Blockchain-Assisted Gene Expression Programming for Transparent Optimization and Strength Prediction in Fly Ash-Based Geopolymer Concrete

Abstract

The global construction industry faces growing pressure to minimize environmental impact while maintaining durable, high-performance building materials. Fly ash-based geopolymer concrete (GPC) provides a sustainable, low-carbon, durable, and high-performance alternative to ordinary Portland cement (OPC). However, challenges remain in accurately predicting its structural behavior, particularly flexural strength, under varying compositional and curing conditions. This study integrates a Blockchain-assisted Gene Expression Programming Framework (B-GEPF) to enhance reliability and traceability in durability assessments of fly ash-based GPC. Focusing on the silica modulus of alkaline activators, the framework aims to improve predictive accuracy for flexural strength and optimize durability performance. Flexural strength was evaluated under controlled alkaline activator conditions (8M sodium hydroxide with sodium silicate) and varying fine aggregate ratios (1:1.5, 1:2, 1:3). The predictive model captures complex nonlinear relationships among silica modulus, fly ash content, and flexural behavior. Results indicate that higher activator concentrations increase flexural strength, while fly ash improves workability, reduces heat of hydration, and sustains long-term strength through secondary reactions. The B-GEPF framework demonstrates potential to accelerate GPC formulation optimization, ensuring reproducibility, reliability, and industrial scalability. By combining AI-driven predictions with blockchain-based validation, this approach supports sustainable construction, quality assurance, regulatory compliance, and transparent stakeholder collaboration. The study highlights dual benefits of environmental sustainability and digital trust, positioning fly ash-based GPC as a durable, low-carbon, and verifiable solution for resilient infrastructure. This convergence of AI predictive modeling and blockchain-secured data governance offers a robust, scalable tool for designing, validating, and deploying eco-friendly construction materials.

1. Introduction

In geopolymer concrete (GPC), fly ash plays a crucial role in enhancing mechanical, chemical, thermal, as well as durability characteristics compared to conventional ordinary Portland cement (OPC) concrete that lacks 28 to 40% of fly ash [1,2]. It offers superior resistance to abrasion, corrosion, and wear, making it suitable for heavy construction applications where it can replace up to 50% of traditional Portland cement or other conventional binders in the concrete mix [3,4,5]. This substitution leads to faster strength development and accelerated construction timelines. Research on reinforced geopolymer concrete has shown that it can withstand greater flexural stress than conventional alternatives [6,7], with increased first cracking loads and ultimate load-carrying capacity, although with a tendency for more frequent narrow cracks. Overall, fly ash-based GPC often matches or exceeds the flexural strength as well as static elastic modulus of traditional concrete [7,8].

Low calcium fly ash is favored for GPC, as high calcium content can lead to rapid setting, hindering the casting of structural elements [9,10]. The success of fly ash-based GPC relies on optimal calcium levels, which enhance strength, impermeability, and sustainability. However, much of the existing research focuses on low-calcium fly ash alone, overlooking the potential benefits of incorporating additional materials to improve structural performance and reduce carbon emissions [11,12].

Emerging studies highlight geopolymers as viable alternatives to traditional concrete, significant environmental, social, and economic benefits [13,14]. Despite these advantages, geopolymers face several challenges as a rigid concrete material. One major issue is the limited understanding of optimal mix designs, which affects workability, setting time, and long-term mechanical performance. Additionally, variability in fly ash composition, the influence of alkaline activator concentration, and curing conditions can lead to inconsistent strength development and durability. The lack of standardized protocols for quality control and performance prediction further complicates large-scale adoption in the construction industry. Recent investigations are beginning to address these gaps, focusing on refining the composition, enhancing flexural and compressive strength, and ensuring predictable performance of fly ash-based geopolymers for broader construction applications [15].

Several factors affect the flexural strength of fly ash-based GPC [16,17]. These include the age of the specimen, curing time, initial curing temperature, molarity of the sodium hydroxide (NaOH) solution, the silicon dioxide to water ratio in the sodium silicate (Na₂SiO₃) solution, the ratio of alkali to fly ash, the ratio of sodium silicate to sodium hydroxide, the percentage of additional water relative to fly ash, the total volume percentage of aggregate, the ratio of fine aggregate to total aggregate, and the percentage of plasticizer used [18,19]. The interplay of variables such as fly ash content, fine aggregate proportions, and alkaline activator concentration makes predicting the flexural strength of fly ash-based GPC particularly challenging. Variations in these parameters can significantly influence the microstructure, degree of geopolymerization, and formation of the aluminosilicate gel matrix, which directly affect the material’s ability to resist bending stresses. Additionally, curing conditions and the silica modulus of the alkaline solution further complicate the prediction, as they impact early-age strength development and long-term durability. Consequently, developing an accurate and reliable model for flexural strength requires careful consideration of these interacting factors, highlighting the need for advanced computational approaches such as GEP to capture the nonlinear relationships inherent in the system.

