Performance, sustainability, and machine learning of UHPC with coarse aggregates: a critical review

Practical applications

This review aims to outline decision criteria or specifications for engineers and researchers to choose UHPC mixtures depending on the requirements of their specific performance, durability, sustainability, and economics. The findings indicate that, for large-scale structures, the use of coarse aggregates can reduce cement consumption while improving dimensional stability, economic feasibility, and environmental sustainability. It also pinpoints the need for optimizing the various parameters like the gradation of aggregate, dosage of Fibre, curing period, and cement content to enhance ITZ performance and crack resistance. Additionally, machine learning methods help with property predictions and rapid mix-design optimization of UHPC materials, cutting down on experimental work and speeding the process of establishing sustainable UHPC systems.

1 Introduction

Ultra-High-Performance Concrete (UHPC) is an advanced cementitious material characterized by high mechanical properties, high resistance to wear and tear, and high resistance to hostile environmental/dynamic loading factors (Gong et al., 2022). These are mainly realized by optimization of particle packing, a very low ratio of water to binder, large cementitious content, and incorporation of steel Fibres that gives a dense and homogeneous microstructure (Millon et al., 2009; Saqif and El-Tawil, 2025). This has led to increasing interest in UHPC in critical infrastructure applications in bridges, Défense, precast elements, and impact-resistant systems (Li et al., 2020; Xue et al., 2020; Wang et al., 2022; Kravanja et al., 2024).

Conventionally, the UHPC mixtures do not contain coarse aggregates to remove the weak interfacial transition zone (ITZ) and evenly distribute stress in the matrix (Yin et al., 2023; Shen et al., 2024). This approach improves tensile strength, ductility, and crack resistance because there is higher Fibre dispersion and matrix continuity (Li and Yu, 2019). But, due to the lack of coarse aggregates, cement usage is greatly increased, production is more expensive, and carbon emissions are much higher, which poses a challenge to scale and sustainability (Gao et al., 2024; Wang et al., 2025).

In recent years, research has made more emphasis on incorporating coarse aggregates in UHPC as a solution to enhance sustainability, dimensional stability, and the general material efficiency without necessarily losing the mechanical performance (Zhuang et al., 2024; Zhong et al., 2025). Research has found that the compressive strength of UHPC with rough aggregates, with the right particle packing density and a good mix of proportions, may reach compressive strengths in an even better manner than that of an aggregate-free UHPC. Moreover, coarse aggregates help to decrease cement usage, enhance dimensional stability, reduce autogenous shrinkage, reduce heat of reaction, decrease the risk of thermal cracking, and therefore are beneficial in large-scale structural applications (Matiushin et al., 2024)The addition of coarse aggregates can, however, result in the heterogeneity of the matrix and compromise the strength of the interfacial transition zone (ITZ), which can negatively impact tensile strength, fracture, and ultimately long-term longevity. Therefore, the trade-off between the mechanical performance and sustainability is one of the main issues in the design and optimisation of the UHPC systems (Marvila et al., 2021; Li et al., 2025).

Simultaneously, machine learning (ML) technology integration stands out as a groundbreaking technology in concrete technology. Predictive models such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), and ensemble learning algorithms can successfully forecast compressive strength, tensile behaviour, durability performance, and mix design optimization (Hussain and Soliman, 2025). These models enable multi-objective optimization by modelling nonlinear interactions of material constituents, such as aggregate content, Fibre dosage, and binder composition. Despite these advancements, recent applications of ML in UHPC are hampered by the unavailability of datasets, the absence of generalization, and weak integration with physics-based models and the lifecycle assessment frameworks (Mansour et al., 2023; Zandifaez et al., 2024).

Moreover, UHPC also demonstrates complicated behaviour during dynamic loading, like impact, blast, and seismic loads (Liu et al., 2024; Niu et al., 2025). Availability or unavailability of coarse aggregates has a tremendous effect on crack initial formation, crack propagation, and dissipation at the differing strain rates (Yu et al., 2021). Parallel to the fact that aggregate-free UHPC shows better ductility and crack behaviour, the inclusion of an aggregate in UHPC is advantageous due to the better load transmission and shrinkage resistance (Akeed et al., 2022; Matiushin et al., 2024). Nevertheless, there has been limited research on multiscale studies involving material behaviour, structural behaviour, and computational modelling.

Therefore, a systematic and integrated review is required to critically assess the function of coarse aggregates in UHPC and some of the newer machine learning methods of performance prediction and optimization. The current paper fulfils this requirement by systematically analysing the literature published between January 2000 and May 2026, covering mechanical performance, durability, sustainability, and computational advances. The goal is to have a clear state of trade-offs, research gaps, and future directions on how to come up with optimized, sustainable, and intelligent UHPC systems. Several review articles have examined various aspects of UHPC, such as constituent materials, mechanical performance, durability, sustainability, and structural applications (Du et al., 2021; Amran et al., 2022; Kravanja et al., 2024). However, most existing reviews focus on individual themes rather than providing an integrated assessment of UHPC. In particular, the combined influence of the incorporation of coarse aggregate, the behaviour of the interfacial transition zone (ITZ), sustainability, and machine learning-based optimization has not been comprehensively evaluated within a single review framework. Table 1 explains the comparison of UHPC with present and published review articles. Furthermore, conflicting conclusions concerning the effect of coarse aggregates on tensile performance, crack propagation, and long-term durability have yet to be critically reviewed. Therefore, the present review addresses these research gaps by systematically comparing UHPC with and without coarse aggregates, integrating recent advances in machine learning for performance prediction and mix design optimization, and proposing future research directions for the development of sustainable, intelligent, and cost-effective UHPC Systems. Figure 1 represents the general structure of the interconnection between the material composition, microstructure, performance, sustainability, and machine learning in UHPC.

