As an indispensable basic material in civil engineering, concrete has increasingly important environmental and resource challenges. High energy consumption is required for conventional concrete production, with considerable greenhouse gases being emitted. Costs are powered by resource shortage, which makes sustainability concerns aware. These challenges are expected to be addressed by Geopolymer concrete (GPC). The theoretical basis for GPC was founded in 1979 when the “Geopolymer” concept of Davidovits was proposed [1]. The reaction mechanism of geopolymers, as shown in 1 [2]. Chemical bonds in aluminosilize minerals are initially disturbed by alkaline activation. The minerals dissociate in [SiO4]4- Tetraedrons and [AlO4]5- Tetraedrian monomers. As the resolution progresses, the dimer formation is initiated by a monomer interactions by structural dismantling. Then polymers are generated by reactions between dimers and additional monomers. These polymers species are subjected to structural rearrangement and ultimately polycondense to amorphine sodium aluminosilize hydrate gel (Nash) with three-dimensional network configuration. The disorganized configuration, the high specific surface and the abundant micropores enable effective pore adsorption and filling, which give geopolymers superior mechanical properties and durability [3]. The usage channels for industrial waste are significantly expanded by the development of the GPC. Environmental -loaded by -products are converted into high -quality building materials [4]. In addition, GPC production significantly reduces greenhouse gas emissions, with the reported Co.2 Emissions typically 60 % -80 % lower than that of OPC [5]The significant potential for reducing global warming.
However, concrete structures in the service are inevitably exposed to complex environmental factors, under which the FT cycle is recognized as a critical mechanism that induces a deterioration in durability. Under cold climate zones or severe seasonal temperature fluctuations, pore water freezes in concrete at low temperatures, which creates expansive tensions by 9 % volumetric expansion. The subsequent thaw induces contraction stresses when melting ice cream. This repetitive FT cycling damages the internal microstructure and ultimately leads to a deterioration in strength and durability impairment [6]. In addition, the surface scaling is pronounced with increasing FT cycles, a phenomenon that is explained by the adhesive scaling theory [7]. The existing research system on the FT resistance of GPC distinguishes OPC both in the chemical composition and in microstructural properties.
Existing research has a relatively comprehensive characterized FT resistance performance in GPC [8]Present [9]Present [10]Present [11]. Zhao et al. [8] Systematically examined the influence of the slag (GGBFS) dose (10 %, 30 %, 50 %) on the FT durability of GPC. The results showed that an increase in the GGBFS content significantly improved the FT resistance, with 50 % GGFBS-containing GPC 225-foot cycles comparable to OPC. The critical FT endurance threshold was observed after 125 cycles. SEM observations showed an initiation of microesses in 125 cycles and crack network formation at 225 cycles in GPC-50. Wu et al. [9] Improved FT performance of GPC by including fiber fiber reinforcement polymer powder (GfK) as a partial replacement powder. Two formulations were developed. GFRP/GGBFS -based and GFRP/Flugasche (FA) -based GPC. All samples showed a mass loss of less than 5 % after 300 feet. The endurance limit rose from 200 to 300 cycles for GGBFS-based GPC and from 150 to 175 cycles for GPC to FA. Gfrp/GGBFS composite materials showed a superior FT resistance that was due to its homogeneous and compact microstructure. Enes Ekinci et al. [10] Improved frost resistance of GPC by nano-Silica (NS), Micro-Silica (MS) and styrene-butadiene latex (SBL) -ED. The results show that individual NAOH-activated copies have 23.6 % higher compressive strength than NA2Siio3 + Naoh -activated counterparts. Optimal doses were identified as 2 % NS and 5 % SBL, while MS 5 % for NA2Siio3 + NAOH systems and 3 % for only NAOH systems. N/A2Siio3 + NAOH -activated rehearsals showed a better FT resistance of 18.4 % after 300 cycles than just NAOH samples. Shima Pilehvar et al. [11] examined the influence of microcapaltic phase change materials (MPCM) on the physical and mechanical properties of GPC and OPC on Pozzolan-based GPC and OPC under FT conditions. The results showed that the deterioration of concrete induced by FT cycles, which were mainly induced from interface micro-cracks between the cement-capable materials, aggregates and MPCM. The addition of MPCM significantly improved the FT resistance of GPC, which led to a lower reduction in compressive strength, while the strength of samples without MPCM decreased significantly. At 0 ° C, the initial setting time of Portland cement paste was extended, while the paste was shortened by Pozzolan -based paste. Increasing the MPCM content has effectively reduced the final period of both pastes. Existing relevant studies focused on the optimization of the FT resistance of GPC using modification methods such as the use of minerals, nanoadditives or phase change materials. These studies have shown the properties of internal crack spread and pore development within the material through microstructure analysis that provide ideas and methods for the design and microstructure analysis of GPC under FT conditions. Building on these foundations, this research constructs a simulation model on the MESO scale based on pore structure development in order to further investigate the laws of GPC strengthening.
