

Fig. 9.

View of modules in dedicated web application.

Fig. 10.

Temperature readings in web application.

The testing process began with initial calibration and baseline checks conducted on day 0 to ensure the accuracy and reliability of the sensors and data transmission system. Sensors were placed in a water bath maintained at a stable temperature of 20 °C ± 0.5 °C. Their readings were compared against those from a calibrated laboratory thermometer, with an acceptable deviation set at ± 0.5 °C. Additionally, a communication check was performed by streaming sensor readings from the water bath to the database server for a duration of at least 30 min. This step verified the correct frequency of data reception, accurate timestamping, and the absence of data loss. Following successful calibration and verification, the sensors were embedded in the concrete cubes, and continuous data acquisition began. During the early hydration phase, spanning the first 72 h, temperature data was recorded continuously, focusing on the critical early-age period when hydration-induced temperature changes occur rapidly. Laboratory conditions were kept stable, with an ambient temperature of approximately 20 °C. To ensure cross-verification, a separate reference thermometer was placed near the specimens to monitor ambient temperatures. Throughout the data acquisition phase, validation checks were conducted to confirm that all embedded sensors transmitted data successfully to the communication module and that the database server logged each reading accurately. The data flow was monitored to ensure no intervals were missed, and the temperature progression was analyzed to verify it aligned with expected hydration behavior, such as an initial rise within the first 24 h. To eliminate potential problems with the uncertainty of the battery life of the sensors (tests have shown that the current prototype can operate on a battery for 3 days), each of the four temperature-measuring prototypes was permanently connected to a power source, by micro-USB cable connected directly to the internal socket. Every specimen has four measurement points from which the mean value was calculated. Compressive strength testing was performed at specific early-age intervals, namely at 1 day, 3 days, and 7 days, using a standard compression testing machine. Strength measurements were recorded for both the sensor-embedded and control specimens, providing ground truth values for comparison. Selected key records has been shown in Table 2.

Table 2.

Selected temperature and concrete strength data.

Cube ID	Age (h)	Measured temperature [avg.] (°C)	Measured compressive strength (MPa)
C1-A	24	25.0 ± 0.5	9.2
C1-B	72	21.4 ± 0.5	18.9
C1-C	168	20.6 ± 0.5	29.7
C2-A	24	25.3 ± 0.5	9.4
C2-B	72	21.5 ± 0.5	19.3
C2-C	168	20.3 ± 0.5	30.2
C3-A	24	24.9 ± 0.5	8.9
C3-B	72	21.3 ± 0.5	18.7
C3-C	168	20.3 ± 0.5	29.5
C4-A	24	25.1 ± 0.5	9.1
C4-B	72	21.6 ± 0.5	18.8
C4-C	168	20.4 ± 0.5	29.9

The collected temperature data was utilized to train the predictive model, for series A, B and C – 188, 288 and 670 datapoints were collected respectively, for every measured specimen arithmetic average from four thermometers readings were calculated. Compressive strength values were interpolated for the values of temperatures between intervals. Compounded number of records for training were 2680 records. The data was split as follows, training set (60%), selection set (20%) and testing set (20%). The machine learning model use time in hours and temperature readings in Celsius degrees to predict early concrete compressive strength in megapascals. Details are presented in Fig. 11.

Fig. 11.

Scatter plots – (A) Concrete compressive strength vs. time and (B) Concrete compressive strength vs. temperature.

Model architecture and training details

A Deep Neural Network (DNN) was selected for its ability to model complex nonlinear relationships between input features and the target variable. Architecture details:

• Input Layer: 2 neurons (time in hours and temperature readings in Celsius degree)

• Hidden Layers:

  First Hidden Layer: 2 neurons with Hyperbolic Tangent (tanh) activation function.

• Second Hidden Layer: 1 neuron with linear activation function.

• Output Layer: 1 neuron to predict the continuous target variable (compressive strength in MPa).

• Feature scaling and bonding layers has been used

Loss Function: Normalized Squared Error (NSE) was chosen as the loss function to minimize the difference between the predicted and actual strength values.

Optimization Algorithm: The quasi-Newton algorithm was used for its efficiency.

Hyperparameters:

• Learning Rate: Set to 0.001, providing a balance between convergence speed and stability.

• Number of Epochs: The model was trained over 37 epochs, allowing sufficient iterations for convergence.

• Early Stopping: Implemented to prevent overfitting, stopping training if validation loss did not improve over 10 consecutive epochs.

Hyperparameters were selected based on initial experiments and tuning using grid search and cross-validation to achieve optimal performance without overfitting. Training and selection errors are presented in Fig. 12.

Fig. 12.

Training and selection errors.

