Artificial intelligence techniques in advanced concrete technology: A comprehensive survey on 10 years research trend

Advanced concrete technology is the science of efficient, cost‐effective, and safe design in civil engineering projects. Engineers and concrete designers are generally faced with the slightest change in the conditions or objectives of the project, which makes it challenging to choose the optimal design among several ones. Besides, the experimental examination of all of them requires time and high costs. Hence, an efficient approach is to utilize artificial intelligence (AI) techniques to predict and optimize real‐world problems in concrete technology. Despite the large body of publications in this field, there are few comprehensive surveys that conduct scientometric analysis. This paper provides a state‐of‐the‐art review that lists, summarizes, and categorizes the most widely used machine learning methods, meta‐heuristic algorithms, and hybrid approaches to concrete issues. To this end, 457 publications are considered during the recent decade with a scientometric approach to highlight the annual trend/active journals/top researchers/co‐occurrence of key title words/countries' participation/research hotspots. In addition, AI techniques are classified into distinct clusters using VOSviewer clustering visualization to identify the application scope and their relationship through the link strength. The findings can be a beacon to help researchers utilize AI techniques in future research on advanced concrete technology.


Different sciences
Ability of a machine imitate intelligent human behavior Application of AI that allows a system to automatically learn and improve from experience Application of ML that uses complex algorithms and deep neural networks to train a model Learning computers to process data in a way that is inspired by the human brain F I G U R E 2 Schematic relationship and scope between artificial intelligence (AI) and its subsets.  25 1965 Handling the concept of partial truth Group method of data handling (GMDH) Ivakhnenko 26 1968 Inductive approach Support vector machine (SVM) Boser, et al. 27 1992 Optimal margin classifiers Genetic programming (GP) Koza 28 1992 Natural selection process based on tree based function Extreme gradient boosting (XGB) Chen 39 2021 Aquila's behaviors in nature during the process of catching the prey metaheuristic algorithms for each technique. The inspired concept behind most of the techniques belongs to the phenomenological and behavioral from the nature and creatures. 22 Based on this, the techniques are further classified in accordance with imitating the evolutionary process in nature (evolutionary-based) and animal behavior in a group (swarm-based). 23

RESEARCH METHODOLOGY
First and foremost, accurate identification of published articles in the field under review is essential to achieve a comprehensive review. To this end, publications were collected using databases. Due to the difficulty of searching for any pertinent publication, it is necessary to delimit the research boundary. These boundaries are based on the title word, purposes, methodology, and significant contributions to select the publication within the research scope. At a glance, details of the methodology for the current survey are provided in Figure 4.

Bibliometric analysis
In this step, the literature search in the database was retrieved using keywords related to the research area. According to the purpose of this survey, the selected keywords were: "Cementitious materials", "Concrete", "Cement mortar", "Cement paste", and so forth to cover the field of advanced concrete technology and "Artificial intelligence or AI

Scientometric analysis
In this step, bibliometric data from published articles are utilized to map the network and evolution of related topics in accordance with large-scale scientific data sets. VOSviewer software (version 1.6.18) is used in this review. It utilizes the visualization of similarities (VOS) technique. It provides a facility of visualization and mapping of a knowledge field with the aim of analyzing its intellectual landscape. 40,41 This feature leads us to use scientometric analysis in the present survey. It enables young research scholars to gain a global perspective of research trend of using AI techniques in advanced concrete technology. In this regard, the following analyses were utilized to reveal research trend: publication year, participation of journals and countries, co-authorship analysis, title words clustering, and research hotspots.

RESULTS AND DISCUSSION
After the procedures and screenings mentioned in the previous section, a total of 457 articles were carefully selected to carry out a comprehensive survey in the current study area. Full details about the list of acronyms and extracted features in both the fields of AI and advanced concrete technology for each selected paper are listed in Tables A1 and B1, respectively. In the following subsections, seven aspects of the analysis are depicted: (5.1) annual publications' trend; (5.2) most contributing research sources; (5.3) trend of countries' participation; (5.4) co-authorship analysis; (5.5) clustering AI techniques by title words; (5.6) research hotspots; and (5.7) hybrid AI models.

