Artificial Intelligence Empowering the Transformation of Building Maintenance: Current State of Research and Knowledge

Abstract

With the acceleration of urbanization and the continuous expansion of building stock, building maintenance plays a critical role in ensuring structural safety, extending service life, and promoting sustainable development. In recent years, the application of artificial intelligence (AI) in building maintenance has expanded significantly, markedly improving detection accuracy and decision-making efficiency through predictive maintenance, automated defect recognition, and multi-source data integration. Although existing studies have made progress in predictive maintenance, defect identification, and data fusion, systematic quantitative analyses of the overall knowledge structure, research hotspots, and technological evolution in this field remain limited. To address this gap, this study retrieved 423 relevant publications from the Web of Science Core Collection covering the period 2000–2025 and conducted a systematic bibliometric and scientometric analysis using tools such as bibliometrix and VOSviewer. The results indicate that the field has entered a phase of rapid growth since 2017, forming four major thematic clusters: (1) intelligent construction and digital twin integration; (2) predictive maintenance and health management; (3) algorithmic innovation and performance evaluation; and (4) deep learning-driven structural inspection and automated operation and maintenance. Research hotspots are evolving from passive monitoring to proactive prediction, and further toward system-level intelligent decision-making and multi-technology integration. Emerging directions include digital twins, energy efficiency management, green buildings, cultural heritage preservation, and climate-adaptive architecture. This study constructs, for the first time, a systematic knowledge framework for AI-enabled building maintenance, revealing the research frontiers and future trends, thereby providing both data-driven support and theoretical reference for interdisciplinary collaboration and the practical implementation of intelligent maintenance.

1. Introduction

In the context of accelerating global urbanization, the unprecedented growth in both newly constructed and existing buildings has placed building maintenance and management at the forefront of ensuring structural safety, extending service life, and promoting sustainable development [1]. Effective routine maintenance can reduce equipment downtime, prolong the lifespan of components, and enhance operational safety, thereby playing a critical role in preserving the functionality and value of buildings [2]. However, current maintenance practices still rely predominantly on traditional approaches such as manual inspections and reactive repairs [3,4]. These methods are not only labor-intensive and time-consuming but also heavily dependent on the subjective judgment and experience of personnel, making it difficult to achieve comprehensive and consistent inspection results [5]. As maintenance tasks become increasingly complex, teams are confronted with challenging environments and insufficient equipment, particularly in high-risk areas such as façade inspections and work at height, which further exacerbates safety hazards. Against this backdrop, the limitations of traditional maintenance approaches in terms of efficiency, precision, and safety assurance have become increasingly evident, rendering them insufficient to meet the growing demands of modern cities for building safety and sustainable management [4,6]. In response to these challenges, the application of emerging technologies in the field of building maintenance has become an important focus in both academic research and engineering practice [7]. In recent years, the rapid advancement of digital technologies, particularly artificial intelligence (AI), has opened new possibilities for innovation in maintenance strategies [8]. AI technologies enable automated monitoring of building conditions, intelligent diagnostics, and predictive maintenance, significantly improving the efficiency and accuracy of maintenance operations, reducing operational costs, and effectively preventing major safety incidents [9]. Therefore, against the backdrop of traditional maintenance methods being unable to meet the demands of modern building management, systematically advancing the application of intelligent technologies in building maintenance can not only significantly enhance efficiency and safety but also provide essential support for achieving high-quality and sustainable development in the industry [10,11].

Artificial intelligence (AI) has emerged in recent years as a rapidly advancing frontier technology, encompassing core subfields such as machine learning, deep learning, computer vision, and natural language processing [12]. Machine learning endows computers with the ability to improve performance through data-driven learning, while deep learning, through multi-layer neural networks, has greatly enhanced the accuracy of pattern recognition and prediction [13]. Computer vision enables machines to “interpret” the content of images and videos [14]. Breakthroughs in these core AI technologies have laid the foundation for their application in buildings maintenance. In recent years, scholars both domestically and internationally have conducted extensive research on building structural inspection, equipment operation and maintenance, and energy efficiency management, achieving notable progress [15]. For example, in the field of structural health monitoring, researchers have proposed automated defect recognition methods based on computer vision [16]. By utilizing unmanned aerial vehicles (UAVs) to capture images of building façades and applying deep learning-based image segmentation algorithms, these methods enable rapid and accurate identification and quantitative assessment of surface defects such as concrete cracks [17]. Such AI-powered inspection systems can also be integrated with Building Information Modeling (BIM) platforms or intelligent operation and maintenance systems to automatically generate façade health diagnostic reports and maintenance recommendations, thereby assisting facility managers in implementing timely preventive maintenance [18]. In the area of mechanical and electrical equipment operation and maintenance, AI-driven building management platforms integrate Internet of Things (IoT) sensor data with predictive analytics models to continuously monitor the operational status of key equipment such as heating, ventilation, and air conditioning (HVAC) systems and elevators, allowing for early prediction of potential failure trends [19,20].

Compared with traditional manual inspections and reactive repair methods, artificial intelligence (AI) offers significant advantages for building maintenance [21]. By leveraging sensors and intelligent algorithms, AI enables round-the-clock automated monitoring, greatly improving the timeliness and accuracy of hazard detection [22]. Through the processing of large volumes of historical and real-time data, AI enhances maintenance decision-making while reducing the costs of manual inspections and operations [23]. Applications such as deep learning-based image analysis have substantially improved both the efficiency and accuracy of defect identification [24]. Moreover, the integration of AI with technologies such as the Internet of Things (IoT) and Building Information Modeling (BIM) has accelerated the digital and intelligent transformation of building management systems, enabling performance optimization, energy reduction, and fault prediction [25]. Consequently, AI has become a central driving force for innovation and upgrading in the field of building maintenance [26].

Current research in this field primarily focuses on predictive maintenance of building facilities, intelligent anomaly detection and defect recognition, multi-source data integration and big data management systems, as well as intelligent operation and maintenance (O&M) management supported by data-driven decision-making [27,28]. These efforts have collectively advanced the automation and intelligence of building maintenance technologies [8]. However, the existing literature largely lacks systematic bibliometric and visualization studies that comprehensively examine the overall knowledge structure, thematic hotspots, and technological evolution of the field. As a result, no global and structured cognitive framework has yet been established, and it remains difficult to capture the latest developments in knowledge innovation and technological breakthroughs in a timely manner. To address this gap, this study adopts a bibliometric approach based on the Web of Science (WOS) Core Collection, selecting the relevant literature from the past 25 years concerning the application of artificial intelligence in building maintenance. The research systematically reviews the knowledge landscape, thematic hotspots, and developmental trajectory of the field. Using visualization tools such as ArcGIS 10.8.1 and VOSviewer (v1.6.20), we construct knowledge maps and research trend visualizations to comprehensively reveal the dynamic evolution of research hotspots and emerging frontiers. The specific objectives of this study are as follows:

RQ1: To reveal the evolution of research themes, the distribution of core authors and institutions, and the clustering characteristics of knowledge networks through bibliometric and scientometric analyses, thereby outlining the overall research landscape of the field.

RQ2: To critically examine the technical, managerial, and governance challenges encountered in the application of AI, identify major research topics, and explore their underlying knowledge structures.

RQ3: To analyze research hotspots and frontiers, while also identifying inherent limitations in existing studies and potential directions for future research.