TABLE 1. Comparison between UHPC review articles and the present review.

1.1 Research significance

This review advances the current understanding of Ultra-High-Performance Concrete (UHPC) by presenting an integrated evaluation of coarse aggregate incorporation from the perspective of mechanical performance, durability, sustainability, interfacial transition zone (ITZ) behaviour, and machine learning-based optimization. Unlike the previous review articles, which primarily focus on individual aspects of UHPC, the present study systematically compares the UHPCs with and without coarse aggregates to identify the associated performance trade-offs, unresolved scientific challenges, and emerging research opportunities. Furthermore, the recent advances in sustainable material design, advanced microstructural characterisation, and data-driven approaches for mix design optimization are critically synthesized. By identifying existing research gaps and proposing future research directions, this review provides a comprehensive framework for the development of sustainable, intelligent, and high-performance UHPC systems for research efforts and practical engineering applications.

2 Methodology

The paper is based on a systematic literature review in order to critically analyse the behaviour of Ultra-High-Performance Concrete (UHPC) with and without coarse aggregates, as well as the combination of machine learning methods for the design of materials and to predict the performance. To guarantee the reliability of the review process, its reproducibility, and thoroughness, a well-organized and open methodology was adopted.

2.1 Data sources and search strategy

The literature search was conducted using Scopus, Web of Science, Google Scholar, and SpringerLink to identify the peer-reviewed studies published between January 2000 and May 2026. The database searches were performed between August 2025 and May 2026. The publication period was selected to capture the development of UHPC research from its early development through the most recent literature available at the time of the search.

Relevant studies were found using a combination of keywords, such as Ultra-High-Performance Concrete (UHPC), coarse aggregate, mechanical properties, durability, sustainability, and machine learning in UHPC. To narrow-down on the search and provide complete coverage on the topics relevant to the search, these keywords have been combined with Boolean operators (AND, OR).

The literature search was conducted using a predefined search protocol. Combinations of keywords and Boolean operators were employed to ensure a comprehensive and systematic search of the available literature. Separate thematic search strings were developed to identify studies addressing coarse aggregate incorporation and mechanical performance, durability, sustainability, and machine learning applications in UHPC. Representative search strings included: (“Ultra-High-Performance Concrete” OR “UHPC”) AND (“coarse aggregate” OR “aggregate incorporation”) AND (“mechanical performance” OR strength OR flexural OR tensile) (“Ultra-High-Performance Concrete” OR “UHPC”) AND (durability OR chloride OR “freeze-thaw” OR permeability OR shrinkage) (“Ultra-High-Performance Concrete” OR “UHPC”) AND (sustainability OR “life cycle assessment” OR LCA OR carbon OR emissions); and (“Ultra-High-Performance Concrete” OR “UHPC”) AND (“machine learning” OR “artificial intelligence” OR ANN OR SVM OR SVR). The search syntax was slightly modified, where necessary, to comply with the indexing and search requirements of the individual database.

2.2 Selection and screening of the studies

The selection of the studies was done through a systematic filtering process in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines. Database searches resulted in an initial pool of approximately 150 publications. The studies were screened based on titles and abstracts to eliminate irrelevant records. After preliminary screening, approximately 120 records remained for further assessment. Following the screening, 20 records were excluded because they were outside the scope of the review, leaving 100 articles for full-text assessment. The full-text articles were evaluated against the predefined inclusion and exclusion criteria, considering their relevance to UHPC with and without coarse aggregates, mechanical and durability performance, sustainability, and ML applications. Out of 100 full-text articles assessed, 37 were excluded because they did not satisfy the eligibility criteria, lacked sufficient technical information, or were not peer-reviewed. Consequently, 63 studies were included in the final systematic review, as shown in Figure 2. Because of the heterogeneity in study designs, mixture compositions, testing conditions, and reported outcomes, a statistical meta-analysis was not performed; instead, the evidence was synthesized qualitatively and thematically.

The screening procedure was conducted sequentially through identification, duplicate removal, title and abstract screening, full-text eligibility assessment, and final inclusion; the selected studies were subsequently subjected to data extraction and thematic classification according to their research focus. This procedure was adopted to maintain consistency and transparency in study selection and to ensure that the final evidence base was directly aligned with the review’s objectives.

2.3 Inclusion and exclusion criteria

Predefined inclusion and exclusion criteria were used to screen the selected studies to ensure the consistency, reliability, and relevance of the selected studies. These criteria were meant to represent high-quality literature pertinent to Ultra-High-Performance Concrete (UHPC) with and without coarse aggregates, including studies addressing mechanical performance, durability, sustainability, microstructural behaviour, and ML applications. Studies that did not satisfy the predefined research scope or lacked sufficient technical information were excluded. Table 2 provides a detailed description of the inclusion and exclusion criteria.

TABLE 2. Provides a summary of the inclusion and exclusion criteria to select studies.

The inclusion and exclusion criteria were established prior to the literature search to ensure a systematic, transparent, and unbiased selection process. Each study was evaluated against these predefined criteria during the title, abstract, and full-text screening stages. Only studies that fulfilled all inclusion criteria and provided sufficient technical, experimental, or computational information relevant to the objectives of this review were retained for further analysis.

2.4 Data extraction and classification

The selected studies were analysed and classified according to the relevant performance indicators such as compressive strength, tensile strength, durability properties, sustainability, and the cost-effectiveness of each category (Ashteyat et al., 2025; Dong et al., 2025). Studies involving machine learning applications were categorized according to the model type, input parameters, and prediction objectives.

Additionally, the literature was systematically categorized into thematic groups, including.