Model validation and testing

After training, the model's performance was validated using the testing dataset. This step involves feeding the test data into the model and comparing the predicted strength values against the actual values obtained from laboratory tests. Performance metrics used:

• Mean Squared Error (MSE): Measures the average squared difference between predicted and actual values.

• Root Mean Square Error (RMSE): The square root of MSE, providing error units consistent with the target variable (MPa).

• Coefficient of Determination (R²): Indicates the proportion of variance in the dependent variable predictable from the independent variables.

The model's performance on the test dataset is summarized as follows:

• MSE: 0.021 MPa²

• RMSE: 0.143 MPa

• R² Score: 0.996

Figure 13 comparing predicted compressive strength to measured values for the test specimens showed a tight clustering around the 1:1 line, indicating strong predictive accuracy. Relative error distribution was presented in Fig. 14. The residual errors did not exhibit any systematic bias relative to curing time or temperature range.

Fig. 13.

Predicted vs. actual concrete compressive strength values.

Fig. 14.

Relative error distribution.

Evaluation criteria

The evaluation of the prototype system was based on several key performance aspects, encompassing sensor performance, data transmission and logging, data integrity and preprocessing, and the preliminary feasibility of the predictive model. Sensor performance was assessed by verifying the accuracy and stability of temperature readings. Accuracy was confirmed by ensuring deviations from reference measurements were within ± 0.5 °C. Stability was evaluated by examining the data for erratic fluctuations that could not be attributed to known hydration behavior. Reliable sensor performance was essential for collecting meaningful data during the curing process. Data transmission and logging focused on the completeness and accuracy of the recorded data. Over 95% of the expected data points were required to be successfully transmitted and logged on the server. Additionally, all data points were checked for correct timestamping in chronological order, ensuring no temporal drift occurred. Data integrity and preprocessing involved verifying the absence of unexpected outliers in the dataset. If outliers were detected, appropriate data cleaning steps were applied to either remove or flag them. Normalization techniques, such as min–max scaling, were confirmed to have been correctly implemented, with normalized features aligning with expected ranges to maintain consistency in data representation. Preliminary modeling feasibility was evaluated by analyzing the outputs of the predictive model. Although high accuracy was not expected at this stage, the predicted compressive strengths were checked to ensure they fell within a plausible range, avoiding negative or unrealistically high values. Furthermore, the model’s predictions were examined for trend consistency, ensuring they followed the expected progression of compressive strength development over time as hydration advanced.

Key advantages of the system

The performance of the prototype system was thoroughly evaluated against established criteria, focusing on sensor performance, data transmission and logging, data integrity and preprocessing, and the preliminary feasibility of the predictive model. The results highlighted several key advantages of the system. The temperature sensors embedded within the concrete specimens demonstrated high accuracy and stability, with deviations from reference measurements remaining within the specified ± 0.5 °C range. This level of precision ensured the reliability of the collected data, which is critical for monitoring hydration processes where temperature fluctuations play a significant role in strength development. The stability of the sensors, with no evidence of erratic fluctuations unrelated to hydration behavior, further validated their suitability for consistent and meaningful data collection. Data transmission and logging were similarly robust. Over 95% of the expected data points were successfully transmitted to and logged on the central database server, meeting the completeness criterion. Additionally, all data points were correctly timestamped in chronological order, with no evidence of temporal drift. This ensured that the recorded data accurately reflected the progression of the curing process and could be used effectively for subsequent analysis and modeling. In terms of data integrity and preprocessing, the system successfully identified and managed any unexpected outliers in the dataset. Data cleaning processes were effective in either removing or flagging such anomalies, maintaining the integrity of the dataset. Normalization techniques, including min–max scaling, were correctly applied, ensuring that the normalized features aligned with expected ranges and provided consistent inputs for the predictive model. The preliminary feasibility of the predictive model was demonstrated through its ability to generate compressive strength predictions within a plausible range. While high accuracy was not anticipated at this early stage, the model consistently avoided producing negative or unrealistically high values. Furthermore, the predicted strength values followed the expected progression over time, closely aligning with the general trends of compressive strength development as hydration advanced. This trend consistency validated the model’s ability to respond logically to variations in the input data and provided a strong foundation for further refinement. Overall, the system performed exceptionally well across all evaluation criteria, highlighting its potential as a reliable and effective tool for real-time monitoring and prediction of concrete hydration and strength development. These results underscore the system's readiness for further optimization and eventual deployment in practical construction settings.