Annual publications' trend
First of all, it is interesting to review the history of articles published in the field of using AI techniques to predict the characteristics in advanced concrete technology during the last decade. Figure 5 shows distribution of articles published on using AI techniques in the form of three categories: single model, hybrid model, and optimizing algorithm from Apr 2013 to Mar 2023. The figure reveals a fluctuating trend until 2018, which did not exceed 25 articles per year. In the following, a significant growth in the number of articles has been observed in the last 3 years (2020-2022). During the years 2020 and 2021, this upward trend reached respectively 59 and 98 articles published per year. However, the highest number of publications was in 2022 with 113 publications, more than twice the number in 2019. With the continuation of this trend, the number of articles is expected to increase significantly at the end of 2023. Taking a close look at the type of articles published each year, it can be seen that not only single-mode models but also hybrid models have attracted more attention in recent years. So that the number of articles published with hybrid models in 2022 and 2021 is more than the total of this type of articles in the previous 8 years ( [2013][2014][2015][2016][2017][2018][2019][2020]. This trend indicates that researchers are more willing to utilize AI techniques in advanced concrete technology as an alternative to expensive and time-consuming laboratory methods; therefore, more research is expected in the future. After that, it seems necessary to specify the extent of use of each technique. For this purpose, Figure 6 shows the variety and distribution of different methods used in advanced concrete technology. As can be seen, among the nine techniques that have attracted the most attention among researchers, ANN with 370 articles, followed by SVM, and GA with 69, and 45 articles published, respectively were the top three techniques in this scope with the highest amount of participation. In addition, the annual publications' trend of each technique is shown separately. By examining them, it can be seen that methods such as RF, XGB, and GWO have attracted the attention of researchers in the last 2 years. The findings of this section help researchers to take a closer look at which AI techniques were most or least used by previous researchers.

Most contributing research sources
The publisher information and participation percentage for leading journals that published more than 10 articles in the field of using AI techniques in advanced concrete technology are listed in Table 2. Among these journals, the "Constr  Optimizing algorithm  1  2  1  2  1  3  8  5  1  Hybrid model  3  1  3  3  4  3  7  1 0  2 3  2 1  6  Single model  15  11  12  21  15  21  32  41  70  91  20 No. of published articles   For a comprehensive review of methods, any number of methods used in an article is included in this figure.
Build Mater." has the most number of articles published equal to 99 with the participation percentage of 21.66, which indicates a significant difference from other journals. In the following, two other top journals are "Struct. Concr." and "Eng. Struct." with 26 and 24 articles, respectively. These three journals account for about a third of the articles in this field. For a closer look, Figure 7 illustrates the annual distribution of the number of articles published for each journal. It is clearly observed that the journal of "Constr Build Mater." was a leading journal in recent years. Besides, three journals of "Struct. Concr.", "Struct.", and "Case Stud. Constr" have contributed significantly to the publication of articles in this field in 2022. Hence, the analysis of research sources will assist researchers to quickly find relevant research articles from sources and facilitate the selection of a suitable journal to publish their manuscripts in the future.

Trend of countries' participation
In the field of current research, it is necessary to familiarize the readers with the countries that contribute the most in the publication of research. In this regard, Figure 8 illustrates the corresponding authors' country map for top 10 countries in the field of studying the characteristics of cementitious materials using AI techniques during last decade. As seen in the figure, China has contributed the most research with 65 articles published (14.22%), followed by Iran, India, United States, and Australia with 57 (12.47%), 39 (8.53%), 28 (6.13%), and, 24 (5.25%) respectively. These five countries have contributed to almost half of the studies in this area. After that, Vietnam, Turkey, South Korea, Greece, and Canada are also emerging countries in this field. In the next ranks, researchers in Egypt, Algeria, Saudi Arabia, Iraq, and Pakistan have also conducted research on the using AI techniques in advanced concrete technology. The variety of AI techniques used by the researchers of each country in their research is examined in Figure 9. It shows that Chinese researchers utilized a variety of AI techniques such as ANN, SVM, RF, XGB, PSO, and Fuzzy in their studies, while the research of other countries has been mostly related to ANN model. Hence, analyzing the map of active countries provides the opportunity to establish international research collaborations and exchange new ideas.

Co-authorship analysis
The density visualization of co-authorship with at least three articles published of an author in the field of using AI techniques in advanced concrete technology is shown in Figure 10. As can be seen, the active authors in this field are clearly recognizable. The largest contribution belongs to Panagiotis G. Asteris with the largest number of articles published, followed by Ali Behnood, and Emadaldin Mohammadi Golafshani. For a more detailed review, visualization of each author's publication year and co-authorship connectivity is depicted in Figure 11. In this figure, the color spectrum indicates the year of publication of the articles, so that the blue and red nodes represent the oldest and newest studies, respectively. Besides, the size of the nodes reflects the frequency of the number of articles published by each author. This figure shows that the most contribution in the last 10 years was made by Panagiotis G. Asteris. In addition, the most recent studies were conducted by Mahzad Esmaeili-Falak and Ahmed Salih Mohammed. This visualization analysis aids to better identify the active and outstanding researchers in the field of AI techniques in advanced concrete technology. F I G U R E 10 Density visualization of co-authorship.