• UHPC with and without coarse aggregates
• Interfacial Transition Zone (ITZ) Behaviour
• Durability and mechanical performance.
• Sustainability assessment
• Machine learning-based optimization

The classification served as an opportunity for comparison and critical evaluation of the existing research. Relevant information was systematically extracted from each of the selected studies, including publication details, constituent materials, coarse aggregate characteristics, mix design parameters, mechanical properties, durability performance, sustainability indicator, microstructure observation, machine learning methodologies, and important findings and limitations discussed by the authors. This consistent extraction process helped to maintain consistency in the data collected and also increased the meaningful comparison between the selected studies.

After data extraction, the selected studies were categorized into thematic categories based on their primary research objective and reported outcomes. The major thematic categorized into thematic categories included material composition and mix design, the influence of coarse aggregate on UHPC performance, interfacial transition zone (ITZ) behaviour, sustainability assessment of UHPC, machine learning application, and directions for future research. This thematic classification enabled a structured synthesis of the literature while facilitating the identification of research trends, gaps, and new opportunities in UHPC research.

2.5 Scope and limitations

Even though a thorough and methodological approach was utilized, there are some constraints related to the differences in experimental designs, discrepancies in reporting guidelines, and the unavailability of standard datasets, especially in experiments based on machine learning. Also, the variation in the material composition, curing, and testing carried out in various studies can also influence the direct relatability of the results.

2.6 Review framework

This review adopts an integrated approach combining thematic and comparative analyses to evaluate UHPC performance. The systematic methodology facilitates a systematic and transparent approach to the selection of the relevant literature, which is applied based on pre-defined criteria, whereas the thematic analysis allows grouping the studies on central research topics, such as UHPC in the presence and absence of coarse aggregates, interfacial transition zone behaviour, sustainability, and machine learning applications (Yoo and Banthia, 2019; Yang et al., 2022). Moreover, a comparative analysis including the trade-off between mechanical performance, durability, and cost-effectiveness, as well as environmental impact, is conducted and explained briefly in Supplementary Appendix A. This is an integrated framework giving a complete picture of UHPC behaviour and aids in determining research gaps and future directions (Du et al., 2021).

3 Thematic analysis of UHPC studies

This section presents a systematic analysis of the selected literature to identify research trends, materials developments, and the growing role of computational techniques in UHPC. The analysis includes publication trends, material composition developments, environmental considerations, and the increasing adoption of machine learning approaches.

3.1 Publication trends and research evolution

Figure 3 presents publication trends across different UHPC research themes. As the categories are not mutually exclusive, the publication counts do not correspond directly to the final studies included in the PRISMA-based review. The historical evolution of UHPC research reflects the changing priorities of the construction industry and advances in material science. During the early stage (2000–2010), research primarily focused on understanding the fundamental mechanical properties, durability, and microstructural characteristics of UHPC, as the material was still emerging and required extensive experimental validation for structural applications. Between 2010 and 2020, research shifted towards fibre reinforcement mechanisms, durability enhancement, and large-scale structural implementation, driven by increasing confidence in UHPC performance and growing interest in engineering, precast construction, and infrastructure applications. The 2026 values represent publications identified up to May 2026, corresponding to the end date of the literature search.

A marked increase in publications after 2020 is primarily attributed to the global emphasis on sustainable construction, carbon emission reduction, circular economy principles, and digital material design. Machine learning applications have increased in recent UHPC research, particularly for performance prediction, mix-design optimization, and material characterization. However, ML-based studies remain a relatively limited subset of the reviewed literature compared with conventional experimental investigations. Similarly, sustainability-oriented research expanded considerably due to increased utilization of supplementary cementitious materials (SCMs), reduced cement consumption, and the need to improve the environmental performance of UHPC without compromising its mechanical characteristics. These trends indicate that contemporary UHPC research is evolving from material characterization towards intelligent, sustainable, and performance-based material optimization.

3.2 Material composition and design trends

UHPC mixtures have been developed significantly in terms of mixture composition, in order to integrate performance with sustainability. Conventional UHPC mixtures are based on a large percentage of cement, silica fumes, fine particles, and steel fibres in order to attain high packing of the particles and high mechanical characteristics.

Recent research highlights partial cement replacement by using supplementary cementitious materials (SCMs), which are fly ash, ground granulated blast furnace slag (GGBS), and metakaolin (Amran et al., 2022; Gong et al., 2022; Bahmani et al., 2024). These materials aid in low carbon emissions, although they are strong and durable. They have also introduced nano-materials like nano-silica, which have promoted nano-material density of microstructure and interfacial bond (Yoo et al., 2024).

Additional developments that have been made are the use of hybrid Fibre systems to enhance crack resistance and energy absorption ability, especially when subjected to dynamic loading conditions (Zhuang et al., 2022).

3.3 Coarse aggregates effects on UHPC performance

The usage of coarse aggregates in UHPC has become a significant research focus area on enhancing the mechanical stability, stability performance, as well as sustainability in the long run. The performance and durability properties of UHPC in environmental performance and durability largely depend on whether there are coarse aggregates or not (Ma, 2004; Feng et al., 2022).

It has been established that UHPC that includes coarse aggregates has better dimensional stability, decreased shrinkage, and increased resistance to environmental forces like freeze-thaw exposure (Kevern et al., 2010; Lu et al., 2021). Additionally, proper gradation of aggregates as well as maximized particle packing can help in similar compressive strength performance as conventional UHPC mixtures. Nonetheless, coarse aggregates can raise the heterogeneity in the matrices, especially in the interfacial transition zone (ITZ), which can negatively impact the tensile strength and crack propagation behaviour (Schmidt et al., 2004; Li et al., 2021a).

Therefore, recent experiments have been aimed at maximizing aggregate gradation, size distribution and surface attributes in a bid to enhance ITZ performance without compromising mechanical performance, sustainability as well as cost-effectiveness (Li et al., 2020; Lv et al., 2020; Meng et al., 2022).