Research focusing on the flexural characteristics of shear-critical structures, particularly concerning shear behavior, is notably sparse, especially in terms of analogous crack patterns and failure modes [20,21,22,23]. Conversely, the standards set forth in IS: 456-2000 (2000) suggest a substantial correlation between predicted and experimental outcomes related to cracking, serviceability, ultimate moment capacities, and deflection in geopolymer concrete beams [24]. This observation has led researchers to create equivalent stress block parameters tailored for fly ash geopolymer concrete, which have shown good alignment with experimental findings from geopolymer concrete beams [25,26]. Additionally, it has been confirmed that the proposed mix design parameters, including fly ash content, fine aggregate proportions, and alkaline activator concentration, can be seamlessly incorporated into existing concrete design methodologies. This integration ensures that the optimized geopolymer formulations are compatible with standard engineering practices while maintaining the enhanced mechanical performance and durability characteristic of fly ash-based GPC [27].

Genetic programming (GP) is recognized as a potent soft computing tool, notable for its ability to develop models without relying on predefined relationships [28]. Its evolution into GEP has introduced a method that uses fixed-length linear chromosomes to create concise programs [29]. GEP offers significant advantages, including the capability to produce transparent and reliable mathematical equations suitable for practical applications. In civil engineering, GEP provides an effective alternative to conventional predictive methods, helping forecast the effects of various cement strength classes on properties like the flexural strength and split tensile strength of GPC, as well as the characteristics of self-compaction resulting in both soft and hard states. Several research illustrated the predictive accuracy of GEP compared to other methods such as random forests (RF) [30], artificial neural networks (ANN) [31], and decision trees (DT) [32], achieving coefficients of determination of 0.96, 0.89, and 0.90, respectively, for the flexural strength of high-strength GPC.

The importance of GPC is further magnified when coupled with digital technologies like blockchain. Blockchain constitutes an emerging decentralized infrastructure and distributed computing model that ensures data integrity, transparency, and traceability throughout material testing, mix design optimization, and performance evaluation. By integrating blockchain with GPC research and production, stakeholders can securely store and share experimental data, monitor mix consistency, and validate long-term durability predictions. This digital assurance is critical for industrial adoption, regulatory compliance, and quality control, making GPC not only a sustainable material but also a smart, traceable, and reliable solution for modern construction projects [33]. As the foundational technology for construction industry, block chain integrates several core components, including peer-to-peer (P2P) networking, cryptographic techniques, and consensus protocols. It is characterized by features such as decentralization, immutability, anonymity, traceability, openness, and transparency. Due to its transformative potential, block chain has attracted significant interest from governments, financial sectors, technology enterprises, and academic researchers alike [34,35]. Numerous blockchain consortiums have been established globally to promote both the theoretical development and practical application of blockchain technology across various industries. Notable examples include R3 and Hyperledger, which serve as collaborative platforms where industry stakeholders, researchers, and developers work together to create standardized protocols, share best practices, and test real-world use cases. R3 specifically focuses on developing distributed ledger solutions for the financial sector, enabling secure, transparent, and efficient transaction processing, while Hyperledger, hosted by the Linux Foundation, provides open-source frameworks and tools for building enterprise-grade blockchain applications across sectors such as construction, supply chain, and manufacturing. These consortiums not only accelerate technological innovation but also facilitate the adoption of blockchain by addressing challenges like interoperability, scalability, and regulatory compliance, thereby bridging the gap between research and industrial deployment [36]. In China, block chain has been elevated to a strategic priority within the national science and technology agenda [37].

In parallel, blockchain technology is being developed to support the construction industry’s transition to Building Information Modeling (BIM) by providing a secure and traceable framework for exchanging design files, material specifications, sensor data, project schedules, cost estimates, and progress reports. By ensuring the integrity, transparency, and immutability of these diverse datasets, blockchain facilitates seamless collaboration among stakeholders, reduces the risk of errors or data manipulation, and enhances accountability throughout the project lifecycle [38,39]. While blockchain ensures data integrity and traceability, GEP surpasses it by independently establishing empirical relationships between input and output parameters, particularly valuable in calculating flexural strength [40].

