Next Article in Journal
Proactive Scheduling and Routing of MRP-Based Production with Constrained Resources
Previous Article in Journal
Mapping Drone Applications in Rural and Regional Cities: A Scoping Review of the Australian State of Practice
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Applied Sciences
Volume 15
Issue 15
10.3390/app15158525
applsci-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Personalization of AI-Based Digital Twins to Optimize Adaptation in Industrial Design and Manufacturing—Review

by
Izabela Rojek
1,
Dariusz Mikołajewski
1,
Ewa Dostatni
2,*,
Jan Cybulski
1 and
Mirosław Kozielski
1
1
Faculty of Computer Science, Kazimierz Wielki University, 30 Chodkiewicza St., 85-064 Bydgoszcz, Poland
2
Faculty of Mechanical Engineering, Poznań University of Technology, Marii Skłodowskiej-Curie 5, 60-965 Poznan, Poland
*
Author to whom correspondence should be addressed.
Appl. Sci. 2025, 15(15), 8525; https://doi.org/10.3390/app15158525
Submission received: 24 June 2025 / Revised: 25 July 2025 / Accepted: 30 July 2025 / Published: 31 July 2025
(This article belongs to the Section Computing and Artificial Intelligence)
Browse Figures
Versions Notes

Abstract

The growing scale of big data and artificial intelligence (AI)-based models has heightened the urgency of developing real-time digital twins (DTs), particularly those capable of simulating personalized behavior in dynamic environments. In this study, we examine the personalization of AI-based digital twins (DTs), with a focus on overcoming computational latencies that hinder real-time responses—especially in complex, large-scale systems and networks. We use bibliometric analysis to map current trends, prevailing themes, and technical challenges in this field. The key findings highlight the growing emphasis on scalable model architectures, multimodal data integration, and the use of high-performance computing platforms. While existing research has focused on model decomposition, structural optimization, and algorithmic integration, there remains a need for fast DT platforms that support diverse user requirements. This review synthesizes these insights to outline new directions for accelerating adaptation and enhancing personalization. By providing a structured overview of the current research landscape, this study contributes to a better understanding of how AI and edge computing can drive the development of the next generation of real-time personalized DTs.

1. Introduction

Artificial intelligence (AI)-powered digital twins (DTs) for Industry 4.0/5.0 integrate physical systems with virtual models to improve decision making and process optimization in smart manufacturing. DTs are created by collecting real-time data from sensors embedded in physical assets, which is then used to create a dynamic, virtual replica of the system [1]. AI algorithms process and analyze this data to predict behavior, identify potential failures, and suggest optimizations, improving operational efficiency [2]. Advanced machine learning models enable DTs to evolve and adapt by learning from new inputs over time, increasing their predictive accuracy [3]. Edge–cloud and network computing architectures are used to ensure rapid data exchange and processing, minimizing latency in real-time decision making. The integration of Internet of Things (IoT) devices enhances the ability of DTs to continuously monitor and adapt industrial operations [4]. Industry 5.0 integrates human-centric considerations by enabling collaboration between human operators and AI systems in a DT environment, ensuring that technology complements rather than replaces human expertise [5]. AI-based DTs help achieve more efficient, adaptive, and resilient industrial processes by combining human intelligence with advanced automation, leading to better resource allocation [6]. The ability to simulate different production scenarios allows manufacturers to test and optimize processes before making changes to real-world operations, reducing risk and costs [7]. DTs also support predictive maintenance by identifying potential equipment failures in advance, preventing costly downtime [8,9,10]. DTs support predictive maintenance by continuously monitoring equipment performance and conditions in real time [8]. By analyzing sensor data and historical trends, they can identify signs of wear, defects, or anomalies before a failure occurs. Early detection allows maintenance teams to proactively resolve issues rather than react to failures [9]. As a result, organizations can prevent unexpected equipment failures and significantly reduce unplanned downtime. Ultimately, this leads to increased operational efficiency and cost savings [10]. Cyber-physical security measures must be incorporated into DTs to ensure data integrity and prevent cyber threats that could disrupt industrial operations. The scalability of AI-based DTs enables its application in various industrial domains, from automotive to pharmaceutical manufacturing [11]. The continuous improvement of DTs through reinforcement learning can further enhance process optimization by automatically adjusting production parameters [12]. Collaboration between multiple DTs in a smart factory ecosystem enables synchronized decision making and increased overall efficiency [13,14]. Future advances in quantum computing and generative AI could further accelerate DT capabilities, enabling more complex simulations and deeper insights into industrial processes. Future advances in quantum computing could significantly increase the computational power available for DT simulations, enabling the real-time modeling of highly complex, multidimensional systems. Quantum algorithms can solve optimization and simulation problems currently too resource-intensive for classical computers, increasing the accuracy and speed of DT analysis. At the same time, generative AI can enhance DTs by autonomously generating realistic scenarios, predicting rare events, and filling gaps in incomplete datasets. Together, these technologies could enable DTs to gain deeper insights into system behavior, uncover hidden patterns, and support more precise decision making. As they develop, quantum computers and generative AI will likely expand the scope and scalability of DT applications in industries such as manufacturing, energy, and healthcare.
The aim of this study is to analyze the possibilities and occupations associated with the current and future personalization and optimization of AI-based DTs.
To strengthen the connection between context and the specific objectives of this study on AI-based digital twin personalization in engineering, it was necessary to clearly define the technological and practical gaps in current digital twin implementations. This context highlights the increasing complexity of engineering systems and the growing need for personalized predictive models that adapt to user-specific data. The objectives directly address these contextual challenges by aiming to develop methods for integrating data (behavioral, operational, and environmental) with AI-based digital twin frameworks. This connection ensures that research is focused on improving the efficiency, usability, and accuracy of decisions in specific applications. It is worth striving to bridge theory and practice by proposing scalable personalization strategies based on observed real-world needs.