3.4 Trends in UHPC research of machine learning

ML has emerged as an increasingly investigated approach in UHPC research, particularly for performance prediction, mix-design optimization, and material characterization. The reviewed literature indicates growing interest in data-driven approaches, although their application remains limited by the availability and heterogeneity of experimental datasets, differences in validation strategies, and limited model interpretability. This growing interest is primarily driven by the need to reduce experimental costs and time while improving prediction accuracy with optimised complex relationships among constituent materials and performance parameters.

There are several popular models based on Artificial Neural Networks (ANN), Support Vector Machines (SVM), and hybrid optimization algorithms (Le Nguyen et al., 2024; He et al., 2025). These models can highly predict compressive strength, tensile behaviour, durability performance, and environmental impact.

Machine learning can be used to explore the small and big relationships between mix design variables in an efficient way, such as the content of aggregates, Fibre dosage, and the composition of binders (Li et al., 2018; Li et al., 2021c). Nonetheless, such limitations as the lack of datasets, non-standardization, and generalization are especially vital issues.

3.5 Key observations from literature analysis

As a result of the literature analysis, the following key observations can be made:

• UHPC Research has transformed from strength-oriented investigations towards sustainability-driven and data-driven.
• The use of coarse aggregate has great benefits in terms of cost and environmental savings.
• Aggregate-free UHPC is characterized by better tensile strength and crack resistance because of the enhanced matrix homogeneity.
• Improved performance and durability are achieved with the help of SCMs and nanomaterials in material design.
• Machine learning has become a popular methodology in predictive modelling and optimization; however, there is a need to standardise and validate further.

The observations were supported by the comparative evidence synthesized from the reviewed studies. Across the literature, aggregate-free UHPC is generally associated with improved tensile and crack-resistance characteristics, which are attributed to greater matrix homogeneity and fiber dispersion, whereas the incorporation of coarse aggregates is consistently associated with reduced cement demand, lower autogenous shrinkage, and improved dimensional stability. However, the reported mechanical response is not uniform and depends on aggregate size, gradation, replacement level, curing conditions, fiber dosage, and ITZ characteristics. For ML applications, the reviewed studies predominantly report the use of ANN, SVR, random forest, decision-tree, and hybrid models, with R ², RMSE, MAE, and accuracy commonly used for model evaluation. Table 3 summarizes these model classes, their reported evaluation metrics, advantages, and limitations. Thus, the evidence supports the identified trends while also demonstrating that direct numerical comparison remains constrained by differences in experimental databases, mixture compositions, testing conditions, and model-validation procedures.

TABLE 3. Machine learning models used in UHPC studies.

4 UHPC thematic and comparative analysis

4.1 UHPC without coarse aggregates

UHPC without coarse aggregates represents the conventional formulation approach, which is mainly aimed at reaching a highly homogeneous and dense cementitious matrix. The lack of coarse aggregates removes the weak interfacial zones and facilitates an even distribution of stresses that drastically improves the tensile strength, ductile characteristics, and crack resistance.

Several studies have indicated that aggregate-free UHPC has better mechanical performance, especially on tensile and flexural loading. This leads to improved dispersion of steel fibres within the matrix, which enhances the ability of steel Fibres to bridge and retards the formation and spreading of the cracks (Hou et al., 2018; Cao et al., 2019). The fine structure with the most efficient packing of the particles and the use of fine materials like silica fume are additional factors that result in a lower porosity and higher durability.

Besides mechanical benefits, aggregate-free UHPC has good resistance to adverse environmental factors such as chloride penetration, freezing and thawing, and chemical intrusion. The small matrix design reduces permeability and maximizes long-life stability, which is appropriate for use in critical infrastructure applications that are vulnerable to harsh conditions.

Nevertheless, the omission of coarse aggregates not only makes the cement content large but also makes the material quite costly and harmful to the environment (Zhang et al., 2012; Ren et al., 2018). The high content of binder predisposes to high carbon emissions and high energy use, which casts uncertainty on the sustainability of conventional UHPC formulations. Moreover, the high density of the matrix provides a contribution to the higher degree of autogenous shrinkage, and in this case, the cracking at a young age can occur unless managed carefully.

Scalability is another weakness of aggregate-free UHPC for large structural applications. The low popularity of cement in the mass-building sector is due to its high price combined with the expensive fine materials, notwithstanding the benefits of its performance features.

All in all, UHPC in the absence of coarse aggregates has demonstrated a high tensile strength, durability, and crack-resistance, but its implementation is limited by economic and environmental factors. These disadvantages have led to more recent studies using alternative formulation possibilities, such as adding coarse aggregates to allow a more balanced performance profile.

4.2 UHPC using coarse aggregates

The use of coarse aggregates in Ultra-High-Performance Concrete (UHPC) has become a variant formulation to nearly achieve sustainability, cost-effectiveness, and workability. The use of coarse aggregates, as opposed to the standard UHPC, makes the cementitious material volume smaller, therefore making the production less costly and more environmentally sustainable (Zhong et al., 2025).

The studies have proved that the compressive strength of UHPC using coarse aggregates can be similar to that of the traditional UHPC made without any aggregates when the mix is designed adequately, and the optimum packing of the particles in it is done (Zhuang et al., 2024). Coarse aggregate presence also helps in better dimensional stability and lessens autogenous shrinkage, useful in the application of massive constructions. Moreover, the decreasing cement content causes a decline in heat of hydration and reduces the risk of thermal cracking.

Sustainability-wise, the use of coarse aggregates will produce a much better carbon emission reduction relative to the production of cement. The use of recycled and locally available materials further enhances the environmental sustainability of UHPC (Butler et al., 2013; Al Shouny et al., 2023; Davolio et al., 2025).