Clustering AI techniques by title words
The first thing that comes to mind when searching for and dealing with an article is its title. It is conclusive that the title words of articles pertain to major themes of research. Therefore, mapping title words in the form of clustering provides the comprehensive view in the field of AI techniques used in advanced concrete technology. Figure 12 depicts the co-occurrences of title words with at least three co-occurrences for each title word in 457 collected articles by VOSviewer. The title words co-occurrence map shows the title words such as artificial neural network, concrete, compressive strength, mortar, fiber, and support vector machine have bigger nodes compare with the rest of the title words. Each title word is connected to other ones by a branch with a specific color, while branches of the same color indicate the co-occurrence of title words in the form of a cluster. In the following, top five title words related to the AI techniques used in advanced concrete technology will survey. Not to mention, full details all 457 collected studies including the classification of AI techniques in the prediction and optimization models and different scopes of advanced concrete technology are presented in Table B1.

Artificial neural network (red cluster)
The most widely used AI technique utilized in studies belongs to the title words of artificial neural network and its abbreviation (i.e. ANN). By filtering out other title words as seen in Figure 13, artificial neural network forms the biggest F I G U R E 12 Clustering co-occurrence of title words for artificial intelligence (AI) techniques used in advanced concrete technology.

F I G U R E 13
Clustering network of artificial neural network (ANN) along with number of co-occurrence for the highest frequencies.
cluster with several intra-clusters so that it connects with wide fields in advanced concrete technology. More specifically, artificial neural network, concrete, compressive strength, and fiber-reinforced concrete have depicted the most frequent co-occurrence. Selecting the top six title words based on the number of occurrence from the field of artificial neural network are; concrete (244), followed by compressive strength (172), fiber-reinforced concrete (42), mortar (30), flexural strength (29), and self-compacting concrete (24). 5.5.2 Support vector machine (pink cluster) Figure 14 shows clustering network of support vector machine along with number of co-occurrence for the highest frequencies. Support vector machine is another AI technique that has received more attention in the field of artificial neural network, compressive strength, concrete, random forest, and recycled aggregate concrete, respectively with number of occurrence of 37, 29, 25, 9, and 4. These results indicate that the most studies using support vector machine focused on concrete and property of the compressive strength.

Genetic algorithm (yellow cluster)
Genetic algorithm has taken a large scope in the field of using AI techniques in advanced concrete technology. In the past decade, researchers have mostly used genetic algorithm combined with artificial neural network (31 occurrences) to optimize and predict the properties and behavior of cementitious materials. For example, fields of such as concrete, compressive strength, flexural strength, bond strength, and shear strength were used in genetic algorithm (see Figure 15).

Particle swarm optimization (blue cluster)
Particle swarm optimization is usually used as one of the optimization techniques with prediction techniques such as artificial neural network. Figure 16 depicts the common title words in articles published with particle swarm optimization. The most common ones were artificial neural network and compressive strength with 22 and 14 occurrences, respectively.

5.5.5
Fuzzy system (orange cluster) Fuzzy system is one of the AI techniques that has the maximum concurrency with artificial neural network (19 occurrences) in the articles. As can be observed from Figure 17, selecting the top two title words based on the number of occurrence from the field of fuzzy system are compressive strength (13) and shear strength (4).

F I G U R E 14
Clustering network of support vector machine (SVM) along with number of co-occurrence for the highest frequencies.

F I G U R E 15
Clustering network of genetic algorithm (GA) along with number of co-occurrence for the highest frequencies.
F I G U R E 16 Clustering network of particle swarm optimization (PSO) along with number of co-occurrence for the highest frequencies.

F I G U R E 17
Clustering network of fuzzy along with number of co-occurrence for the highest frequencies.