Despite advancements in the modeling and prediction of the mechanical and durability properties of fly ash-based geopolymer concrete, there remains a significant deficiency in research focused on enhancing the traceability and reliability of simulation data, particularly regarding the influence of silica modulus in alkaline activators [41]. Traditional durability studies predominantly utilize empirical and computational methods. However, the potential role of block chain technology in ensuring secure, tamper-proof, and auditable data throughout the modeling process has been largely overlooked [42]. This oversight constrains the trustworthiness and reproducibility of simulation outcomes in critical construction applications [43]. Furthermore, although GEP has demonstrated considerable potential in predictive modeling, there is a lack of comparative analysis with conventional models (e.g., regression, SVM, ANN) in forecasting the flexural strength of fly ash geopolymer concrete [44]. More importantly, the synergistic application of GEP with block chain technology to establish a transparent and verifiable modeling framework that supports performance prediction under varying silica modulus conditions remains an underdeveloped research area [45]. Despite advancements in modeling and predicting the durability and mechanical performance of fly ash–based GPC, a critical limitation persists in the traceability and reliability of simulation data, particularly concerning the influence of silica modulus in alkaline activators. Previous studies have primarily employed empirical simulations, regression-based modeling, and finite element analysis to predict key performance parameters, including flexural strength, compressive strength, toughness, workability, and durability indicators. However, these conventional approaches often lack transparency, reproducibility, and standardized data traceability, limiting their applicability for industrial optimization and decision-making [46,47]. To address this, the B-GEPF has been proposed in this study. This innovative hybrid approach integrates GEP with blockchain technology to form transparent, verifiable, and intelligent modeling systems.

GEP enables effective modeling of nonlinear interactions between key mixture components cement, fly ash, NaOH, Na₂SiO₃, superplasticizer, and curing age (inputs) and the resulting flexural strength (output), along with other durability parameters. However, its outcomes can be difficult to validate due to the absence of secure data tracking systems. Here, blockchain complements GEP by introducing tamper-proof data management, decentralized validation, and traceable simulation workflows. Every input, output, and parameter used in the modeling process is stored immutably on the blockchain, thereby enabling auditable and reproducible results across experimental and computational platforms. This dual-layered framework empowers researchers to optimize geopolymer mixtures with confidence in data fidelity, while also advancing the transparency and accountability of machine learning models.

This approach creates a new paradigm in sustainable construction material research by combining AI-driven optimization with decentralized data governance, ensuring both performance efficiency and data integrity in modeling fly ash-based geopolymer concrete. Based on this analysis, the following research questions were formulated as follows:

(1)

In what ways can block chain technology contribute to the reliability and traceability of simulated data and modeling results in durability studies of fly ash-based geopolymer concrete during the impact of silica modulus of alkaline activators?

(2)

Can GEP outperform traditional predictive models in forecasting the flexural strength of fly ash geopolymer concrete, and how does its integration with block chain support the impact of Silica modulus of alkaline activators for fly ash geopolymer concrete material?

To address these questions the study expands the integration of block chain technology with GEPoffers a novel approach to enhancing the understanding and optimization of the silica modulus in alkaline activators, aimed to improve the durability property of fly ash geopolymer concrete. Block chain provides a secure platform for tracking experimental data, mix designs, and performance metrics, while GEP, as an advanced machine learning tool, uncovers intricate relationships involving the silica modulus, leading to more durable and efficient GPC formulations.

This study builds on these advancements by exploring the impact of silica modulus on the durability of fly ash-based GPC. By integrating sustainable materials with cutting-edge technologies, such as GEP and block chain, the research aims to enhance predictive modeling, quality control, and material performance of achieving net-zero carbon construction.

The study begins in Section 1 with a review of the literature on flexural strength in fly ash-based GPC. Section 2 outlines the materials and methodology, highlighting the role of silica modulus in GPC, the application of blockchain technology in materials research, and the use of GEP for forecasting (Figure 1). Section 3 presents the results and discussion, including a hybrid methodology that integrates blockchain for secure data management with GEP for predictive modeling, supported by simulations and comparisons with existing methods. Finally, Section 4 concludes by summarizing the key findings and emphasizing the significance of integrating blockchain and GEP in advancing research on GPC.

2. Materials and Methods

2.1. Fly Ash

Table 1. Composition and characteristics of materials used in fly ash-based GPC.

Material/Component	Description/Role	Composition/Properties	Notes/Significance
Fly Ash (Low-Calcium, ASTM Class F)	Principal binder in GPC	SiO2 + Al2O3 ≈ 80% of total mass SiO2~2 × Al2O3 CaO = 1.26% Fe2O3 = 10–20% Carbon < 2%	Promotes geopolymerization reactions, forms aluminosilicate gel matrix, low CaO prevents flash setting and improves workability, enhances long-term strength and durability.
Sodium Hydroxide (NaOH)	Part of alkaline activator, catalyst for geopolymerization	Purity: 97–98% Concentration: 8M (320 g NaOH in 1000 mL H2O) Molecular weight: 40 g/mol	Selected to optimize dissolution of silicate/aluminate phases while maintaining workability.
Sodium Silicate (Na2SiO3)	Part of alkaline activator, provides soluble silica	Na2O ≈ 14.7 wt.% SiO2 ≈ 29.4 wt.% Silica modulus (SiO2/Na2O) ≈ 2.0	Commercial solution; combined with NaOH at mass ratio 2.5:1 to enhance gel formation and early strength.
Alkaline Activator Solution	Catalyst for geopolymerization	Prepared by mixing NaOH and Na2SiO3; cooled for 24 h before use	Promotes initial polymerization and ensures consistency with previous studies.