Nevertheless, the heterogeneity in the cementitious whole of coarse aggregates is introduced, especially in the interfacial transition zone (ITZ) between the surface of coarse aggregates and the enveloping paste (Li et al., 2025). Weaker bonding and increased porosity in this region than in the bulk matrix are usually observed and can negatively impact tensile strength, fracture behaviour, and crack propagation resistance.

Research has shown that coarse aggregates can lead to discontinuity in the distribution of the fibres, minimizing the effects of the fibre bridging mechanisms. Consequently, UHPC containing coarse aggregates will tend to have lower tensile strength as well as worse crack control than the UHPC without aggregates. Limitations are especially important in the case of flexural and dynamic loading (Li et al., 2021b).

In response to these challenges, recent studies have worked on how to optimize the size of aggregates, gradation, and surface to increase the performance of ITZ. Surface treatment of aggregates, addition of nanomaterials, and more efficient packing techniques of the aggregates have been suggested to increase the bond between the aggregates and the cement mass. Table 4 concludes the comparative performance of UHPC with and without coarse aggregate.

TABLE 4. Comparative performance of UHPC with and without coarse aggregate.

In conclusion, UHPC with coarse aggregates is a balanced solution as it enhances cost, feature, and sustainability, while maintaining acceptable mechanical performance. Nevertheless, it is necessary to carefully optimize the mix design and microstructural properties to alleviate the negative impact of ITZ behaviour and tensile behaviour (Cheng Jun, 2017).

The comparative analysis reveals that, due to its homogeneous matrix and improved dispersion of fibres, aggregate-free UHPC generally exhibits superior tensile strength, ductility, and crack resistance compared to UHPC incorporating coarse aggregates. By contrast, UHPC mixes with coarse aggregates offer several benefits, including lower cement consumption, reduced autogenous shrinkage, improved dimensional stability, enhanced economic feasibility, and reduced environmental impacts. Experimental studies have also demonstrated that properly designed coarse-aggregate UHPC can achieve very high compressive strength; for example (Lv et al., 2020),reported compressive strength exceeding 200 MPa for coarse-aggregate UHPC. However, conflicting findings regarding compressive strength and long-term durability have been reported in the literature. These differences are primarily attributed to variations in aggregate mineralogy, particle size distribution, aggregate replacement level, fibre dosage, curing conditions, and interfacial transition zone (ITZ) characteristics. Therefore, the successful incorporation of coarse aggregates depends not only on adding coarse aggregates to the mix but also on optimized mix design.

4.3 UHPC interfacial transition zone (ITZ) behaviour

The interfacial transition zone (ITZ) is vital in the control of the mechanical performance and durability of the Ultra-High-Performance Concrete (UHPC), especially in mixes involving coarse aggregates. ITZ is a crucial microstructural region surrounding aggregate particles, often characterized by higher porosity and weaker bonding compared to the bulk matrix (Li et al., 2025).

In traditional UHPC, where there are no coarse aggregates, large aggregates cause an extremely fine and dense matrix of minimum interfacial discontinuities (Matiushin et al., 2024). This homogeneity promotes stress distribution and leads to high tensile strength, ductility, and resistance to cracks. The homogeneous distribution of small particles and Fibres also enhances the ability of the microstructure of the composite.

Compared to this, the introduction of coarse aggregates brings about discrete ITZ that serve as possible weak areas within the material. When aggregates are stiffer than their surrounding matrix, stress becomes concentrated in the form of aggregates under load, thus favouring the initiation of cracks at the ITZ. Moreover, the porosity of this area is generally higher and has a lower packing density than the cement paste, which can weaken the connections between the surfaces of aggregates and cement.

When the tensile and flexural loads are present, the effect of the ITZ is especially important; in that case, crack propagation depends on local microstructural behaviour. Cracks tend to initiate and extend at the ITZ when using off-the-shelf aggregates in UHPC, decreasing the total crack resistance. This is enhanced more when the distribution of the Fibre is not evenly distributed, or its structure is disturbed by the inclusion of aggregates.

To alleviate the negative impact of ITZ, various measures have been recommended in recent studies. These involve optimization of aggregate gradation to enhance packing density, surface-treated aggregate can be used to enhance bonding, and the use of nano-materials like nano-silica can be used to tune the microstructure. A better particle packing method can lower the porosity in the ITZ as well, resulting in an increase in mechanical performance.

Moreover, Fibres and ITZ interaction are important factors in bridging cracks and dissipating energy. Fibres can also bridge cracks across the ITZ, compensate local weaknesses, and enhance ductility in highly optimized systems. But this must be closely regulated in the orientation and distribution of fibres in the matrix.

Recent experimental studies have provided deeper insights into the evolution of ITZ through advanced microstructural characterization techniques. The use of coarse aggregates could result in an increase in the local porosity and microcracks in the ITZ; optimized coarse aggregate packing within the cement paste may increase the density of the ITZ and enhance the bond between aggregates and cement paste, and the addition of nano-silica or SCMs could lead to a denser microstructure and better bond between aggregates and cement paste. X-ray Computed Tomography (X-ray CT) has been used to visualize the three-dimensional distribution of aggregates, fibres, and internal crack networks, demonstrating that fibre orientation and aggregate morphology play a major role in determining crack paths and initiation points. In addition, nanoindentation tests revealed significant differences in local stiffness and hardness within the ITZ, indicating that the quality of the ITZ strongly affects the mechanical performance of UHPC. These experimental observations demonstrate that optimizing the ITZ through improved particle packing, aggregate surface modification, and controlled fibre distribution is essential for enhancing the mechanical performance, durability, and long-term structural reliability of coarse aggregate UHPC.

In general, the ITZ can be regarded as one of the determining factors of the performance of coarse-grained UHPC systems. Although it presents certain issues regarding strength and durability, the right mix design and optimization of the microstructure can boost the properties of ITZ to a considerable degree and allow to create more efficient and environmentally friendly UHPC compositions.