Distribution of research hotspots on cementitious materials by title words
It is interesting to know which research hotspots in the field of advance concrete technology in AI techniques have received more attention from researchers during the last decade. Therefore, by scrutinizing the topics of the collected articles, the top five topics by number and percentage of participation are shown in Figure 18. As can be clearly seen, the investigation of various aspects of strength properties of cementitious materials using AI techniques was the most popular topic with 332 articles and a significant participation rate of 83.8%. With a more detailed look at the strength properties, it can be found that the most focus was on the compressive strength of cementitious materials with 216 articles (65% participation), followed by flexural, tensile, and shear strength respectively with 51, 33, and 32. Afterwards, "modulus of elasticity", "slump", "chloride penetration", and "fracture parameters" were four other top topics. Also, Figure 19 depicts the number of researchers' topics along with the reviewed approach in each year for five top topics. It clearly shows that in the last decade, the largest number of articles in the field of evaluating the strength properties of cementitious materials has been done with the ANN model. A closer look indicates that in recent years, topics of "modulus of elasticity" and "fracture parameters" along with "SVM" and "XGB" methods have received more attention from researchers. It is noteworthy that full details about the objective of all 457 collected studies are presented in Table B1. Consequently, knowing the topics studied so far helps researchers to get acquainted with various topics for future research.

Hybrid AI models with metaheuristic optimization algorithms
In advance concrete technology because of facing with complex optimization issues, it induces researchers to utilize metaheuristic optimization algorithms in predictive models in order to improve the strengths of single-mode model and cover its faults. It is clearly seen from Figure 20, the number of articles published in the field of hybrid AI models with metaheuristic optimization algorithms were few until 2019. From 2020, this trend started to growth, arriving at 23 and 21 articles in 2021 and 2022, respectively. Table 3 presents more details of these studies including cementitious material type, prediction model, optimization model, target output, and comparing the results of single and hybrid models. The most widely used of prediction models utilized in hybrid methods were ANN and SVM, while the most popular optimization methods were GA and PSO. The results of studies indicate that hybrid models achieved better results than single-mode model. Using hybrid AI with metaheuristic optimization algorithms is of growing interest worldwide, and therefore, further studies are expected to continue in upcoming future.

CONCLUSIONS, RECOMMENDATIONS, AND OUTLOOK
This study for the first time surveys the research trend of utilizing AI techniques in advanced concrete technology through the scientometric analysis approach from 457 articles published during the recent decade (Apr 2013-Mar 2023). The salient findings are as follows: 1. The trend of annual publications depicted a significant growth in article published during the years of 2021 and 2022 so that the publications in these years were almost equal with the total of articles published in the previous 8 years (2013-2020). It indicates the tendency of researchers to utilize these techniques as an affordable approach instead of the traditional laboratory procedures. 2. The analysis of research sources showed that about a third of the articles in this field belong to top three journals: "Constr Build Mater.", "Struct. Concr." and "Eng. Struct.". This analysis provides an opportunity for researchers to quickly access articles from sources and introduces the relevant journals in this scope to publish their manuscripts. 3. The top five countries with the highest amount of participation were China, Iran, India, United States, and Australia, respectively, which shows the dispersion of the participating countries in the continents of the world. Also, Vietnam, Turkey, South Korea, Greece, and Canada were emerging countries in this field. This visualization allows to better identify the active and outstanding researchers for research collaborations and exchange of new ideas in future research.
4. By clustering AI techniques, it was found that five main clusters of used techniques are ANN, SVM, GA, PSO, and Fuzzy. Among of these, ANN formed the biggest cluster with several intra-clusters. The in-depth analysis of research hotspots indicated that the most important objectives of these AI techniques were: "strength properties", followed by "modulus of elasticity", "slump", "chloride penetration", and "fracture parameters". The topic of strength properties with 4 sub-sets of compressive, flexural, tensile, and shear strength was the most popular topic. Also, in recent years, SVM and XGB methods have received more attention from researchers. The full details about the scope, AI techniques used and objective of all 457 collected studies are presented in the Table B1. 5. In advanced concrete technology, due to facing complex optimization problems, researchers have been more inclined to utilize optimization algorithms in combination with predictive models in recent years. Also, the most widely used of optimization algorithms were GA and PSO. Considering that hybrid models have outperformed than the single-mode model, it expects to see more studies in this scope in the upcoming years. 6. One of the strengths of some techniques such as GP is the formulation of parameters based on the mathematical relationships defined in the model. On the other hand, the ANN and XGB methods have received much attention due to its high ability to integrate with optimization techniques, which distinguish them from other methods. The evidence of that is the high number of studies conducted with these methods in the last decade.
Further research can be targeted to expand the present study. It is recommended examining the strengths and weaknesses of each technique can be considered as idea for future research. Besides, similar studies need to be planned at future intervals to follow the evolutionary trend of AI techniques in advanced concrete technology and to help monitor its development.

CONFLICT OF INTEREST STATEMENT
The author declares no conflict of interest.

DATA AVAILABILITY STATEMENT
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Text Acronym Text Acronym
In the field of artificial intelligence