This study systematically reviews the application of artificial intelligence (AI) in building maintenance through bibliometric analysis, encompassing technologies such as machine learning, deep learning, and computer vision. It summarizes the practices and potential of these technologies in predictive maintenance, health monitoring, defect detection, and energy efficiency optimization. From a systemic perspective, the research explores the integration of AI with intelligent operation and maintenance platforms, highlighting its value in proactive maintenance, risk prevention, and collaborative optimization. The ultimate aim is to provide valuable insights for future academic research and industry practice, thereby advancing the safety, cost-effectiveness, and sustainability of building maintenance.

2. Literature Review

Building maintenance refers to a series of planned, executed, supervised, and continuously improved activities aimed at ensuring the safety, reliability, energy efficiency, and regulatory compliance of a building and its mechanical and electrical systems throughout its life cycle [29,30]. During their service life, buildings inevitably undergo varying degrees of aging and performance degradation. Such deterioration may stem from untreated structural defects in the early construction phase, insufficient maintenance during operation, or long-term exposure to natural environmental factors such as wind and rain erosion or seismic activity [7]. Therefore, systematic maintenance activities are both necessary and urgent to slow the aging process and preserve the safety and functional continuity of buildings [4,31].

In general, building maintenance strategies can be categorized into planned maintenance and unplanned maintenance [4,32]. Planned maintenance is carried out under a predefined maintenance scheme, in which maintenance cycles and component replacement schedules are determined based on the operational conditions, performance requirements, and service life of equipment [33,34]. Its core objective is to prevent failures under normal operating conditions, maintain system stability and reliability, and reduce life-cycle costs [35]. This type of maintenance is characterized by its systematic and controllable nature, facilitating rational resource allocation and transparency in operation and maintenance processes [36]. In contrast, unplanned maintenance lacks prior arrangement and is typically triggered by user complaints, unexpected failures, or system malfunctions [37]. It is highly reactive, requires immediate response, and often incurs unpredictable maintenance costs and downtime losses, exerting substantial impacts on the continuity of building operations [7].

Within the framework of planned maintenance, preventive maintenance and corrective maintenance constitute two fundamental strategies [38]. Preventive maintenance centers on “proactive control,” aiming to reduce the likelihood of failures and slow performance degradation through early intervention [39]. It can be further divided into three forms. First, scheduled maintenance is performed at fixed time intervals, irrespective of the actual condition of the components [33]. Although simple to implement, it may lead to excessive maintenance [40]. Second, condition-based maintenance relies on periodic or continuous monitoring to collect equipment operation data and assess its health status, enabling more targeted maintenance decisions [41,42]. Third, predictive maintenance utilizes performance degradation models or trend analyses to estimate potential failure times and conduct interventions before failures occur, thereby optimizing maintenance timing while minimizing operational interruptions [43].

In contrast, corrective maintenance is performed after a component or system failure has occurred, with the aim of restoring functionality [44]. This strategy may take place as planned maintenance, in which repairs are deliberately scheduled once performance declines to a certain level, or as unplanned emergency maintenance, which involves addressing unpredictable failures immediately to prevent further damage [45,46]. Depending on risk levels and available resources, unplanned corrective maintenance may be classified into deferred repair—applied to non-critical components—and immediate repair, which is required for critical systems where safety and operational continuity are of paramount importance [47]. The specific technical processes of building maintenance are presented in Figure 1.

A substantial body of research has proposed integrated information technology solutions for the full life cycle of building projects, including Computerized Maintenance Management Systems (CMMS), Building Information Modeling (BIM), and Supervisory Control and Data Acquisition (SCADA) platforms [48,49,50]. These systems primarily rely on documentation and predefined rules, however, and demonstrate limited adaptability to complex and dynamically changing environments [51]. Maseda et al. (2021) developed a data-driven anomaly detection framework based on SCADA that can identify potential failures in advance without relying on fixed alarm thresholds, and automatically trigger maintenance work orders, thereby reducing downtime and improving equipment availability, demonstrating a feasible pathway toward predictive maintenance [52]. The review by Abideen et al. (2022) indicates that BIM is primarily used for information integration and model visualization during the operation and maintenance (O&M) phase, while its application in decision-making and value assessment remains insufficient, largely due to issues such as inconsistent information standards, difficulties in cross-departmental coordination, and the absence of dynamic update mechanisms [53]. Rui Calejo Rodrigues et al. (2023) proposed a CMMS selection model for Integrated Maintenance Systems (IMS), emphasizing that the system should support equipment asset registers and performance monitoring, and be capable of integration with BIM and IoT to enhance preventive maintenance and resource management efficiency [54].

However, in practical building maintenance, CMMS is often still limited to log recording and work order tracking, with underutilized functional potential [55].With the advancement of IoT and big data technologies, artificial intelligence (AI) methods—including machine learning, computer vision, reinforcement learning, and large-scale language models—have increasingly been incorporated into the field of building maintenance [56]. By automatically learning from massive multi-source datasets, these methods enable higher degrees of automation and predictive capability, shifting maintenance management from reactive mode to proactive prevention [57]. Matteo Interlando et al. (2024) integrated Vision Transformers with ConvNeXt for multi-class façade defect recognition, significantly improving fine-grained detection accuracy and enhancing inspection efficiency and automation [58]. Jihan Zhang et al. (2025) proposed an AI-enabled digital twin framework that integrates UAV data acquisition, deep learning-based defect detection, and GeoBIM semantic registration to achieve automated exterior wall defect recognition and centimeter-level 3D localization; field validation on high-rise buildings in Hong Kong demonstrated scalability and positional accuracy within the 1–5 cm range, offering an extensible digital twin solution for structural health monitoring and intelligent maintenance in complex built environments [59]. Siliang Chen et al. (2025) introduced a human–AI collaborative large-model framework for O&M of building energy systems, achieving complex fault diagnosis through few-shot learning and knowledge enhancement, with an accuracy of 96.3%, demonstrating the scalability of large models in energy operation scenarios [60]. Bo Pang et al. (2025) combined infrared thermography with an improved YOLOv8 to enhance the identification of concealed defects such as cracks and water seepage, improving detection accuracy by 18.7% and effectively overcoming the limitations of visible-light inspection [61]. Zhou Xiaoling et al. (2026) developed an automated multi-type façade defect detection system based on UAV imagery and deep learning. By constructing a multi-scale segmentation model using EfficientUNet++, the system achieved high-precision identification of cracks, spalling, water leakage, and glass breakage across different façade materials (including concrete, ceramics, masonry, and glass), verifying strong generalizability and robustness, and providing a scalable AI framework for intelligent façade inspection and urban infrastructure maintenance [62].

3. Materials and Methods

3.1. Data Sources and Retrieval Strategy

The data used in this study were obtained from the Web of Science (WOS) Core Collection database, provided by Clarivate Analytics. As one of the most widely used multidisciplinary literature retrieval tools worldwide [63]. WOS covers a broad range of disciplines, including natural sciences, social sciences, arts, and humanities [64]. With its comprehensive coverage, powerful search capabilities, and timely data updates, WOS ensures high-quality bibliographic support for this research.