4.4 Sustainability assessment of UHPC

Ultra-High-Performance Concrete (UHPC), due to its own high cement content and consequent environmental impact, has recently made the sustainability of UHPC a significant point of research (Gao et al., 2024; Farahzadi et al., 2025). The traditional UHPC mixtures correspond to a strong dependence on cement and fines to obtain a high level of compacting the particles together and the enhanced mechanical properties, leading to enhanced carbon emissions and energy use. Consequently, optimization of the sustainability of UHPC, but without a reduction in performance, has become an essential research goal (Abbas et al., 2016). Figure 4 presents the environmental exposure distribution of UHPC studies.

Part of the cement replacement with alternative cementitious materials (SCM) includes fly ash, ground granulated blast furnace slag (GGBS), and metakaolin as some of the main vectors that can aid in achieving greater sustainability. These materials can reduce the carbon footprint while maintaining, and in some cases improving, long-term strength and durability through pozzolanic reactions (Sameer et al., 2019; Kravanja et al., 2024). The reviewed durability studies further indicate that UHPC performance under environmental exposure is strongly dependent on mixture composition and curing conditions, while long-term field validation remains limited.

The use of coarse aggregates is another viable approach towards enhancing sustainability. The substitution of a part of the cementitious matrix by aggregates will decrease the total cement demand, allowing to reduce carbon emissions and the cost of the material (Lande and Terje Thorstensen, 2023). Research has shown that coarse aggregate UHPC has better environmental performance than traditional aggregate-free UHPC in general, with optimized mix designs.

Moreover, recycled aggregates and the use of local materials have been considered in order to improve the environmental impact of UHPC. These solutions decrease the number of emissions connected with transportation and advance the ideas of the circular economy in the construction industry. Nevertheless, recycled materials must be carefully used, and the quality of materials, as well as their mechanical behaviours and other terms contributing to longevity, have to be taken into consideration.

Regardless of these developments, there are still obstacles in creating a balance between sustainability and performance. The cement content can be reduced to have an influence on early-age strength and durability, whereas the coarse aggregates can bring about the heterogeneity of the microstructure. A multi-objective optimization strategy is thus needed to optimize strength, durability, costs, and environmental impact.

The value of life cycle assessment (LCA) in assessing the sustainability of UHPC has also been highlighted in recent studies. LCA offers a broad perspective of the environmental impact of the various phases, such as material fabrication, transportation, and construction, as well as the service life. Yet, the use of LCA in the UHPC study is scarcely proposed and has yet to be developed (Farahzadi et al., 2024).

All in all, to enhance the sustainability of UHPC, a set of issues, such as optimization of the materials, effective use of resources, and sophisticated analysis, will be required. The coarse aggregates and SCMs are integrated with lifecycle and evaluation methods, which is a promising direction for establishing environmentally sustainable UHPC systems.

While several studies have demonstrated the environmental benefits of using coarse aggregates and supplementary cementitious materials (SCMs) in UHPC production, significant inconsistencies exist in the sustainability assessment methodology used. The reviewed literature indicates that integrating machine learning (ML) with life-cycle assessment (LCA) can support strength prediction and the evaluation of environmental impacts during multi-objective mix optimization. Most life cycle assessment (LCA) studies have a functional unit of 1 m³ of UHPC; varying life-cycle system boundaries were adopted between cradle-to-gate and cradle-to-grave, making direct comparison of environmental impacts challenging. Cement production remains the largest contributor to the global warming potential (GWP) of UHPC because of its high clinker content, while steel fibres also contribute significantly owing to their energy-intensive manufacturing process. The use of supplementary cementitious materials (SCMs) like silica fume, fly ash (FA), metakaolin, and ground granulated blast furnace slag (GGBS) as partial replacement materials for cement is consistently reported to be contributing to a reduction of carbon emissions without significantly degraded mechanical properties. In addition, variables such as transportation distance, material sourcing, and manufacturing processes also influence the overall environmental footprint of UHPC but are often excluded from existing sustainability assessments. Future sustainability evaluations should thus be based on harmonized functional units, a standardized system boundary, and holistic life-cycle assessment (LCA) strategies that include environmental, economic, and mechanical performance indicators to allow for meaningful comparisons between different UHPC systems. The reviewed studies indicate that AI-guided optimization has the potential to simultaneously improve mechanical performance, reduce environmental impacts, and control material costs, demonstrating the value of multi-objective optimization for sustainable UHPC mix design.

4.5 UHPC in machine learning applications

The growing complexity of Ultra-High-Performance Concrete (UHPC) mix design, with many parameters that interact with each other, including binder composition, aggregate content, Fibre dosage, and curing conditions, has resulted in using machine learning (ML) techniques as effective predictive and optimization tools. Conventional experimental methods can be time-consuming and resource-intensive in comparison with ML models, which can be used to generate quick and efficient material performance forecasts (Su et al., 2025; Yuan et al., 2025).

Some of the machine learning algorithms that have been studied in UHPC research, among others, are Artificial Neural Networks (ANN), Support Vector Machines (SVM), Support Vector Regression (SVR), decision trees, and ensemble learning techniques (Yuan et al., 2023; Hussain and Soliman, 2025). These models can be used to model nonlinear interactions between input variables and output responses, such as compressive strength, tensile strength, durability indices, and workability (Yu et al., 2021; Taffese et al., 2024). Table 3 provides an overview of machine learning models and applications in UHPC. Figure 5 illustrates the machine learning framework used for prediction and optimization of UHPC properties.