2. Materials and Methods

2.1. Dataset

Our bibliometric analysis aimed to investigate the research landscape and the state of knowledge and practice in the area of planning and implementing personalized adaptive DTs. For this purpose, we used bibliometric methods to analyze scientific publications and answer the formulated research questions in order to identify key areas such as the current state, origin and evolution of research topics, origin of publications (institutions, country, and, if possible, also funding mode), and the most influential authors and articles.
Three important research questions serve as a framework for reviewing the literature on the personalization of AI-based digital twins in engineering:
  • Research question 1 (RQ1): How do current AI techniques enable the personalization of behavior and decision making in digital twin systems across various engineering domains?
  • RQ2: What are the technical and ethical challenges associated with the personalization of AI-based digital twins, particularly in terms of data privacy, model transparency, and user trust?
  • RQ3: To what extent do existing frameworks and architectures support the scalability, interoperability, and real-time performance of AI-based personalized digital twins at the edge and IoT?
Such a selected approach allowed a comprehensive understanding of current research and industry trends in the field of DTs to be obtained. By interpreting bibliometric data, this study has the potential to enrich current analyses and discussions and establish a solid foundation for future research.

2.2. Methods

In this study, we searched three major bibliographic databases: Web of Science (WoS), Scopus, and dblp. They were selected for their wide coverage of studies and rich citation data that supports in-depth bibliometric analysis of DTs (Table 1). We applied filters to the searches to focus on the relevant literature, narrowing the scope to original articles in English. After filtering, a manual review of each article was conducted to ensure that it met the inclusion criteria, which helped determine the final sample size. Three reviewers were involved in the process, and inclusion/exclusion decisions were made independently. Any discrepancies were resolved by consensus among at least two of the three reviewers. Key features of the dataset were then analyzed, including known authors, research groups/institutions, countries, topic clusters, and emerging trends. This allowed for mapping the evolution of key terminology and major research developments in the field. Where possible, temporal trends were tracked to monitor changes in research coverage over time, and publications were grouped into topic clusters, revealing relationships between different research areas. This process highlighted key topics and subfields within the research area.
The study was based on selected elements of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines for bibliographic reviews, focusing on aspects such as rationale (item 3), objectives (item 4), eligibility criteria (item 5), information sources (item 6), search strategy (item 7), selection process (item 8), data collection process (item 9), synthesis methods (item 13a), synthesis results (item 20b), and discussion (item 23a). For the bibliometric analysis, tools embedded in the Web of Science (WoS), Scopus, and dblp databases were used. This selected review methodology supports bibliometric and scientometric studies, often allowing for refined categorization by concepts, research areas, authors, documents, and sources.
The rationale for using a bibliometric approach in this study is its ability to complement the PRISMA framework by providing a quantitative perspective on the structure and evolution of a research field. While PRISMA was primarily designed for systematic reviews and qualitative content analysis, bibliometric analysis enables the identification of publishing trends, influential authors, collaboration networks, and topic clusters. This dual-method approach increases the depth and breadth of the review by combining structured content assessment with insights based on bibliographic metadata. It also helps uncover hidden patterns and gaps in the literature that may not be apparent using qualitative methods alone. Therefore, integrating bibliometrics with PRISMA supports a more comprehensive and transparent synthesis of current knowledge on AI-based DT personalization in engineering.