To ensure the systematicity and accuracy of the literature retrieval process, this study followed the principles of “comprehensiveness, representativeness, and consistency” in designing the expanded set of keywords. The artificial intelligence-related terms were selected to encompass major research directions and algorithmic categories within AI, while the building maintenance-related terms corresponded to application domains such as building operation and maintenance, facility inspection, and structural health monitoring. In terms of logical structure, the retrieval strategy was constructed using Boolean operators: the OR operator was applied to expand synonymous and semantically similar terms, and the AND operator was used to define the intersection between AI-related technologies and the building maintenance domain, thereby ensuring thematic focus and relevance in the search results. The literature retrieval was conducted in Web of Science on 15 June 2025, using the Topic Search (TS) method, which matches terms within the title, abstract, author keywords, and KeyWords Plus fields. This approach ensured both the comprehensiveness and representativeness of the retrieved publications. The search query was designed as follows: TS = ((“artificial intelligence” OR “machine learning” OR “deep learning” OR “neural network*” OR “computer vision” OR “expert system*” OR “intelligent system*” OR “AI”) AND (“building maintenance” OR “facility maintenance” OR “structural maintenance” OR “building inspection” OR “facility inspection” OR “structural inspection” OR “infrastructure maintenance” OR “infrastructure inspection”)). The time span for the search was set from 1 January 2000 to 1 June 2025, in order to comprehensively capture the major developmental stages and the latest progress in the field of artificial intelligence applied to building maintenance research.

The initial search yielded a total of 503 relevant publications. To improve the accuracy of the analysis, the results were further restricted to English-language documents, reducing the number to 499. Subsequently, based on document type, only Articles and Proceeding Papers were retained as high-quality sources, leaving 460 papers. Finally, an independent screening and verification were conducted by a researcher: studies highly relevant to the topic were retained, while those that were tangential, incomplete, or duplicate were excluded. This process resulted in a final dataset of 423 highly relevant publications, which served as the basis for the subsequent bibliometric analysis. The detailed retrieval and screening process is illustrated in Figure 2.

3.2. Data Analysis

This study is based on 423 publications related to artificial intelligence and building maintenance, and conducts a systematic bibliometric analysis using Microsoft Office Excel 2021, ArcGIS 10.8.1, VOSviewer (v1.6.20), RStudio, and Bibliometrix v5.0.1 (R package). Specifically:

Microsoft Office Excel 2021 was used for basic statistical analyses, including annual publication trends, core journal distribution, research category statistics, and publication outputs by major authors and institutions. The extracted results provided foundational data support for subsequent visualization.

VOSviewer (v1.6.20) was employed to construct national collaboration networks, institutional collaboration networks, author collaboration networks, and keyword co-occurrence networks.

For the keyword analysis, the minimum occurrence threshold was set to 6, resulting in 91 high-frequency keywords selected from a total of 1795.

For the author collaboration analysis, the minimum publication threshold was set to 2, yielding 75 authors visualized from a pool of 1542.

For the institutional collaboration analysis, the minimum publication threshold was set to 3, resulting in 51 institutions visualized from 663.

In each network, nodes represent countries, institutions, authors, or keywords; node size reflects publication output or occurrence frequency, and connecting lines represent collaborative or co-occurrence relationships.

RStudio was used to generate the thematic map and the keywords distribution over time using the Bibliometrix package. The thematic map was configured with Field = All Keywords, Number of Words = 250, Minimum Cluster Frequency = 5, Number of Labels = 3, and the Walktrap clustering algorithm. The time range for the keyword temporal distribution analysis was set to 2005–2025.

ArcGIS 10.8.1 was primarily used to create the geographic distribution map of publications. A choropleth mapping approach was employed to visualize publication outputs across countries or regions, thereby illustrating the spatial distribution characteristics of research activities.

4. Results

4.1. Publication and Disciplinary Distribution Trends

Analyzing the annual publication trend provides researchers with a systematic understanding of the developmental dynamics of the field, thereby offering data support for formulating informed research strategies [65]. Figure 3 illustrates the annual number of publications in the field of artificial intelligence applications in building maintenance from 2000 to 2025. Over the past 25 years, the annual output has exhibited distinct stages of growth. Between 2000 and 2016, the annual number of publications remained at a consistently low level, typically ranging from 0 to 1 per year, with a cumulative total of only 12 publications by 2016. Research during this period was relatively fragmented, and the overall development of the field progressed at a slow pace.

Since 2017, the annual number of publications has shown steady growth. In 2017, 2018, and 2019, the numbers reached 4, 10, and 18, respectively, with the growth rate gradually accelerating. After 2020, the annual output increased sharply, rising to 34 in 2020, 47 in 2021, 61 in 2022, and 71 in 2023. By 2024, the annual number of publications peaked at 116. The cumulative total also grew rapidly, reaching 257 by 2023 and 423 by 2025. The relatively smaller number for 2025 can be attributed to the fact that data collection was completed in May. A polynomial regression analysis of the annual publication trend yielded a coefficient of determination (R²) of 0.9936, indicating an excellent fit and a highly consistent relationship between publication output and time. Overall, 2017 marked a turning point in research activity, with the field entering a phase of accelerated development after 2020 and reaching its peak in 2024. The rapid increase in publication volume reflects the deepening integration of AI technologies into the domain of building maintenance and the continuously rising research interest in this area.

The impact factor (IF) is one of the most important international indicators for evaluating the academic influence of journals. It not only reflects a journal’s status and role in scientific communication but also serves as a critical measure of its academic value. Likewise, the number of citations a paper receives is a key metric for assessing its quality and scholarly significance, as highly cited studies often indicate substantial influence and contribution to the field. Our analysis identified a total of 423 relevant publications distributed across 81 journals.

Table 1. Top Ten Core Journals by Publication Volume.

Ran	Institutions	NP	NC	IF (2024)
1	Journal Of Building Engineering	26	420	7.4
2	Buildings	25	165	3.1
3	Automation In Construction	13	1004	11.5
4	Applied Sciences-Basel	13	400	2.5
5	Energy And Buildings	10	388	7.1
6	Energies	10	139	3.2
7	Sustainability	9	54	3.3
8	Building And Environment	7	242	7.6
9	Ieee Access	7	108	3.6
10	Sensors	6	241	3.5

Note: NP = Number of Publications; NC = Number of Total Citations; IF = Impact Factor.

lists the top ten core journals in terms of publication output within this research domain.

In terms of publication volume, the Journal of Building Engineering ranks first with 26 papers, accounting for 6.1% of the total, far surpassing other journals and highlighting its prominent influence in the field of building maintenance and intelligent applications. Following closely, Buildings and Automation in Construction occupy the second and third positions with 25 and 13 papers, respectively, underscoring their central roles in research on building technologies and automation. Notably, Automation in Construction has an impressive impact factor (IF) of 11.5 and leads by a wide margin with 1004 citations, indicating not only a considerable number of publications but also exceptionally strong academic influence. This positions the journal as an authoritative platform for research on the application of artificial intelligence in building maintenance.

From the perspective of impact factor (IF2024), in addition to Automation in Construction and Building and Environment, both the Journal of Building Engineering and Energy and Buildings also demonstrate strong performance, each with an IF exceeding 7.0. This further confirms the international influence of these journals within the field. Overall, the top ten journals not only include mainstream outlets in architecture and construction but also encompass interdisciplinary fields such as intelligence and sustainable development. This highlights the multidimensional research hotspots of artificial intelligence applications in building maintenance and provides valuable references for future journal selection and research planning.

By conducting a statistical analysis of research categories for publications on artificial intelligence in building maintenance indexed in the Web of Science, we found that studies in this field are primarily concentrated in engineering and computer science. As shown in

Table 2. Top 20 Research Categories.

Rank	Web of Science Categories	NP	Rank	Web of Science Categories	NP
1	Construction Building Technology	146	11	Computer Science Theory Methods	23
2	Engineering Civil	137	12	Engineering Mechanical	19
3	Engineering Electrical Electronic	47	13	Telecommunications	19
4	Engineering Multidisciplinary	45	14	Engineering Industrial	17
5	Energy Fuels	44	15	Physics Applied	17
6	Computer Science Artificial Intelligence	40	16	Instruments Instrumentation	16
7	Computer Science Interdisciplinary Applications	39	17	Chemistry Multidisciplinary	15
8	Computer Science Information Systems	31	18	Automation Control Systems	13
9	Materials Science Multidisciplinary	26	19	Environmental Sciences	13
10	Green Sustainable Science Technology	24	20	Engineering Manufacturing	12

Note: NP = Number of Publications.