The numerical simulation in the mesoscala serves as a powerful analytical instrument to characterize the damage development mechanisms and the prediction of the technical performance of materials. Peng et al. [12] Developed a mesoscala model to analyze the mechanical reactions of concrete under FT state and showed the effectiveness of the model when recording frost-induced mechanical degradation, whereby the maximum aggregate size had a limited influence on the FT resistance. On this basis Liang [13] Advanced this approach by adding the deterioration of the main voltage strength into the modeling of Mesosskala and a computing framework was determined that simulates FT-induced damage patterns and spatial strength distribution. This model was used to analyze the multixial stress behavior of FT-related concrete, which proves to be effective when identifying damage modes for the evaluation of durability.
With the deepening of the studies on GPC, artificial intelligence algorithms have proven to be powerful tools with unique advantages. In order to reduce the experimental workload and optimize resource efficiency, Cao et al. [14] Used three approaches for machine learning. Support the vector machine (SVM), Multilayer Perceptron (MLP) and XGBOOST (XGB) to predict GPC pressure strength. The XGB model showed a superior predictive accuracy with a determination coefficient (RR2) of 0.98, significantly exceed SVM (0.91) and MLP (0.88). The errors (MAE, MSE, RMSE) from XGB were small, which were 1.49, 3.16 and 1.78. The contribution of each parameter was also evaluated by sensitivity analysis. A Thao Huynh et al. [15] Developed machine learning methods (Ann, DNN, reset) to predict the compression strength of GPC. And the performance was rated by R2RMSE and MAPE. The sensitivity analysis showed that the FA/Agregate ratio in particular influenced the pressure resistance. The residual model is optimal in predictive accuracy. Khoa Tan Nguyen et al. [16] Compare DNN and reset when predicting pressure strength properties of FA-based GPC. The model was trained and verified by 335 sets of experimental data. The results showed that reset had a better performance, R 0.9927, RMSE 1.2687 and MAE 0.5536. Mohamed AbdellaTief et al. [17] Used models Random Forest (RF), support vector regression (SVR) and extreme gradient boosting (XGB) to estimate the pressure capacity of geopolymer concrete (UHPGPC) of ultra-hole performance. Based on 128 data modeling, the analysis showed that the XGB model achieved the highest predictive accuracy (RR2 is over 0.84) and the steel fiber content and the ratio of liquid to binder (L/B) in particular influence the pressure strength of UHPGPC.
In the case of individual use for FT damage research by GPC, the analysis of the microma duct, the modeling of MESO standards and the deep learning learning show each of the three methods certain restrictions. The microscala analysis can record exactly information such as the crackback path and pore features, but it has difficulty quantifying the correlation between damage development and reducing the performance of macroscala during FT processes. The modeling of MESO standards is strongly based on input parameters. However, the most important parameter data for GPC are scarce than that for OPC that require experimental data to support modeling and validation. Although Deep Learning has predictable skills, it suffers from a scarcity of high -quality data records. Currently, data that cover the micro macro correlation of the GPC-FT performance is extremely limited, and model forecasts are missing the interpretation of the physical mechanism. It is therefore necessary to combine experiments in order to clarify the causes of the strength closure, which means that the integration of the three technologies is particularly important. Therefore, GPC samples were manufactured in this study and FT cyclical tests were carried out in order to measure the change in compressive strength under various FT cycles. SEM and MIP were used to follow the development of pore features and the microstructural dismantling during the FT cycles. A three-dimensional numerical simulation model in the mesoskaly measurement-sensitive simulation model of GPC under FT conditions was determined. Based on the results of the numerical simulation model, the correlation analysis of strength disorder factors was carried out. Three deep learning models (BPNN, Ann, CNN) were used to predict the strength of the GPC under FT conditions. The predictive performance was evaluated, whereby the optimal model was validated by case studies, which confirmed the reliability of the methodology. This research approach not only realizes the scale bridging from microscala to macroscala, but also the data shortage in the GPC modeling effectively, but also improves the physical interpretability of research consequences by the synergy of several methods and overcomes the limits of traditional individual research methods. This study determines an effective prediction framework for the GPC strength under FT conditions, whereby environmentally friendly material applications are brought up in extreme climate zones and sustainable constructions are supported.