Areas for improvement

While the system demonstrated significant strengths, the testing phase also revealed several limitations that must be addressed to improve its reliability and overall performance. One notable issue was related to wireless communication between the communication modules and the database server. Interference and signal range limitations were observed, particularly in areas with high levels of electromagnetic interference or where significant distances separated the devices. These conditions led to unreliable data transmission, resulting in occasional gaps in the collected data. Such inconsistencies can compromise the accuracy of strength predictions, as uninterrupted data flow is essential for reliable modeling. To mitigate these issues, additional access points and signal amplifiers were deployed during testing. Although this solution improved transmission stability, it increased the system’s complexity and overall cost. Future iterations of the device may benefit from the integration of more robust wireless communication technologies or alternative methods of data transmission to ensure consistent performance across varied environments. Power supply challenges with some temperature sensors also emerged during the testing phase. Interruptions in data collection occurred due to suboptimal power source placement and the limited battery capacity of the sensors. These interruptions hindered continuous monitoring, which is critical for accurately capturing the hydration process and ensuring the predictive model’s effectiveness. Addressing these power supply issues will require the implementation of more efficient power management strategies, such as incorporating longer-lasting batteries, solar-powered sensors, or energy-harvesting technologies. Reliable power supply solutions are essential for maintaining uninterrupted data collection and enhancing the system's robustness. Another area of concern was the use of connectionless data transmission protocols. While this approach minimized delays between measurement and data transmission, it also increased the risk of data loss, particularly in conditions of weak signal strength. Any gaps in the transmitted data can adversely affect the precision of the predictive model's results. Signal monitoring was implemented for each module to address this issue, enabling operators to track and resolve transmission problems in real-time. This improvement enhanced data reliability, but further refinements in transmission protocols and signal strength management will be crucial to fully eliminate data loss and improve overall system performance. These limitations highlight areas for further development, including improving wireless communication reliability, optimizing power supply systems, and enhancing data transmission protocols. Addressing these issues will significantly strengthen the system’s capability to operate effectively in diverse and challenging construction environments.

Limitations and challenges

Despite the demonstrated potential of machine learning (ML) models in predicting the compressive strength of concrete, several inherent limitations and practical challenges remain. These factors must be acknowledged and addressed to ensure that the application of ML in this domain yields robust, reliable, and scalable results.

Data requirements

One of the primary challenges is the need for large, high-quality, and diverse datasets. Achieving high predictive accuracy is contingent upon training models on comprehensive datasets that cover a wide range of mix proportions, material sources, curing conditions, and environmental factors. Data scarcity or insufficient variability can lead to models that perform well on specific datasets but fail to generalize beyond them. In practice, this necessitates extensive collaboration with industry partners, concrete producers, and testing laboratories, as well as the continuous integration of newly collected data. Standardization in data reporting formats and the adoption of open-access data repositories could further alleviate these issues, enabling the broader research community to compare results more effectively and accelerate model improvement.

Overfitting and model complexity

Overfitting, in which a model memorizes patterns specific to its training dataset rather than learning generalizable relationships, remains a common hurdle. Without careful oversight, complex ML models—especially those with many parameters—may inadvertently tailor themselves to nuances in training data that do not represent broader trends. We mitigated this risk in our approach through the incorporation of techniques like cross-validation, regularization, dropout layers (for neural networks), and early stopping criteria. These methods help ensure that the model learns underlying principles rather than just fitting noise. Nonetheless, the constant evolution of ML methods and new regularization strategies requires ongoing vigilance and experimentation to maintain robust predictive capabilities.

Generalization across diverse conditions

Ensuring that predictive models can generalize effectively to conditions beyond their original training scenarios remains a critical challenge. Variations in cement types, aggregate characteristics, admixtures, and environmental conditions can significantly influence the mechanical properties of concrete. A model trained exclusively on data from a specific geographic region or a limited set of conditions may experience degraded predictive performance when applied in different contexts. Addressing this challenge requires strategies to enhance model adaptability, such as transfer learning, domain adaptation, or hybrid modeling approaches that integrate first-principles engineering knowledge with data-driven methods. Future research should focus on developing and evaluating models capable of handling such variability. One promising direction is the creation of consensus modeling frameworks, which integrate multiple models fine-tuned for specific conditions, thereby enhancing robustness and adaptability across heterogeneous environments. Additionally, datasets capturing greater diversity in concrete mix designs, curing conditions, and environmental settings are essential. This includes incorporating data from mixes with different cement types and admixtures, curing conditions with varying temperatures and humidity levels, and environmental settings ranging from controlled indoor laboratories to outdoor field sites. To assess generalizability, robust cross-validation techniques must be implemented to evaluate model performance across these diverse scenarios. Demonstrating consistent predictive accuracy under varied conditions would provide strong evidence of the model's applicability and reliability in real-world settings. Addressing these aspects in future research will significantly expand the scope and impact of predictive systems, providing more reliable tools for diverse construction applications.