, the top five research categories are Construction Building Technology (146 papers), Engineering Civil (137 papers), Engineering Electrical Electronic (47 papers), Engineering Multidisciplinary (45 papers), and Energy Fuels (44 papers). The high publication volume within these categories indicates that research on AI in building maintenance is developing along a composite trajectory: with civil engineering and building technology as the core, driven by computer science, and extending into multiple domains such as energy, materials, and environment. This trend not only provides a solid disciplinary foundation for the deep integration of AI and building maintenance but also establishes both theoretical and practical support for future cross-disciplinary collaborative innovation.

4.2. Geographic Distribution of Publications

Analyzing the global publication landscape of artificial intelligence in building maintenance from a national perspective helps to reveal the level of research engagement and the leading forces in the international academic community [66]. Figure 4 illustrates the global distribution of research outputs in this field. Specifically, China ranks first with 80 publications, accounting for 18.7% of the total, followed by the United States with 52 publications (12.1%). Italy ranks third with 37 publications, while Canada (26), the United Kingdom (25), and India (24) closely follow. These countries, supported by well-established research systems, international collaboration networks, and policy incentives, have achieved rapid application and transformation of AI technologies in the field of building maintenance.

From the perspective of global regional distribution, Asian countries (such as China, India, Vietnam, Malaysia, South Korea, and Singapore) and European countries (including Italy, the United Kingdom, Germany, France, Poland, Sweden, and Norway) are particularly active, while North America (the United States and Canada) and Oceania (Australia) also demonstrate strong competitiveness. This distribution pattern indicates that research on artificial intelligence in building maintenance has moved beyond the traditional dominance of developed countries, evolving toward a new trend of global multi-centered development and regional collaboration. With the continuous iteration of relevant technologies and the deepening digital transformation of the construction industry, it is anticipated that more emerging economies will actively participate in international academic exchange and technological innovation, thereby driving the geographical distribution of this field toward greater diversification and dynamism.

To further reveal the global academic collaboration landscape of artificial intelligence in building maintenance, this study employed VOSviewer to construct a country-level scientific collaboration network (see Figure 5). In the network, each node represents a country, with node size indicating the number of publications produced by that country in this field. The links between nodes represent collaborative relationships between countries, with thicker lines signifying stronger and more frequent cooperation.

As shown in Figure 5, the collaboration network exhibits characteristics of multi-centered and clustered structures. Core-node countries are closely interconnected, whereas peripheral countries are generally engaged in limited or single-link collaborations. China, the United States, Italy, the United Kingdom, and Canada occupy central positions in the network, with larger node sizes and dense linkages, underscoring their dominant and pivotal roles in international academic cooperation. China not only leads in publication output but also demonstrates remarkable strength and breadth in international collaborations, maintaining partnerships with nearly all major research-active countries. This highlights its high level of openness and international influence. In contrast, countries such as Ukraine, Qatar, Nigeria, Greece and Lebanon display relatively limited collaboration, characterized primarily by unidirectional or small-scale partnerships. This reflects their comparatively lower level of participation in research on AI in building maintenance, which may be constrained by factors such as academic resources, policy environments, or institutional capacity.

Overall, international collaboration in this field demonstrates a globalized pattern centered on major technological powers such as China, the United States, and European countries, characterized by multi-regional cooperation and multi-level interconnections. Looking ahead, as international cooperation mechanisms continue to improve and the research capacity of emerging countries strengthens, the global collaboration network is expected to become more open and balanced, thereby providing a solid foundation of international cooperation for the sustainable development of this field.

4.3. Influential Authors

The analysis of prolific authors and core institutions helps to clarify the primary contributors in the field of artificial intelligence applied to building maintenance, the distribution of academic resources, and potential directions for future collaboration.

Table 3. Top 10 Authors by Publication Output.

Rank	Author	NP	NC	H-Index
1	Hoang, Nhat-Duc	7	175	45
2	Andrés J Prieto	5	44	15
3	Cao, Minh-Tu	4	73	20
4	D’Orazio, Marco	4	7	8
5	Gunay, Burak	4	101	16
6	Shen, Weiming	4	101	58
7	Yang, Chunsheng	4	101	22
8	Bernardini, Gabriele	3	7	25
9	Di Giuseppe, Elisa	3	7	20
10	El-Rayes, Khaled	3	13	32

Note: NP = Number of Publications; NC = number of total citations; H-Index = Hirsch Index.

presents the top ten authors in terms of publication output. In terms of the number of publications, Hoang Nhat-Duc ranks first with seven papers, demonstrating sustained research commitment and high activity in this domain, followed by Andrés J. Prieto (five papers) and Cao Minh-Tu (four papers). With respect to total citations, Hoang Nhat-Duc likewise leads with 175 citations, indicating that his research work has received widespread academic attention and recognition. Notably, although Shen Weiming has published only four papers, his total citation count reaches 101 and his H-index is 58—the highest among all listed authors—reflecting the exceptional academic value and influential impact of his contributions.

In summary, while publication quantity reflects an author’s level of research activity to some degree, citation count and the H-index provide a more accurate measurement of academic influence. The findings indicate that Hoang Nhat-Duc performs strongly in both research output and citation impact, whereas Shen Weiming achieves significant influence with comparatively fewer publications. The work of such high-impact scholars plays a crucial guiding role in advancing theoretical development and methodological innovation within this field.

To reveal the research collaboration patterns and academic network structure in the field of artificial intelligence applied to building maintenance, this study employed VOSviewer to visualize author publication data and generate an author collaboration network (Figure 6). In the figure, each node represents an author, while the edges indicate collaborative relationships. The colors differentiate distinct collaboration clusters, and the density and positioning of the edges reflect the closeness of collaboration and the core structure of the academic network.

From the overall structure of the collaboration network, the relationships exhibit a “multi-core with weak interconnections” pattern, meaning that several scholars have formed relatively stable collaboration clusters, while connections between these clusters remain comparatively limited. Hoang Nhat-Duc, Shen Weiming, and Gunay Burak occupy central positions in the network, demonstrating broad collaborative reach and strong academic influence, indicating their organizational capacity and cohesive roles in advancing the field. For instance, Hoang Nhat-Duc forms a stable collaborative cluster with Prieto A. J. and Nguyen Ngoc-Mai, showing both high productivity and strong scholarly impact. Shen Weiming’s collaboration network is highly concentrated, involving close cooperation with scholars such as Svidt Kjeld and Tinoco-Ruiz J. G., highlighting his significant role in facilitating international academic collaboration. Additionally, Gunay Burak exhibits dense network linkages with collaborators such as El-Rayes Khaled and Haddad Assed, reflecting active efforts to expand collaborative boundaries within the research domain.

In conclusion, core authors form academic communities through stable collaborative relationships, playing a central role in driving the coordinated development of research and enhancing the influence of the field. Future research should further strengthen cross-team and cross-regional collaboration to promote knowledge integration and methodological innovation, thereby advancing the deeper application of artificial intelligence technologies in building maintenance.

4.4. Institutional Contributions

Institutional analysis helps to reveal the research landscape and core strengths within the field, providing insight into the academic influence and collaboration networks of leading institutions [67]. This information serves as a valuable reference for future research planning and opportunities for inter-institutional collaboration.