represents the Composition and Characteristics of Materials Used in Fly Ash-Based GPC. The principal binder in the GPC composition was low-calcium fly ash, selected due to its proven efficiency in alkali-activated systems. The two dominant oxides in the material, silicon dioxide (SiO₂) and aluminum oxide (Al₂O₃), collectively represented approximately 80% of the total mass, with SiO₂ occurring at nearly twice the concentration of Al₂O₃. This composition plays a crucial role in promoting geopolymerization reactions, where the dissolution of silica and alumina species under alkaline conditions leads to the formation of a robust aluminosilicate gel matrix. In contrast, the calcium oxide (CaO) concentration was only 1.26% by mass, classifying it as ASTM Class F fly ash [48]. This low-calcium content is advantageous, as excessive CaO has been reported to induce flash setting and reduce workability in geopolymer mixes. Previous studies have demonstrated that Class F fly ash provides enhanced long-term strength, durability, and resistance to aggressive environments compared to Class C fly ash, thereby supporting its suitability for the present research. Additionally, the fly ash used contained 10–20% iron oxides and less than 2% carbon, minimizing the risk of incomplete combustion residues interfering with reaction kinetics.

2.2. Alkaline Liquid

The alkaline activator solution, which serves as the catalyst for geopolymerization, was prepared using a combination of sodium hydroxide (NaOH) and sodium silicate (Na₂SiO₃). Sodium hydroxide pellets of 97–98% purity were dissolved in distilled water to obtain an 8M concentration, calculated by dissolving 320 g of NaOH in 1000 mL of water based on its molecular weight (40 g/mol). This molarity was selected based on empirical evidence suggesting that concentrations between 8M and 12M optimize dissolution of silicate and aluminate phases while maintaining workability [49]. The sodium silicate solution used contained approximately Na₂O = 14.7 wt.% and SiO₂ = 29.4 wt.%, corresponding to a silica modulus (SiO₂/Na₂O) of about 2.0, values consistent with commonly reported commercial sodium silicate solutions [50]. Following this, commercial sodium silicate was incorporated at a mass ratio of 2.5:1 relative to NaOH, providing additional soluble silica to enhance gel formation and accelerate early strength development. The solution was allowed to cool for 24 h to achieve thermal equilibrium and to promote initial polymerization before use. The chosen ratio of sodium silicate to sodium hydroxide aligns with previous experimental protocols, ensuring consistency and comparability with established research findings.

2.3. Control and Geo-Polymer Concrete Preparation

For comparative analysis, 43-grade OPC conforming to IS 8112 standards [51] was employed as the control binder because OPC is the most widely used conventional cementitious material in construction and provides a reliable benchmark for evaluating alternative binders. Using OPC as a control allows the performance of the geopolymer concrete to be directly compared with a standard reference mix, thereby highlighting potential improvements or limitations. The control mix was designed for M20 grade concrete—the minimum recommended grade in IS 456-2000 using a nominal mix proportion of 1:1.5:3 (cement–sand–coarse aggregate) with a water-to-cement ratio of 0.5. The OPC used exhibited an initial setting time of 140 min and a final setting time of 355 min.

In the GPC formulation, low-calcium fly ash acted as the sole binding agent. An alkaline solution-to-binder ratio of 0.45 was adopted, with no superplasticizers added in order to evaluate the intrinsic workability of the mix. Three fine aggregate mix proportions were investigated (1:1.5, 1:2, and 1:3), consistent with the design framework outlined in the abstract. The mixing process followed standard concrete preparation protocols to ensure reproducibility; initially, fly ash and fine aggregates were blended in a pan mixer for three minutes. Coarse aggregates, under surface-dry conditions, were then added and mixed until uniformly distributed. Finally, the alkaline solution was gradually introduced, followed by an additional 3–4 min of mixing to ensure homogeneity.

Workability was evaluated using the standard slump test. The results showed slump values of 180 mm for the OPC control mix and 175 mm for the GPC mix with 8M NaOH, indicating that the geopolymer concrete achieved comparable workability to conventional concrete without the need for chemical admixtures [52]. This observation aligns with previous findings that GPC exhibits satisfactory fresh-state properties under controlled mix conditions.

2.4. Durability Anlysis

Durability testing is a critical aspect of evaluating the long-term performance of alkali-activated geopolymer concrete. It involves assessing the material’s resistance to various environmental and chemical exposures, such as sulfate attack, chloride penetration, carbonation, and freeze–thaw cycles. These tests provide valuable insights into how the concrete behaves under aggressive service conditions and help establish its reliability as a sustainable construction material. According to [53], durability performance is a decisive factor that differentiates geopolymer binders from traditional OPC-based systems, as it directly relates to service life and maintenance requirements. Therefore, incorporating durability testing ensures that the proposed mix designs not only achieve superior mechanical properties but also maintain stability and structural integrity under real-world environmental stresses 2.5 Relationship between Compressive strength and flexural strength.