The performance of machine learning models varies considerably depending on the dataset size, feature selection, and validation strategy. Most studies evaluated model performance using statistical indicators such as the coefficient of determination (R ²), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE). ANN and ensemble learning methods generally demonstrated superior prediction accuracy due to their capability of handling the nonlinear phenomena among materials of UHPC. However, when training sets were relatively small, Support Vector Regression (SVR) frequently showed better generalization performance. Nevertheless, direct comparison among published models remains challenging because of variations in experimental databases, input variables, data preprocessing techniques, and validation strategies adopted by different researchers.

Among these methods, the ANN-based models have become one of the most popular since they can cope with extensive datasets and their capability to model non-linear interactions. On the same note, SVR models have been found to have high predictive capability, especially when it comes to estimating the strength and performance of durability (Hussain and Soliman, 2025). Multi-objective optimization of UHPC mix design has also caught attention on hybrid models that incorporate optimization algorithms with machine learning methods, where strength, cost, and environmental impact are among the factors that should be considered.

Coarse aggregates and additional cementitious materials and their effect on the performance of UHPC have also been analysed using machine learning. ML models do this by efficiently exploring the most optimal combinations of materials by using various input parameters, and do not require a lengthy process of trial-and-error, as they do through experimental testing.

Despite the considerable predictive potential of machine learning, several challenges continue to limit its practical application in UHPC research. One such limitation is the lack of heterogeneity in experimental databases, and the difficulty in generalizing models trained with one database to another is mainly due to the use of different constituent materials, curing conditions, testing procedures, and specimen dimensions. In addition, prediction performance depends on input features, and small data sets can easily lead to overfitting. Most of the existing studies also use black-box models, which are hard to explain since they do not provide interpretability for the model predictions. Consequently, explainable artificial intelligence (XAI), uncertainty quantification techniques, and standardized benchmark datasets have emerged as important research directions for improving the reliability, transparency, and practical applicability of machine learning models in UHPC.

The reviewed studies indicate that ML models require sufficiently large and continuously updated datasets, while careful data preprocessing is essential for reducing outliers and improving prediction accuracy. The automated database can enhance future UHPC prediction system. To address these issues, recent studies have focused on integrating machine learning with physics-based modelling, digital twin technologies, and multi-objective optimization frameworks. The methods are meant to enhance the accuracy of predictions and physical relevance. Furthermore, future machine learning models should incorporate life-cycle assessment indicators, carbon emissions, economic factors, and uncertainty analysis to support the development of sustainable and intelligent UHPC mix design. Recent studies have shown that integrating machine learning (ML) with life-cycle assessment (LCA) enables the simultaneous optimization of mechanical performance, environmental sustainability, and economic efficiency in UHPC mix design.

On the whole, machine learning provides a good avenue to develop UHPC design through efficient prediction, optimization, and decision-making. Nevertheless, more studies are needed to enhance the quality of data, standardization, and integration of the models to maximize its capabilities in real-life settings. A comparative summary of the UHPC systems with and without coarse aggregates is given in Table 5.

TABLE 5. Comparative summary of UHPC systems.

5 Discussion

The reviewed literature reveals a fundamental trade-off between mechanical performance and sustainability in UHPC with and without coarse aggregates. Traditional UHPC, constituted of no coarse aggregates, has better tensile strength, ductility, and crack resistance because its homogeneous matrix structure is highly effective, as it contains well-dispersed Fibres. The lack of coarse aggregates removes the weak interfacial transition zones (ITZ), whose distribution of stress and fracture resistance is enhanced (Ma, 2004; Lv et al., 2020).

Conversely, UHPC with coarse aggregates has a great benefit in terms of cost-effectiveness and environmental sustainability is shown in Figure 6. The decrease in cement content results in the emission of less carbon and a more economical, sustainable usage, as it is more efficient in massive structures. Coarse aggregates reduce autogenous shrinkage as dimensional stability. These advantages are, however, at the cost of decreased tensile behaviour and enhanced heterogeneity of the matrix. These findings highlight the necessity of achieving a balanced UHPC design that integrates mechanical performance, sustainability, and cost efficiency.

The differences observed between the systems of UHPC with and without coarse aggregate are mainly due to microstructural characteristics and interfacial interactions. The greater homogeneity of the matrix and better Fibre dispersion of aggregate-free UHPC will result in efficient stress transfer and crack-bridging mechanisms. Coarse aggregates presence, on the other hand, creates interfaces of transition zones (ITZs) which can be a source of weak zones when the material is subjected to tensile and flexural stresses. But these impacts can be minimized by optimizing aggregate size distribution, improving particle packing density, surface-modifying techniques, and adding nanomaterials. In this sense, further studies on the design of the UHPC should focus on achieving a homogenized microstructure to achieve better sustainability goals as well as overall performance.

The combined effect of the particle packing density, the characteristics of aggregates, the quality of the interfacial transition zone (ITZ), the distribution of fibres, and the composition of binders explains the performance of UHPC, which was indicated by the collective evidence reviewed in this study. The conflicting findings reported in the literature on the effect of coarse aggregate are primarily attributed to variations in aggregate mineralogy, particle size distribution, degree of aggregate replacement, curing regime, and experimental methodologies. These observations suggest that the successful incorporation of coarse aggregates depends on optimizing the overall mix design instead of treating coarse aggregates as an independent variable.

Moreover, the integration of advanced microstructural characterization and machine learning-based optimization offers a promising pathway for the development of next-generation UHPC systems. Experimental investigation and data-driven prediction models help to optimize mechanical performance, durability, sustainability, and economic viability simultaneously, which allows the design of intelligent, environmentally friendly UHPCs for future structural applications. Taken together, the reviewed evidence indicates that UHPC mix design should not be optimized solely for maximum mechanical strength. A balanced design approach considering mechanical performance, durability, dimensional stability, environmental impact, cost, and material availability is more appropriate. In this context, ML provides a promising tool for identifying nonlinear relationships among mixture parameters and performance indicators, although its practical reliability remains dependent on dataset quality, model interpretability, and validation against experimental evidence.