Table 4. Top 10 Institutions by Publication Output.

Rank	Institutions	NP	NC	AC
1	Duy Tan University	11	231	21
2	The Hong Kong Polytechnic University	10	127	13
3	Islamic Azad University	7	71	10
4	National University of Singapore	6	372	62
5	Shanghai Jiao Tong University	6	163	27
6	University of Illinois at Urbana-Champaign	6	20	3
7	Università Politecnica delle Marche	6	29	5
8	Chongqing University	5	69	14
9	The Hong Kong University of Science and Technology	5	360	72
10	Pontificia Universidad Católica de Chile	5	20	4

Note: NP = Number of Publications; NC = number of total citations; AC = Average Citations per Publication.

lists the top ten institutions ranked by publication volume, providing a clear overview of the research distribution and academic influence of each institution.

In terms of publication output, Duy Tan University ranks first with 11 articles, indicating sustained research investment and a high level of scholarly activity in this field. It is followed by The Hong Kong Polytechnic University (10 articles) and Islamic Azad University (7 articles), both of which demonstrate strong organizational capacity and research productivity in advancing studies on intelligent building maintenance. However, when considering average citation impact, The Hong Kong University of Science and Technology (AC = 72) and the National University of Singapore (AC = 62) emerge as the most influential institutions. Although these two institutions have published only five and six articles, respectively, their total citation counts reach 360 and 372, far surpassing other institutions. This indicates that their research outputs possess high scholarly value and citation influence, reflecting their critical role in theoretical innovation and methodological advancement within the field.

To further elucidate the international collaboration landscape in the application of artificial intelligence to building maintenance, this study employed VOSviewer to visualize institutional publication data. By setting the minimum publication threshold at three papers per institution, an institutional collaboration network map was generated (Figure 7). In this network, each node represents an institution, with node size corresponding to publication volume, connecting lines indicating collaborative relationships, and colors distinguishing different collaboration clusters.

From the overall network structure, institutional collaboration exhibits a distinctly multi-centered cluster pattern. High-output institutions tend to form close cooperative networks around specific regions or research alliances. Duy Tan University and Islamic Azad University form the most prominent cluster (highlighted in red), demonstrating a strong regional synergy between institutions in the Middle East and Southeast Asia. This cluster includes multiple universities from Arab and Asian regions, such as King Khalid University and Prince Sattam Bin Abdulaziz University, characterized by high collaboration strength and dense network connectivity. The Hong Kong Polytechnic University occupies a core position within the green cluster, linking with institutions such as The Hong Kong University of Science and Technology, Shenzhen University, and Southeast University. This indicates that universities in mainland China and Hong Kong have established relatively stable regional cooperation alliances in research on intelligent building maintenance. In contrast, some institutions—such as the University of Illinois and Univ Politecn Marche—are relatively isolated within the network, reflecting limited collaboration ties and suggesting that research resources in this field remain unevenly distributed across regions.

In summary, institutional collaboration in the field of artificial intelligence for building maintenance currently shows a trend centered in Asia, characterized by multi-polar linkages and a gradually increasing degree of internationalization. Strengthening cross-regional research collaboration and knowledge exchange will be essential for enhancing the global level of intelligent building maintenance, accelerating the translation of research outcomes into practice, and promoting broader technological application.

4.5. Knowledge Structure and Clustering Characteristics

By analyzing high-frequency keywords, it is possible to gain deeper insights into the main research directions, technical approaches, and research hotspots in the field [68]. In this study, VOSviewer software was employed to conduct a co-occurrence network analysis of the keywords extracted from the core literature (Figure 8). The co-occurrence relationships among high-frequency keywords effectively map the implicit knowledge structure of the discipline and reveal the evolutionary trajectory of research hotspots [69]. A total of 1795 high-frequency keywords were identified, and four major thematic clusters were generated using the Louvain clustering algorithm. These clusters maintain relative thematic independence while simultaneously achieving close coupling within a complex network structure through cross-domain nodes.

Cluster 1: Intelligent Construction and Digital Twin Integration

This cluster centers on the keywords “artificial intelligence,” “building information modeling (BIM),” “digital twin,” and “internet of things (IoT),” revealing a trend toward deep integration of digitalization and intelligent technologies in the building sector. Current research is evolving from the digitalization of isolated subsystems to the real-time interconnection of multiple platforms, with BIM and digital twins serving as central carriers for building life-cycle data, while IoT provides continuous sensing and data input [70,71]. The introduction of artificial intelligence enables a closed-loop workflow of “perception–analysis–decision–execution,” shifting maintenance practices from experience-driven to knowledge-driven paradigms. The key drivers behind this trend include increasing industry demands for energy efficiency optimization and operational reliability, as well as policy initiatives supporting smart city development [72]. However, challenges remain regarding data standardization, interface interoperability, and model compatibility, which constrain large-scale implementation. Future research is expected to focus on open platforms, semantic interoperability, and knowledge graph development, thereby establishing a unified data foundation for life-cycle intelligent maintenance management.

Cluster 2: Predictive Maintenance and Health Management

This cluster is characterized by the keywords “predictive maintenance,” “fault detection,” and “condition monitoring,” reflecting the ongoing shift in building maintenance from passive response to proactive prediction. By integrating multi-source sensor data with machine learning algorithms, predictive maintenance can identify potential equipment failures in advance, enabling optimal resource allocation and risk prevention [73]. The emergence of this research direction is driven by the dual pressures of increasing building facility aging and rising operational costs [4]. The research frontier is evolving from single-equipment health assessment toward system-level operational optimization, focusing on the integrated balance among energy consumption, service life, and reliability. Future breakthroughs are expected to lie in the development of generalized and interpretable intelligent models that can be deployed robustly in complex building environments [74]. Moreover, through deep coupling with BIM and IoT, it will be possible to establish a truly “self-sensing, self-diagnosing, and self-decision-making” building maintenance system.

Cluster 3: Algorithmic Innovation and Performance Evaluation

This cluster centers on the keywords “machine learning,” “optimization,” and “performance,” reflecting the ongoing evolution of algorithmic frameworks and the deepening of intelligent maintenance methodologies. The progression from early statistical learning to deep neural networks, reinforcement learning, and transfer learning has continuously improved prediction accuracy and operational efficiency [75]. In recent years, the research focus has shifted from model selection to performance optimization and enhanced interpretability, with emerging directions including uncertainty quantification, multi-task learning, and knowledge fusion [76]. The significance of algorithmic innovation lies in advancing intelligent maintenance from being merely “usable” to becoming truly “trustworthy.” Future development should strengthen research on model transparency and computational sustainability, and explore lightweight, deployable AI models that can support real-time decision-making and large-scale application in building operation and maintenance.

Cluster 4: Deep Learning-Driven Structural Detection and Automated Operation and Maintenance

This cluster centers on the keywords “deep learning,” “CNN,” “computer vision,” and “structural health monitoring (SHM),” highlighting the extensive application of visual intelligence in structural inspection and condition monitoring. Automated detection methods based on convolutional neural networks and point cloud analysis enable efficient, non-contact crack identification, damage localization, and surface anomaly detection, significantly improving the real-time performance and accuracy of building safety monitoring [77]. The emergence of this research direction has benefited from advancements in high-resolution imaging and multimodal sensing technologies; however, limited availability of annotated data and insufficient model generalization continue to restrict large-scale deployment in complex real-world environments [78]. Future studies are expected to integrate generative AI, active learning, and knowledge distillation techniques to develop sustainable learning frameworks for structural health monitoring, achieving an intelligent transition from localized inspection to holistic operational management.