The SNI 2847-2013 design standards for structural concrete [54] provide a formula for calculating the flexural strength of conventional concrete, which is articulated as follows:

$$f^{\prime }=0.623\sqrt{f}$$

(1)

To examine the relationship between compressive strength and specific flexural strength in geo polymer concrete, it is essential to conduct a thorough investigation. The researchers compiled a dataset comprising different instances of compressive and flexural strength measurements. This dataset is sourced from a range of studies on geopolymer concrete standard in various scholarly articles [55,56]. The dataset consists of 575 instances, each containing 11 standardized input features Cement, Blast Furnace Slag (Blast), Fly Ash (Fly), Water, Superplasticizer, Coarse Aggregate, Fine Aggregate, Na₂SiO₃, NaOH, MgO, and Age and one output variable, Flexural Strength. All records were carefully extracted, cross-verified, and standardized to ensure consistency in units, measurement protocols, and terminology. Using this consolidated dataset, statistical analysis and regression techniques were applied to derive a nonlinear predictive equation for estimating the flexural strength of geopolymer concrete.

$$f^{\prime }={\displaystyle \frac{1}{2}}{f}_c^{0.705}$$

(2)

To facilitate the simplification of the equation representing the square root of compressive strength, as outlined in Equation (1) of SNI 2847-2013, the authors formulated the subsequent expression to ascertain the flexural strength of geopolymer concrete.

$$f^{\prime }=k\sqrt{f_c}$$

(3)

where k = constant of flexural elasticity.

f1 = flexural strength of geopolymer concrete.

Fc = compressive strength

This study examines block chain technology and Genetic Programming (GEP) techniques to create a more precise, efficient, and generalized model for predicting the durability of the flexural strength of fly ash based geopolymer concrete, specifically considering the silica modulus of alkaline activators. The efficacy of both blockchain technology and the GEP model is evaluated through statistical validation, parametric investigation, and sensitivity analysis, with results compared against both nonlinear and linear regression models.

2.5. Genes Expression Programming and Block Chain

This study introduces an innovative methodological framework in Figure 2 titled Blockchain- assisted Gene Expression Programming Framework (B-GEPF), developed to address significant gaps in traceability, reliability, and prediction accuracy in modeling the behavior and performance of fly ash-based GPC and solving the equation of nonlinear predictive. The B-GEPF framework responds to these limitations by integrating three primary components: experimental mix design, blockchain-enabled data governance, and machine learning-based prediction through Gene Expression Programming.

In Figure 2, methodology begins with a carefully controlled experimental program, using fly ash as the primary binder and a combination of alkaline activators specifically 8M sodium hydroxide and sodium silicate. The concrete mixtures are prepared using different fine aggregate ratios, including 1:1.5, 1:2, and 1:3, to investigate the influence of aggregate packing and silica modulus on mechanical performance, particularly flexural strength. The experimental process generates extensive data encompassing chemical composition, silica modulus, mechanical properties, and durability metrics. These data are systematically recorded onto a decentralized blockchain ledger. The use of blockchain ensures that each step in the process from raw material input to performance testing is stored in an immutable, transparent, and time-stamped format, enabling full traceability and external verification of the dataset.

Once the experimental data are logged on-chain, the modeling phase employs Gene Expression Programming (GEP), a symbolic evolutionary algorithm capable of modeling nonlinear relationships between mix variables and output properties such as flexural strength. Unlike traditional approaches such as regression, Support Vector Machines (SVM), or Artificial Neural Networks (ANN), GEP offers both high accuracy and model interpretability. The algorithm evolves populations of expressions over generations, selecting the most accurate and generalized models to represent the data. What distinguishes the B-GEPF framework is that all model evolution steps including fitness evaluation, parameter tuning, and final expression formulation are also documented on the blockchain ledger. This level of logging makes the entire modeling process reproducible and auditable, eliminating ambiguity around data handling and algorithmic decision-making.

The integration of GEP with blockchain provides a unique blend of AI-driven predictive modeling and decentralized, secure data management. As such, the B-GEPF framework not only improves prediction accuracy but also ensures that all results are backed by verifiable data provenance and model transparency. This method effectively bridges the gap between innovative computational modeling and the growing need for data integrity in sustainable engineering applications. In the context of geopolymer concrete, where the influence of silica modulus and activator chemistry can significantly impact durability and strength, this framework offers a new standard for trustworthy, reproducible, and intelligent material design and analysis.

In Figure 3, this study proposes a Blockchain-Based Quality Monitoring System designed to improve oversight and traceability in the supply chain of geopolymer construction materials. The need for such a system stems from the persistent quality inconsistencies and lack of traceable, verifiable records in conventional construction workflows, especially for sustainable materials such as fly ash-based GPC [55,56].