2.3. Data Selection

To refine the search, advanced filtered queries were used, limiting the results to articles in English. In WoS, searches were performed using the “Subject” field (consisting of title, abstract, keyword plus, and other keywords); in Scopus, they were performed using article title, abstract, and keywords; and in dblp, they were performed using manual sets of keywords. The databases were searched for articles using keywords such as “artificial intelligence,” “digital twin,” “personalization,” and “adaptation” (Table 2).
The selected set of publications was then further checked by manually re-screening the articles, and removing irrelevant items and duplicates, which allowed for determining the final sample size (Figure 1). In the article selection process, we employed a combination of methodological quality assessment tools to ensure their reliability and validity. Tools such as Abstrackr and SWIFT-Active Screener were used to semi-automatically screen abstracts and prioritize studies based on relevance and quality. Citationchaser supported forward and backward citation tracking and structured evidence synthesis, providing a comprehensive and systematic literature review. Additionally, we considered journal impact factors (IFs) and CiteScore quartiles, prioritizing sources from journals in the first and second quartiles to enhance the credibility and scientific value of the reference list.
The general summary of the results of the bibliographic analysis is presented in Table 3 and Figure 1, Figure 2, Figure 3 and Figure 4. The review included 85 articles (2018–2025)—no older ones were included. Articles published before 2018 were excluded because integrating personalization with AI-based DTs in engineering is a relatively new phenomenon. Significant advances in AI/ML and real-time data integration, which enable personalization, have only been widely applied in DTs in the last few years. Including older studies would risk including outdated methods or technologies that do not reflect current capabilities or trends. The 2018–2025 timeframe ensures that the review focuses on the most relevant, innovative, and technically feasible approaches. This limited timeframe also aligns with the rapid development of AI-based engineering solutions, providing a clearer picture of current and emerging best practices.
The final set of 85 publications constitutes a representative sample of the literature, reflecting both the breadth and depth of current research on AI-based DTs in engineering. This number was achieved through a rigorous selection process based on relevance, topical saturation, and scientific impact, ensuring the inclusion of key studies and influential works. The dataset encompasses a wide range of perspectives, methodologies, and applications, indicating the coverage of major research trends and directions. Saturation was observed because recurring concepts and findings appeared consistently across the included studies, suggesting a comprehensive representation of the field (Figure 5, Figure 6, Figure 7, Figure 8 and Figure 9). Furthermore, the inclusion of highly cited and peer-reviewed publications strengthens the credibility and scientific relevance of the review.
This review highlights several potential limitations that may impact the scope and comprehensiveness of the findings. First, the inclusion of only English-language sources introduces linguistic bias, potentially excluding relevant studies published in other languages. Second, the literature search was limited to three major academic databases, which, although comprehensive, may have missed relevant work indexed elsewhere. Third, the exclusion of gray literature—such as technical reports, theses, and industry papers—may limit insight into emerging practices and real-world applications not yet covered by peer-reviewed sources. These limitations underscore the need for cautious interpretation and suggest directions for broader future research.

3. Architecture and Devices of Simple DTs vs. Complex AI-Based DTs

In simple DTs, the architecture is lightweight, typically consisting of a basic sensor-to-cloud or sensor-to-edge configuration. It uses IoT devices such as temperature, pressure, or motion sensors to collect data. Data is transmitted via wired (Ethernet) or wireless (Wi-Fi, Bluetooth, Long Range Wide Area Network (LoRaWAN)) communication. A cloud platform (AWS Internet of Things (IoT), Amazon, Seattle, WA, USA, Azure IoT, Microsoft, Redmond, WA, USA, or an on-premises server) stores and processes the data in real time. Dashboards and analytics tools (Grafana 12.0.2, Grafana Labs, Penthouse, NY, USA, Power BI, Microsoft, Redmond, WA, USA) provide simple visualizations. The system uses a rule-based model, meaning that responses are predefined and deterministic. It often uses microcontrollers (ESP32, TSMC, Hsinchu, Taiwan, Arduino, Arduino S.R.L., Monza, Italy) or single-board computers (Raspberry Pi, Raspberry Pi Foundation, Cambridge, UK) for edge computing. Updates are manual or scheduled with little real-time adaptation [15,16,17].
In complex AI-based DTs, the architecture is multi-layered, including IoT, edge computing, AI, and deep learning models in the cloud. It uses advanced edge devices (NVIDIA Jetson, NVIDIA, Santa Clara, CA, USA, Intel Movidius, Intel, Santa Clara, CA, USA) with AI inference capabilities; includes high-precision sensors (light detection and ranging (LiDAR), 3D cameras, industrial-grade IoT sensors) to collect rich data; and supports real-time data processing using AI and ML models, enabling predictive analytics and autonomous decision making. AI algorithms are deployed on AI edge devices or AI cloud services (Google AI models, Google, Mountain View, CA, USA, Azure ML, Microdoft, Redmond, WA, USA, TensorFlow, Google, Muntain View, CA, USA). It integrates digital simulation tools (Ansys Twin Builder, Ansys Inc., Canonsburg, PA, USA, Siemens MindSphere, Siemens, Munich, Germany) for continuous optimization. Self-learning engines enable adaptive updates, making the twin dynamic and evolving over time [18,19,20].