Together, the four clusters constitute the core knowledge structure of artificial intelligence in the field of building maintenance: Cluster 1 provides the data and platform foundation; Cluster 2 reflects the shift in maintenance logic toward intelligent prediction; Cluster 3 offers algorithmic support; and Cluster 4 drives practical application and deployment. The overall research evolution follows the trajectory of digital integration → intelligent sensing → algorithmic empowerment → automated operation and maintenance. The general trend indicates that the field is transitioning from tool development toward a new stage of systematic, sustainable, and autonomous intelligent maintenance. Future research should further strengthen integration across clusters, build an open and shared data ecosystem, and establish interpretable intelligent decision-making systems to enable adaptive management across the full life cycle of buildings.

4.6. Thematic Evolution and Trend Analysis

To reveal the developmental trajectory of research themes and the migration of hotspots in the field of artificial intelligence for building maintenance, we employed RStudio and the bibliometrix visualization tool to systematically construct the keyword temporal distribution heatmap (Figure 9A) and the thematic structure evolution diagram (Figure 9B). The keyword temporal distribution heatmap (Figure 9A) provides an intuitive representation of the emergence and evolution of various research themes. Meanwhile, the thematic structure evolution diagram (Figure 9B), based on the dual dimensions of centrality and density, delineates the structural characteristics and evolutionary pathways of core, foundational, and emerging themes within the field.

4.6.1. Dynamic Evolution of Keyword Temporal Distribution

As shown in Figure 9A, the temporal visualization of keywords clearly reveals the dynamic evolutionary trajectory of research themes in this field. Between 2005 and 2015, related studies primarily focused on foundational themes such as “Fault Diagnosis” and “Feature Extraction.” During this period, keywords appeared infrequently and were dispersed, indicating that the penetration of artificial intelligence technologies into building maintenance was still at an early stage. After 2016, with the introduction of emerging technologies such as machine learning and deep learning, both the variety and density of keywords gradually increased. In particular, keywords such as “Machine Learning,” “Artificial Intelligence,” and “Deep Learning” began to appear with high frequency after 2018, reflecting the consolidation of the technological foundation of the field and a gradual shift in research attention toward the practical application of intelligent algorithms. Moreover, compound keywords such as “Digital Twin,” “IoT,” and “Energy Efficiency” have risen significantly since 2020, demonstrating the advancement of building maintenance research from single-technology approaches toward multi-technology integration and system-level intelligence.

From the distribution on the right side of the heatmap, it is evident that during the period from 2021 to 2025, the concentration of keywords further intensified, with high-frequency themes represented by “Digital Twins,” “Predictive Artificial Intelligence Maintenance,” and “Anomaly Detection.” This indicates that the discipline has entered a new stage characterized by the deep integration of technologies and application scenarios. The vertical hierarchical changes in the keywords clearly illustrate the shift in research focus from fundamental data sensing and monitoring toward advanced objectives such as full life-cycle intelligent operation and maintenance, energy efficiency management, and system-level optimization. This trend not only reflects the upgrading of industry demands but also signals that the future of building maintenance will increasingly rely on the diversified empowerment of artificial intelligence.

4.6.2. Thematic Structure and Evolutionary Pathways

Figure 9B presents the thematic evolution structure diagram based on co-occurrence analysis, further revealing the structural relationships and evolutionary directions of different research themes in the field of artificial intelligence for building maintenance. In this two-dimensional coordinate system, the horizontal axis represents thematic relevance (centrality), while the vertical axis represents thematic development (density). The distribution across the four quadrants clearly delineates the hierarchical importance of research themes as well as the emerging frontiers and hotspots.

The upper-right quadrant represents the “Motor Themes,” exemplified by keywords such as “artificial intelligence,” “performance,” and “optimization.” These themes exhibit both high centrality and high density, indicating that they are the core driving forces in the field. Themes in this quadrant are not only highly connected and influential but also demonstrate a high level of technological maturity, serving as critical nodes for advancing the intelligence and automation of building maintenance. Related research focuses on performance enhancement, energy optimization, and intelligent decision-making enabled by AI, reflecting the parallel progression of technological innovation and industry application.

The lower-right quadrant represents the “Basic Themes,” including keywords such as “machine learning,” “predictive maintenance,” “deep learning,” “buildings,” and “classification.” These themes exhibit high relevance and moderate development, serving as the methodological and practical foundation for the application of artificial intelligence in building maintenance. Themes in this quadrant often act as bridges for knowledge transfer and innovation expansion, supporting the overall progression of the field from theoretical research to practical implementation.

The upper-left quadrant represents the “Niche Themes,” dominated by keywords such as “fuzzy inference system” and “axial compressive behavior,” reflecting in-depth exploration within specialized subfields. Although these themes exhibit relatively low centrality, they possess unique advantages in addressing specific technical challenges and application scenarios. As such, they hold non-negligible value in fostering the diversified development of the technological framework within the field.

The lower-left quadrant represents the “Emerging or Declining Themes,” including keywords such as “heritage buildings,” “climate change,” and “cultural heritage,” which exhibit relatively low density and centrality. These themes largely correspond to emerging fields or cross-disciplinary explorations, with current research still at an early or transitional stage but demonstrating considerable development potential. With the growing prominence of issues such as green buildings and cultural heritage preservation, these themes are expected to evolve into new hotspots of future growth.

5. Discussion

5.1. Major Research Findings (Knowledge Structure)

By combining the temporal perspective of Figure 9A with the structural perspective of Figure 9B in Section 4.6, the bibliometric analysis of literature on artificial intelligence in building maintenance management reveals that the development of this field is characterized by dynamic complexity, with various research directions and hotspots emerging at different stages. Some new research directions expand upon existing themes, while others deepen the investigation of established issues, indicating that the evolution of research focuses and directions is inherently dynamic. Although several review studies have examined the application of Building Information Modeling (BIM) in operations and maintenance or suggested the use of AI to improve facility management, these works have not succeeded in establishing a coherent knowledge framework, making it difficult for readers to access key information in a concise and intuitive manner. Therefore, it is necessary to construct a new, valuable, and comprehensive theoretical knowledge framework for research on AI in building maintenance, in order to present a holistic view of the field and to provide guidance for future studies.

Figure 10 presents the integrated knowledge framework of AI in building maintenance management research, clearly illustrating the interrelationships among major research themes, the primary stakeholders, and potential future research directions, thereby providing valuable references for subsequent studies.

Publication Analysis: The bibliometric results indicate that the number of publications in this research field has shown a continuous upward trend, with research topics becoming increasingly diverse. This reflects the growing scholarly attention to the field and suggests a promising outlook for future development. Basic publication statistics demonstrate that related studies began to rise significantly in the late 2010s, with a sharp increase in annual publications after 2018. In terms of major academic journals, Journal of Building Engineering, Buildings, and Automation in Construction are the top three in terms of publication volume. The most prominent subject categories include Civil Engineering and Building Technology, Computer Science, and Energy and Environmental Sciences, reflecting the interdisciplinary and cross-domain development trend of the field. multidisciplinary and cross-disciplinary development trajectory.

Collaboration Analysis: The collaboration network reveals the dynamic cooperative relationships among research institutions, countries/regions, and authors. Scholars from all major countries are actively engaged in this field, and the leading contributors have been identified to provide references for researchers seeking potential collaborators and appropriate research domains. China, the United States, Italy, Canada, and the United Kingdom are the countries with the closest collaboration. More specifically, the leading institutional participants include Duy Tan University, The Hong Kong Polytechnic University, and Islamic Azad University. Among individual scholars, Nhat-Duc Hoang, A. J. Prieto, and Minh-Tu Cao are identified as the most active collaborators.