Conclusions

This study has presented a comprehensive framework for real-time prediction of early concrete compressive strength by integrating continuous monitoring of the cement hydration process with a custom artificial intelligence (AI) model. The developed system successfully combines sensor technology, wireless communication, and advanced AI algorithms to provide accurate strength predictions. Empirical results demonstrated high predictive accuracy, with an R² value of 0. 996 and an RMSE of 0.143 MPa, indicating the model's reliability and effectiveness in practical applications. The implementation of this system offers significant practical benefits for the construction industry. By providing immediate and accurate predictions of concrete strength, construction managers can make informed decisions regarding formwork removal, load application, and scheduling of subsequent activities. This real-time insight enhances construction efficiency by reducing unnecessary delays, optimizing resource allocation, and potentially shortening project timelines. Moreover, accurate strength prediction contributes to improved safety by ensuring that structural elements meet required strength criteria before proceeding, thus preventing premature loading that could compromise structural integrity. The system also promotes higher quality in concrete structures by enabling better control over the curing process and early detection of potential issues. A key achievement of this work is the successful integration of real-time temperature monitoring with an advanced AI prediction model. The system leverages continuous temperature data collected from embedded sensors within the concrete to feed into the AI model, which then predicts the concrete's compressive strength in real time. This integration addresses the limitations of traditional laboratory methods by providing immediate, accurate predictions, thus optimizing the production cycle of reinforced concrete elements. The system's modular design allows for scalability and adaptation across various project sizes and types. For small projects, such as residential buildings or minor infrastructure works, a simplified version of the system can be deployed with fewer sensors and basic predictive capabilities, offering a cost-effective monitoring solution. In contrast, for large-scale projects like high-rise buildings, bridges, or tunnels, a distributed sensor network with advanced analytics can handle extensive data and provide detailed insights. Customization of the AI model by retraining with project-specific data ensures accurate predictions tailored to unique concrete mixes, environmental conditions, and structural requirements. This adaptability enhances the system's applicability and value across the industry. While the current system demonstrates significant potential, several areas for future development have been identified to further enhance its functionality, reliability, and effectiveness. Enhancing wireless communication technologies is paramount; researching communication methods with higher resistance to interference, such as LoRaWAN, Zigbee, or 5G networks, could improve data transmission reliability, especially on large construction sites or in challenging terrain. Employing low-power communication protocols can also significantly extend sensor battery life, contributing to the overall sustainability of the system. Improving power management for sensors is another critical area. Developing more efficient energy management methods, such as using solar panels, energy harvesting technologies (e.g., capturing energy from vibrations or movement), or advanced long-life batteries, can address the challenge of powering sensors in harsh construction conditions. Introducing intelligent energy management systems that automatically adjust energy consumption based on operational needs can increase the system's autonomy and reliability. Expanding the AI model's capabilities is essential for broader applicability. Training the model on a more extensive and diverse dataset—including various climatic conditions, types of concrete, and other contextual variables—can improve its generalizability and predictive accuracy. Applying more sophisticated AI methods, such as deep learning or reinforcement learning, may further enhance the model's adaptability to changing construction conditions. Integrating the system with existing construction management systems, such as Building Information Modeling (BIM) or Enterprise Resource Planning (ERP), would enable automation of monitoring and reporting processes, increasing efficiency and transparency. Real-time data and strength predictions can facilitate dynamic schedule adjustments, optimizing construction workflows and resource allocation. This integration would significantly enhance the system's practicality and ease of use in the industry. Conducting tests in diverse climatic and geographical conditions—including extreme temperatures, high humidity, or significant weather variability—is crucial for assessing the system's effectiveness and robustness. Insights from these tests can guide adaptations and optimizations tailored to specific local conditions, ensuring reliability across a wide range of environments. Such testing will validate the system's performance under various scenarios, reinforcing its suitability for global application. Developing a modular system architecture allows for easy adaptation to specific project requirements, whether monitoring small structural elements or large, complex engineering projects. Tailoring the system for residential buildings, bridges, high-rise structures, and other contexts demonstrates its versatility and scalability. This modularity ensures that the system can be efficiently customized and deployed across different construction contexts, maximizing its utility and impact. In conclusion, the developed system represents a significant advancement in construction technology, offering a practical, data-driven solution for real-time prediction of concrete compressive strength. By enhancing construction efficiency, safety, and quality, the system addresses critical industry needs. Future enhancements focused on communication technologies, power management, AI capabilities, and system integration will further strengthen its performance and broaden its applicability. The modular design and adaptability across different construction contexts position the system as a valuable tool in advancing smarter, more efficient construction practices, ultimately contributing to the evolution of the construction industry.

Author contributions

A.M. design of electronical backbone of the device, A.M. device design research, P.Z. design of machine learning workflow, P.Z. and A.M. funding acquisition, P.Z. and A.M. patent acquisition, P.Z. writing of the original manuscript P.Z., A.M. and S.G.G. review and editing. All authors reviewed the manuscript.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare no competing interests.