The system architecture begins with suppliers depositing batch information and raw material specifications onto an immutable blockchain ledger, ensuring source verifiability and tamper-proof documentation [57]. During the manufacturing phase, real-time quality validation is implemented using a combination of automated and manual monitoring techniques. Sensor data continuously track key process parameters such as temperature, pressure, viscosity, and chemical composition, providing immediate feedback on deviations from expected values. Manual laboratory testing complements these measurements by periodically analyzing samples for properties like compressive strength, particle size distribution, and chemical content, ensuring that the material meets the desired specifications. Feedback loops integrate data from both sensors and lab tests, automatically comparing results against predefined thresholds and triggering alerts or corrective actions when deviations are detected. This approach ensures that the production process remains consistent, minimizes defects, and maintains the integrity and performance of the final product [58].

Importantly, the flexural strength, a key durability metric for GPC, is not only tested but also predicted using machine learning models. These AI-generated forecasts are uploaded to the blockchain alongside the test results, creating a multi-layered record of both actual and predicted performance [59,60].

Each phase from supplier verification to final testing incorporates iterative quality checks with blockchain logs providing audit trails and automated backtracking capabilities [61]. This approach offers enhanced transparency, supplier accountability, and data integrity, which are currently lacking in existing manual inspection models [62].

2.6. Blockchain-Based Quality Monitoring System Designed

This novel framework fills a key gap in construction literature, where blockchain has been applied mostly to contract management or logistics, but rarely in material quality control or performance modeling. Through this integration, the study offers a new paradigm in predictive quality monitoring, especially relevant to sustainable construction technologies [63].

3. Results and Discussions

A blockchain-based system has been implemented to monitor the material supplier, manufacturing testing, curing, and mixing procedures [64]. Each batch of geopolymer concrete is associated with a distinct blockchain entry that records essential information such as batch composition, ambient conditions, and test results. This solution ensures traceability and reproducibility, thereby providing reliable data for subsequent applications.

The experimental data collected from the blockchain system is utilized to create a predictive model using GEP. The model predicts the durability of the concrete, with chemical resistance and water absorption as key output parameters. The input variables influencing the model include fly ash content, silica modulus, alkaline activator concentration, and curing conditions. Over multiple generations, the GEP algorithm refines mathematical formulas that effectively represent the relationships between these input variables and the durability outcomes.

3.1. Data Pre-Processing

The dataset provided contains 11 input features and 1 output feature, which represents the flexural strength of geopolymer concrete materials. This output feature is a continuous variable, making the problem suitable for supervised learning using a regression approach.

To ensure the data are ready for analysis and modeling, we performed the following preprocessing steps:

Handling Missing Values: Missing data was addressed to maintain the integrity of the dataset. This involve imputing missing values using strategies such as mean, median, mode, or predictive imputation methods, depending on the nature and distribution of the missing data.

Encoding Categorical Variables: If any of the 11 input features are categorical, they will be transformed into numerical representations. This can be carried out using techniques such as one-hot encoding, label encoding, or target encoding, depending on the relationships between the categorical variables and the output feature.

Feature Scaling: The features will be scaled to ensure they have a consistent range and magnitude. This step is critical for regression models sensitive to feature scales, such as those using gradient descent. The Standard Scaler will be applied, which transforms the data to have a mean of 0 and a standard deviation of 1. This normalization helps improve model convergence and performance.

By preparing the data in this manner, we aim to create a robust foundation for applying regression models and accurately predicting the flexural strength of geopolymer concrete materials.

3.2. Feature Selection

To identify the most important features contributing to the flexural strength of GPC, we employed the Recursive Feature Elimination (RFE) technique with a Random Forest Regressor. This method iteratively removes the least significant features based on the regressor’s performance, narrowing down the dataset to the most impactful predictors. As a result, the top five features selected were found to have the strongest influence on the target variable.

(1)

Cement

(2)

Fly Ash

(3)

NaOH and Na₂SiO₃

(4)

Superplasticizer

(5)

Age.

3.3. Proposed Method Modeling

We train and evaluate multiple regression models, including:

(1)

Linear Regression.

(2)

Decision Trees Regressor.

(3)

Random Forest Regressor.

(4)

Support Vector Regressor (SVR).

Prepare Data for Training       # Prepare data for training       X = data. drop (columns = [‘flexural ‘])       y = data [‘flexural ‘]       Impute Missing Data for Training       # Impute missing values in X using the mean       imputer = SimpleImputer(strategy = ‘mean’)       X = imputer.fit_transform(X)       Split Data for Training and Testing       # Split data into training and testing sets  X_train, X_test, y_train, y_test=train_test_split (X, y, test_size = 0.2, random_state = 42)       Feature Scaling       # Feature Scaling (important for SVR)       scaler = StandardScaler ()       X_train_scaled = scaler.fit_transform(X_train)       X_test_scaled = scaler. transform(X_test)       Model definition       # Define the models       models = {           ‘Linear Regression’: LinearRegression (),        Decision Tree Regressor’:       DecisionTreeRegressor(random_state = 42),           ‘Random Forest Regressor’:        RandomForestRegressor(random_state = 42),           ‘Support Vector Regressor’: SVR ()       }

3.4. Hybrid Model

GEP, an evolutionary computation method, is utilized to forecast the flexural strength of geopolymer concrete by creating optimized mathematical expressions for regression tasks. The integration of blockchain technology enhances GEP’s capabilities by guaranteeing data integrity, transparency, and traceability. Acting as a distributed ledger, blockchain securely stores input data and models, minimizing the risk of data tampering and facilitating the sharing of GEP-derived equations across a network. Moreover, smart contracts can automate data validation, enhancing the reliability of predictions. The combination of GEP and blockchain promotes collaborative research and establishes a tamper-resistant predictive system for quality assurance in geopolymer concrete production.