3.1. Ways of Personalizing DTs to Optimize Adaptation to Signal Selection, Gathering, and Preprocessing Level

The personalization of DTs involves dynamically selecting the most appropriate sensors based on changing environmental conditions or system needs, optimizing data accuracy and relevance. AI-based algorithms can assess operating conditions and prioritize critical signals, reducing unnecessary data transmission and focusing on key parameters. Machine learning models can be trained to filter out noise and redundant data, ensuring only relevant information is processed and stored. Implementing AI-enabled preprocessing at the edge (on IoT devices or AI processors at the edge) reduces latency and optimizes real-time decision making before sending data to the cloud. Personalization allows the DT to dynamically adjust the sensor sampling rate based on system requirements, increasing efficiency and minimizing bandwidth usage. Instead of continuously collecting data, DTs can be programmed to collect and process signals only when predefined conditions or anomalies occur. Domain experts can define weighting values for different signals, ensuring that the most important parameters drive decision making and predictions. Machine learning models can be trained to extract the most relevant features from raw sensor data, improving signal quality and optimizing compute resources. Configurable anomaly detection settings allow the twin to adjust sensitivity levels based on specific industry needs, preventing false alarms while ensuring system reliability. The system can learn from user feedback and historical data patterns to improve signal selection, preprocessing methods, and overall performance of the DT over time [21,22,23].

3.2. Ways of Personalizing DTs to Optimize Adaptation to IoT Level

DTs can personalize IoT device settings based on real-time conditions, adjusting parameters such as power consumption, data transmission rates, and operational modes. Deploying AI models at the IoT level enables real-time adaptation, allowing edge devices to process data locally and send only relevant insights to the cloud. Personalizing communications by selecting the most efficient protocol (Message Queuing Telemetry Transport (MQTT), Constrained Application Protocol (CoAP), Open Platform Communications Unified Architecture (OPC UA)) based on network conditions and device capabilities ensures smooth data flow. DTs can optimize data transmission by dynamically adjusting IoT bandwidth utilization, reducing congestion, and prioritizing critical data. IoT devices in the system can be synchronized based on specific operational needs, providing coordinated responses and reducing redundant data collection. AI-based updates enable IoT devices to receive targeted firmware updates based on usage patterns, improving performance while minimizing disruptions. Configurable roles enable specific devices to focus on designated tasks, such as anomaly detection, energy monitoring, or predictive maintenance, optimizing overall system performance. IoT devices can be programmed to dynamically switch between low-power and high-performance modes based on real-time workload and environmental conditions. Customizing authentication methods, encryption levels, and access controls for IoT devices balances security with performance in DT ecosystems. IoT devices can learn from historical data, user feedback, and AI recommendations to improve their operations, increasing long-term adaptability and efficiency [24,25,26].

3.3. Ways of Personalizing DTs to Optimize Adaptation to Edge Computing Level

Thepersonalization of DTs at the edge enables adaptive workload allocation, balancing real-time processing across multiple edge devices based on compute power. Machine learning models can be tuned for specific tasks, enabling edge devices to process critical data locally and reducing reliance on cloud computing. Users can define how data is filtered, aggregated, and transformed at the edge, optimizing processing efficiency and reducing unnecessary data transmission. Edge computing resources such as central processing unit (CPU), graphics processing unit (GPU), and random access memory (RAM) can be dynamically managed based on system requirements, providing optimal performance and energy efficiency. DTs can personalize edge computing to prioritize time-sensitive tasks, ensuring critical decisions are processed with minimal latency. Implementing personalized compression techniques based on data type and network conditions reduces bandwidth usage while preserving critical information. Personalization enables defining security measures such as encryption, access control, and anomaly detection tailored to specific edge environments and threat levels. Instead of continuous computation, edge devices can process data only when certain conditions are met, improving efficiency and reducing energy consumption. AI-based feedback loops enable edge computing frameworks to learn from past performance and dynamically optimize their processing strategy over time. Personalizing when and how data is moved to the cloud versus at the edge provides an optimal balance between latency, cost, and compute [27,28,29,30].