The outer ring displays the set of research keywords: the keyword co-occurrence patterns reveal the major research hotspots and evolutionary trends in the field, outlining the core themes and potential turning points across different periods. The analysis indicates that the thematic structure is primarily organized around four core clusters: “intelligent construction and digital twin integration,” “predictive maintenance and health management,” “algorithmic innovation and performance evaluation,” and “deep learning–driven structural inspection and automated operation and maintenance,” together forming the central lexical network of this research domain.

5.2. Thematic Evolution Trends and Emerging Hotspots

By integrating the keyword evolution pathways with the thematic strategic diagram, it becomes evident that research themes in the field of artificial intelligence for building maintenance management are undergoing a progressive upgrade—from “passive monitoring” to “proactive prediction,” and further toward “intelligent decision-making” and “system integration.” In recent years, “digital twins,” “energy efficiency management,” and “multi-technology integration” have emerged as new research hotspots, while novel application scenarios such as green sustainability, cultural heritage, and climate-adaptive buildings have continuously surfaced. The emergence of these hotspots has not only advanced the levels of intelligence and coordination in building maintenance but has also provided more efficient and scientifically grounded solutions for energy conservation and emission reduction, as well as for the preservation of historic buildings, thereby driving the industry toward greater intelligence and sustainability.

5.2.1. Integration with Digital Twin Technology

With the integration of Artificial Intelligence (AI) and Digital Twin technology, building maintenance is shifting from traditional reactive monitoring toward proactive prediction [19,79]. A systematic review by Siyuan Chen et al. indicates that AI techniques such as supervised learning, deep learning, and reinforcement learning can enhance digital twin models to enable fault prediction and data-driven maintenance strategies [80]. For example, Chen et al. (2023) note that digital twins provide accurate equipment condition recognition and failure forecasting for predictive maintenance (PdM), supporting a transition from “passive response” to “active prevention.” This shift allows for optimization of maintenance scheduling, reduction of downtime, and improvement in system reliability and economic efficiency. However, current industrial applications are still constrained by challenges including scalability, heterogeneous data integration, and model deployment complexity [81]. Future research should continue to advance standardized modeling frameworks, mechanisms for multi-source data fusion, model interpretability, and real-time reliability validation, in order to facilitate the practical implementation of digital twin technology within building maintenance [82].

5.2.2. Energy Efficiency Management

AI is playing an increasingly significant role in building energy efficiency management [83]. According to the review by Bajwa et al. (2024), AI technologies have been widely applied in Smart Building Management Systems (SBMS) for HVAC optimization, lighting control, and renewable energy forecasting, achieving substantial energy savings (with consumption reductions of 20–50%) [84]. Algorithms such as reinforcement learning and deep learning have demonstrated greater flexibility than traditional rule-based approaches by adaptively adjusting control strategies in response to real-time sensor data. Future development requires the establishment of unified frameworks and strengthened interdisciplinary collaboration, along with expanded experimental validation and real-world project implementation, to promote the industrial-scale application of AI in building energy management [85]. By integrating dynamic control with predictive maintenance, energy management can evolve from a narrow focus on “energy-saving” to a more comprehensive approach encompassing “energy efficiency, emission reduction, and sustainable operation [86].”

5.2.3. Multi-Technology Integration

The trend of intelligent building maintenance is increasingly characterized by the integrated coordination of multiple technologies. Under the combined influence of BIM, the Internet of Things (IoT), 5G, virtual/augmented reality (VR/AR), and AI-driven robotics, building maintenance capabilities are being significantly expanded to form comprehensive intelligent maintenance solutions [87,88]. Wong et al. (2025) note that IoT enables real-time data acquisition and monitoring in buildings, VR/AR and digital twins support immersive visualization and simulation, and AI algorithms provide predictive analytics for maintenance decision-making through machine learning [89]. Specifically, Building Information Modeling (BIM), as a collaborative digital platform, ensures shared situational awareness of building conditions across departments; meanwhile, AI techniques such as generative design can search for optimal solutions under constraints such as cost and energy efficiency [90,91,92,93]. The combination of these tools enables scenario-based simulation and continuous condition sensing, thereby optimizing maintenance pathways and enhancing sustainability.

The integration of robotics and AI provides solutions for building inspection in inaccessible or hazardous environments [94,95]. Maryam Kouzehgar et al. proposed a self-reconfigurable façade-cleaning robot equipped with a deep convolutional neural network (CNN) for crack detection. Using a CNN model implemented in TensorFlow, the system analyzes real-time video captured from the robot’s onboard camera to achieve high-precision (approximately 90%) automatic crack identification and obstacle avoidance, significantly improving the safety and automation level of high-rise glass curtain wall cleaning [96]. The convergence of diverse technologies is thus a core pathway for the systematic development of intelligent maintenance. By integrating BIM, IoT, AI, and robotic automation into a unified platform, fragmented maintenance operations can be transformed into a coordinated and intelligent system [97].

5.2.4. Applications of Artificial Intelligence in Green Buildings

Under the dual-carbon policy targets, AI-empowered green and sustainable buildings have become an emerging application scenario in building maintenance [98,99]. Modern green buildings not only aim for energy conservation but also emphasize full life-cycle sustainability and occupant health and comfort [100]. AI technologies are increasingly applied to optimize the operational performance of green buildings [101]. For example, Prabhu Rajaram et al. employed algorithms such as KNN, SVM, Random Forest, XGBoost, AdaBoost, and Naive Bayes to enhance energy efficiency and reduce carbon footprints. Their approach effectively addressed challenges such as inaccurate energy consumption prediction, limited real-time indoor environmental optimization, and insufficient decision support in sustainable smart buildings, achieving coordinated improvements in both energy performance and occupant comfort [102]. Furthermore, AI has been utilized to support green building design and certification decision-making. Mansouri et al. used machine learning models (including decision trees, support vector regression, and XGBoost) to predict LEED green building certification scores, with the XGBoost model achieving the best performance (R² ≈ 0.81), demonstrating the effectiveness of machine learning for LEED credit score prediction [103]. Future research may focus on developing intelligent design assistants aligned with dual-carbon objectives, embedding carbon simulation into BIM platforms, and leveraging generative and reinforcement learning approaches to promote proactive carbon reduction throughout the building life cycle [104].

5.2.5. Applications of Artificial Intelligence in Cultural Heritage Preservation

For historical buildings and cultural heritage, the introduction of AI technologies provides new approaches to monitoring and risk management in this traditionally conservation-oriented field [105]. As heritage structures are exposed to long-term climate change and environmental pollution, traditional manual inspections often fail to detect slow and subtle deterioration processes in a timely manner [106]. By deploying IoT sensor networks in historic buildings to monitor environmental and structural parameters such as temperature, humidity, stress, strain, and air quality, and applying AI-based data analysis, real-time early warning and predictive assessment can be achieved [107]. Neeraparng Laohaviraphap et al. report that IoT enables continuous acquisition of environmental and structural condition data from heritage sites, while AI performs pattern analysis and risk prediction, shifting heritage conservation from passive restoration to proactive prevention. For example, when abnormal moisture levels are detected in wall materials, AI can predict risks such as mold growth or structural weakening and issue early alerts [105]. At the same time, deep learning-based image recognition technologies enable automated identification of surface cracks and weathering, facilitating fine-grained monitoring. Sida Xu and Haonan Chen et al. integrated digital twin technology with an improved YOLO object detection algorithm by using UAV panoramic scanning to construct a high-precision 3D model of the Nanjing Qixia Temple Pagoda. They then applied YOLO within the digital twin environment to detect crack and corrosion damage, enabling multi-angle and cross-season automated defect identification, achieving significantly higher accuracy and efficiency compared with traditional manual inspection [108]. AI is transforming cultural heritage preservation into a proactive and scientific process: through intelligent monitoring and simulation-based analysis, it enhances the foresight, precision, and coverage of conservation actions, thereby greatly improving the capacity for long-term preservation of valuable cultural assets [109].