6 Limitations

Although the methodology of this study was extensive and comprehensive, a number of limitations must be noted. This study is based on published literature, which may involve variations in experimental methods, material compositions, and testing conditions. Such discrepancies may have an impact on the direct comparability of results among studies.

Second, standardized datasets in UHPC studies, especially machine learning cases, are unavailable, which hinders the extrapolation of results. These variations may introduce uncertainties in data interpretation and comparison across studies. Third, short-term mechanical performance is studied the most, whereas long-term durability, long-term lifecycle performance, and behaviour in complex environmental conditions are relatively little explored. Also, there is scant literature on the large-scale strategy of implementing UHPC using coarse aggregates.

Lastly, the combination of machine learning, physical and mechanistic models is still immature, which limits the predictive model reliability and interpretability.

7 Research challenges and future directions

Future research needs are related to overcoming current limitations and the development of optimized UHPC systems. Although there is much progress in the technology of UHPCs, there are many research problems that are yet to be addressed to accommodate its quick application. These involve the optimization of incorporating the coarse aggregate without sacrificing the tensile performance, the lack of standardized durability assessment procedures, the limited validation of sustainable UHPC mixtures at the field scale, and the lack of a comprehensive database of different mixtures that can be used as inputs for machine learning applications. In addition, the inclusion of sustainability indicators, lifecycle assessment metrics, and explainable artificial intelligence techniques in UHPCs design frameworks is still in the early stages. Standardization, comprehensiveness of given datasets, and incorporation of physics-based machine learning methods can improve the reliability of machine learning models as well as their generalization properties and, at the same time, increase their physical interpretability. A summary of the major research challenges, knowledge gaps, and future research directions derived from the literature reviewed is presented in Table 6.

TABLE 6. Research gaps and future directions in UHPC.

Major research gaps obtained from the literature studied are presented in Table 6. While significant progress has been achieved in understanding the mechanical, durability, and sustainability properties of UHPC, a few issues remain unsolved, such as optimization of coarse aggregate, long duration of durability, and utilization of explainable machine learning techniques. It is important to fill such gaps to promote practical implementation and sustainable development of next-generation UHPC systems. Additional studies are needed to maximize aggregate gradation and surface properties to enhance interfacial transition zone (ITZ) functions in UHPC that utilises coarse aggregates. Hybrid Fibre systems and techniques of nano-modification should also be considered to develop crack resistance and durability.

Moreover, lifecycle assessment (LCA) is advisable to be used in the future to assess the environmental impact of UHPC throughout its service life. The recycled materials and alternative binders should also be explored more to enhance sustainability.

Last but not least, to develop next-generation UHPC systems that can be applicable to the field practice, the multi-scale and multi-objective optimization methods must be reflected on mechanical performance, durability, cost, and environmental impact.

Though several capabilities of UHPC, such as performance, durability, sustainability, and machine learning applications, are well understood, a few challenges remain unsolved. Some of those constraints include optimization of future incorporation of coarse aggregate, establishment of standardized durability assessment protocols, incorporation of sustainability metrics in the mix design, and the limited use of explainable AI techniques. A coordinated research approach is needed to overcome these challenges, encompassing material innovation, durability enhancement, sustainability assessment, and digital engineering technologies. To this end, a roadmap for future research is forecasted as shown in Figure 7, towards developing the next-generation sustainable UHPC systems that will be intelligent.

8 Conclusion

This paper presents a systematic research that provides a comprehensive review of Ultra-High-Performance Concrete (UHPC) when compared to its application without coarse aggregates in terms of mechanical performance, durability, sustainability, and the use of machine learning. The results show that aggregate-free UHPC has better tensile strength, ductility, and crack resistance, which can be attributed to the dense and homogeneous structure of the material. Conversely, the use of coarse aggregate in UHPC reduces the need for cement, making it more cost-efficient and environmentally friendly.

Microstructural properties, specifically the behaviour of the interfacial transition zone (ITZ) and Fibre–matrix interaction, have been identified to predominantly dictate the performance differences between these systems. Although joint inclusion of coarse aggregates brings about issues of tensile characteristics and crack propagation, these tendencies can be reduced by optimum mix design and microstructural improvement measures.

Machine learning methods have shown a unique potential in predicting UHPC properties and optimizing the mix design, but their usage is now limited due to their sensitivity to data and the absence of correlation to physical models.

In general, this paper demonstrates the need to find a middle ground embracing both the mechanical performance, sustainability, and cost. Overall, this study provides a foundation for the development of next-generation UHPC systems that achieve an optimal balance between mechanical performance, sustainability, and intelligent design.

This review bridges existing gaps by offering an integrated scientific framework, incorporating coarse aggregate, ITZ behaviour, sustainability assessment, and machine learning-based optimization in a single systematic review. Findings distinguished established knowledge, including the benefits of optimized coarse aggregate incorporation in enhancing sustainability and dimensional stability, from the remaining uncertainties related to long-term durability, a standardized performance evaluation, and the generalizability of machine learning models. In practical terms, the findings from this review offer insights for researchers and engineers engaged in UHPC mixture design to ensure optimal performance, durability, sustainability, and cost-effectiveness. Future research should prioritize the development of standardized experimental databases, explainable artificial intelligence frameworks, physics-informed machine learning models, and comprehensive life-cycle assessment methodologies to accelerate the implementation of intelligent and sustainable UHPC systems in engineering practice.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbuil.2026.1903555/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Roberto Alonso González-Lezcano, CEU San Pablo University, Spain

Reviewed by: Atefeh Soleymani, University of Birjand, Iran

Sivaprakasam Palani, Addis Ababa Science and Technology University, Ethiopia