3.4. Ways of Personalizing DTs to Optimize Adaptation to AI Level

DTs can be personalized by selecting AI models tailored to specific tasks, such as predictive maintenance, anomaly detection, or real-time optimization. AI models in DTs can dynamically modify their learning rates based on performance feedback, providing faster convergence and higher accuracy. Personalization enables DTs to extract and prioritize the most relevant features from data, improving AI-driven insights and decision making. AI models can also be personalized to switch between different algorithms or architectures based on the operating environment or data patterns. Configurable AI settings allow users to specify which aspects (e.g., speed, accuracy, energy efficiency) should be prioritized for decision making and reasoning. AI-driven DTs can automatically tune hyperparameters such as batch size, activation functions, and regularization techniques to improve performance. AI models can customize data augmentation techniques based on specific use cases, providing robust training and improved generalization. AI algorithms can continually evolve based on real-world user feedback, improving their predictions and decisions over time. DTs can optimize where AI computations are performed, dynamically deciding whether tasks should be run on edge devices for low latency or in the cloud for higher compute power. AI models can be configured to incorporate human expertise, adjusting trust thresholds and enabling manual overrides for critical decisions when needed [31,32,33].

3.5. Ways of Personalizing DTs to Optimize Adaptation to Deep Learning Models in the Cloud Level

DTs can be personally connected to learning architectures (e.g., convolutional neural networks (CNNs) for image data, long short-term memory’s (LSTMs) for time series, or transformers for complex patterns) best distributed to applications. Cloud-based ground truth models can be optimized to scale up or down based on workload requirements and cost-effective resource utilization. DTs can differentiate AutoML or Bayesian optimization to personalize hyperparameters such as outcome markers, specified criteria, and layer configurations to obtain model results. Universal ground truth models and cloud-based DTs can personalize learning by first fine-tuning trained models on domain-specific datasets. Workflows can be applied to actually act on, augment, and filter the data that results from the model while reducing unnecessary computation. DTs can personalize how and when models are trained, responding with feedback over time to determine whether batch, online, or resultant learning is more effective. Depending on the application needs, DTs can configure cloud-based outbound models to prioritize inference, in addition to, or energy. The DT configuration allows the twin to decide which parts of the outbound model are off to be run at the edge output (for real-time tasks) and which are off in the cloud (for feature application). DTs can be customized to adapt explained techniques such as SHapley Additive exPlanation (SHAP) or Local Interpretable Model-agnostic Explanations (LIME) to comply with regulations regarding the level of expertise of different users. Models that come from analytics can be applied to gain improved insights, providing feedback to the domain, such as from the results of prediction and adaptation (Table 4 and Table 5) [34,35,36,37].