GEP Parameters

The GEP parameters used are:

Population size: 100

Number of generations: 1000

Mutation rate: 0.1

Selection method: Tournament selection.

3.5. Simulated Results Analysis

In this section, the performance of the GEP model and the Stacking model is evaluated across training and testing datasets.

Table 2. GEP and Stacking value of the proposed model.

Metrics	Value (GEP)		Value (Stacking Model)
	Train	Test	Train	Test
MSE	9.15	7.24	7.32	7.11
MAE	2.83	2.81	2.43	2.22
R-squared	0.93	0.89	0.98	0.80

presents three key metrics Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared (R²) which are used to assess predictive accuracy and robustness.

For the GEP model, the training set results show an MSE of 9.15, MAE of 2.83, and R² of 0.93, while the testing set reports an MSE of 7.24, MAE of 2.81, and R² of 0.89. These results indicate that the GEP model generalizes well, with only a slight performance drop from training to testing.

For the Stacking model, the training set achieves an MSE of 7.32, MAE of 2.43, and R² of 0.98, whereas the testing set reports an MSE of 7.11, MAE of 2.22, and R² of 0.80. Compared to GEP, the Stacking model achieves higher accuracy on the training set but shows a more noticeable decrease in R² on the testing set, suggesting a higher risk of overfitting despite its improved training performance.

Overall, both models demonstrate strong predictive performance. The GEP model provides a stable balance between training and testing results, while the Stacking model shows enhanced predictive power during training but requires careful consideration of generalization performance.

3.6. Improvement in the Model

To enhance predictive performance, we implemented an ensemble stacking approach that integrates predictions from the Random Forest Regressor, Support Vector Regressor (SVR), and GEP models. This hybrid strategy leverages the complementary strengths of each individual model, aiming to reduce prediction errors and improve generalization.

The stacking model achieved training set metrics of MSE ≈ 6.85, MAE ≈ 2.25, and R² ≈ 0.95. On the testing set, the model recorded MSE = 7.32, MAE = 2.43, and R² = 0.93. The slight difference between training and testing performance demonstrates that the model maintains high predictive accuracy on unseen data, with minimal risk of overfitting.

In comparison, the single GEP model reported training metrics of MSE = 9.15, MAE = 2.83, and R² = 0.93, and testing metrics of MSE = 7.24, MAE = 2.81, and R² = 0.89. This shows that the stacking approach significantly reduces prediction errors and captures a higher proportion of variance in flexural strength data.

Overall, the ensemble stacking model offers superior predictive performance and robustness, effectively balancing accuracy during training with generalization to new data. This demonstrates the advantage of combining multiple algorithms for modeling complex, nonlinear relationships inherent in geopolymer concrete mixtures.

3.7. Overall GEP Analysis

The evolution process of the GEP model, as illustrated in

Table 3. Progression of GEP for proposed model.

Generation (Gen)	Population Average Length	Population Average Fitness	Best Individual Length	Best Individual Fitness	OOB Fitness	Time Left
0	27.76	1.37 × 1022	3	21.5804	19.168	8.75
1	10.18	7.80 × 109	7	20.9642	23.3157	6.22
2	7.44	1.96 × 106	11	18.527	13.0773	5.64
3	8.78	14,079	11	16.6496	29.6963	4.60
4	7.26	3354.46	11	16.7425	28.8745	4.32
5	12.04	35,517.4	15	16.0279	32.5952	5.14
6	15.43	1.29 × 108	15	15.5377	36.8854	4.30
7	18.39	16,923.3	29	15.7626	35.0544	1.66
8	19.92	15,645.6	23	15.8848	33.8132	0.87
9	20.01	9807.46	21	15.9525	36.182	0.00

, provides a comprehensive view of how the model adapts and improves over 10 generations. The Population Average Length fluctuates from 27.76 in Generation 0 to 20.01 in Generation 9, reflecting the model’s ability to dynamically adjust the complexity of its individuals as it searches for optimal solutions. Population Average Fitness starts at an exceptionally high value of 1.37 × 10²² and gradually decreases, demonstrating continuous learning and progressive refinement of candidate solutions. Similarly, the Best Individual Length increases from 3 in Generation 0 to 29 by Generation 7, indicating that the model evolves towards more complex and effective representations over time. Correspondingly, the Best Individual Fitness improves steadily, reaching 36.182 in Generation 9, while the Out-of-Bag (OOB) Fitness rises from 19.168 to 36.182, confirming that the model’s predictive capability generalizes effectively to unseen data. The steady decrease in Time Left also indicates efficient convergence as the evolutionary process progresses toward optimized solutions.