5.2.6. Applications of Artificial Intelligence in Climate-Resilient Architecture

With the increasing frequency of extreme weather events and natural disasters, building resilience and climate adaptability have become emerging focal points, in which AI plays a critical role [110]. By applying machine learning to meteorological, atmospheric, and structural response data, AI-driven disaster prediction and response capabilities have rapidly advanced in extreme event scenarios [111]. For example, Alessandro Rocchi et al. analyzed data from 331 municipalities in the Emilia–Romagna region of Italy and, after data standardization and PCA dimensionality reduction, applied K-means clustering to classify the region into four risk categories. They then incorporated variables such as seismic acceleration, building condition, and population exposure to generate comprehensive risk labels, providing a scientific basis for government prioritization of disaster prevention interventions [112]. According to a review by Vagelis Plevris, AI has already been employed in fields such as earthquake early warning, structural health monitoring, damage assessment, and multi-hazard risk mapping, improving early warning accuracy and enhancing structural resilience. However, challenges remain, including uncertainties in climate models, insufficient reliability of intelligent systems under extreme conditions, and difficulties in integrating such systems into regulatory frameworks [113]. Future research directions may explore dynamic building envelopes (e.g., adaptive shading systems, deployable or morphable structural assemblies) and real-time climate control strategies enabled by edge computing, thereby coupling AI-based predictive analytics with automated building control [114,115]. This integration is expected to enhance the capability of buildings to autonomously anticipate, respond to, and recover from extreme climate impacts [116].

5.3. Challenges and Major Issues

Overall, the engineering application of artificial intelligence in building maintenance management still faces numerous challenges. First, data sources related to building maintenance are highly diverse yet lack unified standards [117]. Data quality is uneven, formats are inconsistent, and effective sharing remains difficult. For instance, large volumes of building operation and maintenance data have not been systematically collected or stored, and there are almost no open-access databases for building maintenance (with only a few available for energy consumption) [118]. This severely constrains the efficient integration of heterogeneous multisource data and limits the widespread adoption of AI models. Second, current AI models often struggle to adapt to variations across building types, climatic conditions, and operational scenarios, reflecting insufficient generalization capability [119]. Due to the scarcity and fragmentation of training datasets, the transferability of models across climate zones and different building structures remains limited [120]. On the one hand, differences in operating conditions between regions and buildings lead to poor performance of “well-trained” models when applied in new environments [121]. On the other hand, the “black-box” nature of AI systems renders their decision-making processes opaque, making it difficult for building operators to understand the rationale behind maintenance decisions [122]. This lack of interpretability undermines user trust in AI-generated recommendations and constitutes a significant barrier to real-world adoption in facility maintenance [123]. Third, many existing building maintenance platforms, sensor monitoring networks, and Building Information Modeling (BIM) systems operate in isolation without effective integration, restricting AI from accessing comprehensive and standardized data support. Moreover, different generations of equipment and software often lack standardized interfaces; for example, legacy building systems are not interoperable with modern IoT edge nodes, preventing seamless data exchange [11]. Such information silos increase the complexity of AI deployment, necessitating extensive customized development to bridge disparate systems. Finally, disparities across countries and regions in terms of resource investment, policy support, and technology transfer capacity further exacerbate the uneven global development of intelligent building maintenance.

In the future, efforts should focus on advancing data standardization, enhancing the generalizability and interpretability of models, and optimizing pathways for system integration. At the same time, policies and resource allocation should be reinforced in alignment with regional contexts to ensure the high-quality implementation and sustainable development of artificial intelligence technologies in building maintenance management.

6. Conclusions

Over the past decade, the application of artificial intelligence in building maintenance has exhibited rapid expansion and, driven by the maturation of technologies such as the Internet of Things (IoT), digital twins, 5G, and edge computing, is expected to experience even stronger growth momentum in the coming years. With the implementation of predictive maintenance, unmanned visual inspection, intelligent optimization of Building Management Systems (BMS), and coordinated management of energy and carbon, the industry faces increasingly urgent demands for effective evaluation frameworks encompassing accuracy, reliability, interpretability, interoperability, and cost-effectiveness. This study seeks to systematically reveal the latest developmental trajectories and methodological evolution in the “AI–building maintenance” domain by identifying the most influential research sources, research hotspots, and emerging topics. Using scientific mapping techniques, we constructed knowledge graphs that delineate the theoretical foundations of the field—such as condition monitoring and remaining service life prediction, mechanism–data fusion, and the integration of Facility Management (FM) with BIM/digital twins—as well as recent integrated framework studies. The observed trend highlights the growing prominence of intelligent and sustainable maintenance, with a continuous increase in publications across interdisciplinary areas, including facility management, architecture and environmental engineering, computer science, and energy and environmental sciences, as indicated by the content analysis.

We emphasized the dynamic evolution of the role of artificial intelligence in building maintenance: shifting from passive inspection to proactive prediction, and from isolated optimization to system-level integration (BIM/digital twin/IoT/CMMS collaboration). This progression demonstrates its potential to enhance maintenance decision-making, energy efficiency management, and risk control through predictive capabilities and real-time data analytics. The four thematic clusters are highly interrelated and intricately coupled: Cluster 1 provides the data and semantic foundation; Cluster 2 constructs a closed loop of “monitoring–diagnosis–prediction–intervention”; Cluster 3 functions as the engine of methodological innovation and evaluation standardization; and Cluster 4 connects the perception–execution chain and feeds new evidence back into the digital twin for continuous learning. Based on this thematic structure, this paper proposes a comprehensive evaluation framework that integrates artificial intelligence with traditional facility management and reliability engineering approaches. Anchored in a digital twin platform, the framework spans data governance, model development and interpretability, business process orchestration, value assessment, and risk control, thereby establishing an end-to-end management and decision-making cycle. This ensures transparency, traceability, and fairness in technological implementation, while enhancing the feasibility of building maintenance management and fostering social trust.

Several critical gaps remain in the existing literature: insufficient data standardization and interoperability, particularly manifested in semantic inconsistencies across BIM, digital twins (DT), and IoT systems; limited generalization and interpretability of models, with inadequate transferability across building types, climatic conditions, and operational contexts; challenges in large-scale deployment due to technological readiness level (TRL) bottlenecks, where transitions from laboratory validation to practical implementation remain problematic; the absence of open benchmark datasets and reproducibility mechanisms, restricting horizontal comparison and validation of different algorithms; and systemic deficiencies in ethical, privacy, and compliance evaluations, including shortcomings in data governance, auditability, and accountability.

7. Limitations and Future Perspectives

This study employed bibliometric methods, drawing on the core dataset of the Web of Science database, to systematically review research hotspots and developmental trajectories of artificial intelligence in building maintenance management. However, certain limitations remain. First, the analysis was based on a single mainstream database, without incorporating additional sources such as Google Scholar, ProQuest, SpringerLink, and CNKI. This may have resulted in the omission of important studies, thereby affecting the comprehensiveness and representativeness of the findings. Future research should expand the range of databases utilized to better capture global research trends in the field of intelligent building maintenance and to continuously track the latest advancements.