4. Resilience and Sustainability of the AI-Based DTs

The resilience of AI-based DTs is a critical challenge due to their reliance on data quality, real-time processing, and adaptation to changing environments. Hybrid DTs, which combine physics-based models with AI-based simulations, face challenges in maintaining consistency and robustness, as discrepancies between model types can lead to inaccurate predictions [38,39]. Multi-level DTs, which integrate multiple hierarchical model layers (e.g., component, system, and enterprise level), struggle with synchronization and consistency across levels, which impacts their resilience [40,41]. Reused DTs, which are adapted from existing models for new applications, can inherit outdated assumptions and biases, reducing their effectiveness in dynamic environments [42,43,44]. AI-based DTs are dependent on massive amounts of sensor data, making them vulnerable to cyberattacks, sensor failures, and hostile AI threats that weaken their resilience. The constant evolution of AI models requires constant updates, which can introduce new errors or lead to model drift, which complicates the long-term stability of DTs [45,46]. Hybrid DTs can have explainability issues because AI-generated insights can conflict with deterministic physics-based models, reducing trust in decision-making processes. Multi-level DTs require scalable data integration, and failures at one level can cascade through the system, leading to instability in the connected processes. Reused DTs often struggle to adapt to new operational contexts due to differences in physical conditions, data sources, or regulatory constraints [47,48,49]. Ensuring the stability of AI-based DTs requires balancing computational efficiency with the need for high-fidelity simulations, which can be resource-intensive. AI-based predictions in DTs can deteriorate over time due to changing real-world conditions, requiring robust retraining mechanisms to ensure long-term resilience. Hybrid DTs can suffer from model divergence if AI components evolve independently of physics-based rules, leading to inconsistencies. Multi-level DTs must address the challenges of interoperability between different AI models and simulation frameworks to remain resilient across domains. Reused DTs need adaptive learning and self-correction mechanisms to ensure they remain relevant and accurate in new contexts. The resilience of AI-based DTs depends on their ability to minimize resource consumption, ensure the ethical use of AI, and maintain long-term operational effectiveness [50,51,52,53].
Studies have shown how the level of decomposition, model size, and submodel structure affect the accuracy, efficiency, and adaptivity of DTs [54,55]. Research has shown that the level of decomposition in DT models—the way a complex system is broken down into smaller components—directly affects the granularity and accuracy of simulations. Finer decomposition can lead to more accurate modeling of individual parts, but it can also increase computational load and complexity. Similarly, the overall model size affects performance: larger models can represent systems more comprehensively but require more processing power and memory, which can limit real-time responsiveness. The structure and interconnectivity of submodels also influence adaptability, determining how easily the DT can be updated or reconfigured when system parameters change. Therefore, finding the right balance between decomposition, model size, and submodel architecture is crucial to optimizing the performance and scalability of DT applications. Studies indicate that excessive decomposition can lead to computational inefficiencies, while insufficient decomposition can result in oversimplified models that lack precision. The size of models and submodels directly affects simulation speed and real-time responsiveness, requiring optimization techniques to balance detail and efficiency [56,57]. The structural organization of submodels affects interoperability, and poorly structured submodels can create bottlenecks in digital twin performance [58,59]. Despite progress, there is still a need for fast-response models that can dynamically adapt to changing conditions while maintaining computational efficiency [60,61,62,63,64]. Adaptive personalized DTs require modular architectures that facilitate the seamless integration of AI-based algorithms and domain-specific models [65,66]. Recent research has focused on developing hybrid modeling approaches that combine data-driven AI techniques with domain knowledge to enhance adaptivity. Improving performance depends on efficient data processing pipelines that support real-time decision making while minimizing computational overhead [67,68,69]. Integrating multiple algorithms into DTs requires a robust framework that ensures consistency, synchronization, and error management; hence, future research should emphasize scalable and interoperable designs that enable the rapid deployment of personalized DTs across domains [70].
AI-based DTs with partial domain adaptation are advanced virtual replicas of physical systems that leverage AI to improve real-world performance while selectively adapting to new environments [71,72]. AI-based DTs with partial domain adaptation can transfer knowledge from one environment to another, reducing the need for extensive retraining and accelerating implementation in new environments. This adaptability allows DTs to remain effective even when operational conditions or data distribution change, for example, when transferring from one urban infrastructure to another or between different hospital systems. By improving modeling efficiency and response time, DTs support real-time optimization and resource management across various sectors. For example, in autonomous systems, they help vehicles adapt to new conditions with minimal manual intervention, increasing operational reliability. Furthermore, by enabling smarter resource utilization and predictive analytics, they contribute to more sustainable practices in energy consumption, waste reduction, and long-term system planning. These models integrate real-time sensor data and machine learning to dynamically simulate, predict, and optimize operations. Partial domain adaptation ensures that the DT learns only the relevant changes from the new domain without retraining the entire model, reducing computational costs [73,74]. This technique is especially beneficial when transferring knowledge between similar but not identical environments, such as different manufacturing plants or smart grids. Optimizing latency and power consumption is critical for real-time applications that require efficient resource allocation and AI-driven predictions [75,76,77]. Edge computing and federated learning can be integrated with DTs to reduce communication latency and improve energy efficiency. AI algorithms such as reinforcement learning help balance performance tradeoffs, ensuring minimal processing latency while conserving energy. By dynamically adjusting parameters, AI-based DTs can optimize task scheduling, workload distribution, and resource allocation. These models also support predictive maintenance, reducing downtime and unnecessary energy consumption [78,79,80,81,82]. AI-based DTs with partial domain adaptation increase efficiency, adaptability, and sustainability in various industries, from smart cities to healthcare and autonomous systems.