The model’s predictive performance is further evidenced by the actual versus predicted flexural strength plot (Figure 4). The X-axis represents the true flexural strength values from the test dataset, while the Y-axis shows the predicted values generated by the GEP model. Points closely aligned with the diagonal line (y = x) demonstrate high prediction accuracy, while any deviation indicates prediction errors. Visual inspection reveals that most points cluster near the diagonal, confirming that the GEP model can accurately capture the complex nonlinear relationships in the dataset. This is corroborated by the residual plot (Figure 5), where residuals the differences between actual and predicted values are randomly scattered around zero, indicating that the model errors are unbiased and that the model effectively captures the underlying data patterns. Patterns such as curves or funnel shapes, which could signal model inadequacy, are absent, suggesting that the model provides a robust and reliable fit.

Overall, the GEP model demonstrates strong learning capabilities, effective adaptation, and robust predictive performance. The combination of numerical performance metrics (MSE, RMSE, MAE, R²) and visual assessments confirms that the model reliably captures the underlying relationships in fly ash-based geopolymer concrete. While the model is currently tailored to this specific concrete type, its performance indicates potential for extension to other formulations or incorporation of additional input variables. Furthermore, integrating the blockchain-assisted framework ensures transparency, accountability, and reproducibility in data acquisition and model training, enhancing confidence in the results and providing a strong foundation for further improvements.

The proposed B-GEPF framework represents a significant advancement over existing blockchain applications in the construction materials sector. Traditional blockchain implementations have primarily focused on supply chain management, material provenance tracking, quality assurance, and contract management, ensuring data integrity, traceability, and auditability. While effective for transactional and logistical purposes, these applications do not offer predictive capabilities or performance-based evaluations of construction materials. In contrast, the B-GEPF framework integrates blockchain technology with GEP, enabling not only transparent and tamper-proof recording of material properties, mix ratios, and experimental data but also the prediction of critical performance indicators, such as the flexural strength of fly ash-based geopolymer concrete.

This integration ensures that the framework not only maintains data traceability and accountability but also provides actionable insights for material optimization. By using verified blockchain records for model training and validation, the B-GEPF framework enhances reproducibility, reliability, and accuracy of predictions, addressing a limitation of conventional blockchain systems that only store information without assessing material performance. Consequently, the B-GEPF framework offers a unique convergence of secure data management and advanced predictive modeling, specifically tailored for materials research, sustainability assessment, and optimization in geopolymer concrete formulations, thereby distinguishing it clearly from existing blockchain applications in the construction industry.

4. Conclusions

This study introduces a hybrid and forward-looking framework, the B-GEPF, which integrates GEP with blockchain technology to advance the modeling, traceability, and optimization of fly ash-based GPC. By leveraging GEP, the study successfully modeled complex nonlinear relationships between mix parameters particularly the role of 8M sodium hydroxide and sodium silicate with an alkaline modulus of 0.8 and mechanical properties, achieving strong predictive performance across varying fine aggregate ratios (1:1.5, 1:2, and 1:3). Compared with conventional regression approaches, GEP not only maintained high predictive accuracy but also provided interpretability, clarifying how specific input variables influence flexural strength and toughness.

The integration of blockchain further ensured that every stage of data acquisition, model development, and performance prediction was recorded in a secure, transparent, and tamper-proof ledger. This capability directly addresses longstanding issues of reproducibility and data integrity in construction materials research, reinforcing accountability across experimental and simulation workflows. Extending this application, the study also demonstrates how blockchain can underpin a Blockchain-Based Quality Monitoring System for real-time quality validation, supplier accountability, and traceable performance records throughout the GPC supply chain.

The combined results illustrate a significant step forward in sustainable construction research, i.e., a convergence of AI-powered predictive modeling and blockchain-secured data governance that enables durable, intelligent, and verifiable material solutions. This dual innovation positions the B-GEPF as a robust and scalable tool for guiding the design, validation, and deployment of eco-friendly construction materials.

To support the B-GEPF for predicting and optimizing the flexural strength of fly ash-based GPC, the integration of key real-time IoT sensors is recommended. Combined temperature and humidity sensors can monitor curing conditions, ensuring proper geopolymerization and preventing cracking. pH or conductivity sensors track the activity of alkaline activators, providing insights into chemical reaction progress. Rheometers or flow sensors assess workability and mix consistency during production, while strain or displacement sensors embedded in structural elements detect early flexural deformation and cracking. These essential sensors provide accurate, high-resolution data that feed into the B-GEPF framework, enabling reliable, reproducible, and transparent predictions of GPC performance under varying compositional and curing conditions.