5. Discussion

Compared to other studies on AI-based DTs in engineering, our review highlights a stronger emphasis on personalization, particularly in tailoring simulations and predictions to specific users, environments, or operational profiles. Several studies, for example, on aerospace and manufacturing systems, have used experimental data—such as sensor logs and operational feedback—to train adaptive DTs that could adjust maintenance schedules or performance parameters in real time. These findings showed that personalized DTs in smart manufacturing reduced downtime by 18% using adaptive learning models. Other studies, particularly in energy systems, have shown that AI-assisted DTs using user-specific consumption data can optimize energy efficiency and predictive control, improving accuracy by up to 25% compared to static models. A recurring theme in the scientific literature is the use of ML algorithms—especially reinforcement learning and deep neural networks—to dynamically adapt DT operation to individual conditions. In comparison, our review indicates a growing shift toward hybrid models, combining physics- and data-driven approaches that offer both interpretability and adaptability. It is worth noting that personalization was less emphasized in earlier studies, which focused more on general system modeling and simulation fidelity, suggesting a recent methodological evolution. Furthermore, research in biomedical engineering has highlighted the use of personalized DTs in diagnostics and patient-specific treatment simulations, confirming the versatility of this concept across various domains. Although technical implementations vary—from cloud infrastructures to embedded edge computing—the trend toward the user-centric personalization of DTs is gaining traction. Our synthesis shows that personalized DTs consistently lead to improved outcomes in terms of accuracy, efficiency, and system responsiveness, reinforcing the value of integrating adaptive AI frameworks with DT engineering (Table 6).
The answer to RQ1 highlights state-of-the-art AI methods (deep learning, generative AI) used to customize digital twins for individual users, resources, or environments. The answer to RQ2 highlights technological gaps and societal concerns that impact the adoption and responsible implementation of personalized digital twins. In response to RQ3, infrastructure readiness and limitations in current DT deployments were assessed, particularly in distributed or resource-constrained environments.

5.1. Advantages of the Proposed Approach

The data collection methods described in Section 2 provide the foundation for the comparative analysis presented in Section 3 and Section 4, but the relationship between them requires further explanation. Section 2 presents the tools, sampling strategies, and criteria used to gather information on digital twin architectures and performance metrics, and Section 3 and Section 4 explain how these specific datasets impacted the categorization of “simple” and “complex AI-based” digital twins. Section 3 discusses architectural differences and device integrations, and Section 4 discusses resilience and sustainability metrics.
The proposed novel approaches to AI-based DTs for Industry 4.0/5.0 offer numerous advantages across various industrial applications. First, they enhance real-time decision making by integrating AI-driven predictive analytics, enabling proactive responses to operational challenges. The use of advanced machine learning models allows DTs to continuously improve by learning from new data, increasing accuracy in fault detection and process optimization. Cloud–edge hybrid computing architectures ensure low-latency processing, making DTs more responsive in time-sensitive industrial environments. The integration of IoT devices enables seamless data collection and synchronization between physical and virtual systems, improving operational efficiency. AI-driven DTs support predictive maintenance by identifying equipment failures before they occur, reducing downtime and maintenance costs. In smart manufacturing, DTs optimize production scheduling by simulating various scenarios, leading to better resource allocation and increased productivity. Industry 5.0-oriented DTs emphasize human–AI collaboration, enhancing worker safety and productivity by providing AI-assisted decision support. The implementation of cyber-physical security measures ensures data integrity and protection against cyber threats, making DTs more reliable. AI-based DTs facilitate mass customization by dynamically adjusting production parameters to meet individualized consumer demands. The scalability of these approaches allows for integration across multiple industries, including automotive, aerospace, pharmaceuticals, and energy sectors. Multi-agent DT systems enable interconnected decision making, improving supply chain management and logistics. The ability to simulate and optimize energy consumption in industrial processes contributes to sustainability and cost savings. AI-enhanced DTs streamline compliance with industry regulations by automatically monitoring and adjusting processes to meet safety and environmental standards. The combination of reinforcement learning and generative AI enables the creation of highly adaptive DTs capable of responding to unforeseen disruptions. Finally, the integration of quantum computing in AI-based DTs has the potential to accelerate complex simulations, unlocking new possibilities for industrial process innovation [82,83,84,85,86].
These findings bridge the gap between theoretical innovation and practical implementation, accelerating the responsible evolution of digital twins in real-world environments. These findings have significant practical implications, offering practical implications for implementing AI-based digital twins across various industries. By improving predictive accuracy and making real-time decisions, these systems can optimize operations in manufacturing, energy, healthcare, and logistics. Integrating AI enables digital twins to dynamically adapt, reducing downtime, improving maintenance strategies, and minimizing operating costs. Policymakers can use these findings to establish standards and guidelines that ensure the safe, transparent, and ethical use of digital twins. In industrial applications, the findings support the development of more resilient supply chains and streamlined production lines through improved simulation and scenario testing. For researchers, the results highlight key areas where AI models require further refinement to better represent complex physical systems. The work also highlights the importance of data quality and interoperability, which are key to scalable and portable digital twin applications. Future AI-based DTs could incorporate the study’s recommendations to improve personalization and generalization and reduce bias, supporting broader adoption across various sectors. Furthermore, these findings could inform workforce training and retraining initiatives, ensuring operators can effectively interact with AI-based systems.

5.2. Limitations

Despite significant progress, previous and current approaches to AI-based DTs for Industry 4.0/5.0 still face several limitations. First, the high computational requirements of AI-based DTs pose a challenge for real-time processing, especially in resource-constrained environments. Many existing DT implementations struggle with data silos, which makes seamless integration across industrial systems difficult. The reliance on large volumes of real-time data raises concerns about data security and privacy, especially when processing sensitive industrial information. Interoperability remains a major challenge, as different industries and manufacturers use different standards, preventing seamless communication between DTs. The high cost of developing and deploying AI-based DTs limits accessibility, especially for small- and medium-sized enterprises (SMEs). Current AI models are often not explainable, making it difficult for human operators to trust and interpret the recommendations generated by DTs. Many DTs are heavily dependent on cloud computing, which can cause latency and dependency on stable internet connections. The complexity of integrating AI-based DTs into existing industrial workflows often requires the extensive reconfiguration and retraining of personnel. Limited generalizability means that DTs optimized for one industrial process may not be easily adapted to others without significant retraining. While predictive maintenance is a key application, false alarms and inaccurate failure predictions can still occur, leading to unnecessary interventions or missed issues. The scalability of current DT frameworks remains a challenge, especially in large industrial ecosystems with many interactive DTs. AI models used in DTs require continuous updates and retraining, which can be time-consuming and computationally expensive. The lack of standardized validation methods for AI-based DTs makes it difficult to assess their reliability and performance across industries. In Industry 5.0 applications, the human–AI collaboration aspect is still underdeveloped, as most DTs focus primarily on automation rather than improving human decision making. Finally, current approaches often struggle to effectively integrate sustainability indicators, which limits their ability to optimize energy consumption and reduce the environmental impact of industrial processes [87,88,89,90,91].

5.3. Directions for Further Studies

Future research on AI-based digital DTs for Industry 4.0/5.0 should focus on increasing computational efficiency using advanced AI techniques such as federated learning and neuromorphic computing. Developing standardized interoperability frameworks and protocols will enable seamless integration across industrial systems and vendors [92,93,94]. Further research is needed on privacy-preserving AI methods such as differential privacy and secure multi-party computation to protect sensitive industrial data. Exploring hybrid edge-of-the-cloud computing solutions can optimize real-time processing while reducing the dependence on centralized cloud infrastructure. Developing self-learning DTs using continuous and reinforcement learning will enable adaptive optimization without the need for frequent retraining [95,96,97,98]. Extending DT capabilities to support multi-agent collaboration will improve supply chain coordination and distributed decision making. Improved explainability and transparency in AI models will increase trust and usability for human operators, especially in Industry 5.0 applications. Research into integrating AI-driven DTs with digital ethics frameworks can help address issues related to bias, fairness, and accountability. Exploring the use of generative AI to generate synthetic data can improve the training and validation of DTs in cases where real-world data is limited. Developing robust validation and benchmarking methods will ensure the reliability and accuracy of AI-driven DTs across industries [99,100]. Research into the application of quantum computing to DT simulations can accelerate complex optimizations and enhance problem-solving capabilities. Incorporating sustainability-focused AI models will allow DTs to optimize energy efficiency, waste reduction, and environmental impact. Improving human–AI collaboration in DTs through natural language processing and augmented-reality interfaces will improve decision making and human performance. Future research should explore blockchain integration to provide secure, transparent, and tamper-proof data management in DT ecosystems. Finally, improving the scalability of AI-based DTs through modular architectures and lightweight AI models will support broader adoption across industry sectors [101,102,103,104].

6. Conclusions

AI-based DTs are advanced virtual representations of individuals that integrate real-time data based on AI/ML and simulation to support personalized decision making, particularly in industry and smart environments. Our analysis revealed the growing integration of ML, IoT, and high-fidelity modeling, enabling adaptive, responsive DTs with significant potential for pre-industrial applications. Despite these advances, challenges remain in areas such as data privacy, real-time synchronization, computational efficiency, and ethical issues. Future research is expected to focus on improved multimodal data fusion, improved AI-based predictions, and more intuitive human–twin interactions. Emerging technologies such as quantum computers could further enhance personalization capabilities and the simulation speed. We also identified growing interest in applying this technology to education, productivity, and mental health. This study contributes to this understanding by mapping the current landscape and identifying key directions for future research. Ensuring fairness, security, and user-centric control will be key to the responsible and effective development of personalized DT systems.

Author Contributions

Conceptualization, I.R., D.M., E.D., J.C., and M.K.; methodology, I.R., D.M., E.D., J.C., and M.K.; software, I.R. and D.M.; validation, I.R., D.M., E.D., J.C., and M.K.; formal analysis, I.R., D.M., E.D., J.C., and M.K.; investigation, I.R., D.M., E.D., J.C., and M.K.; resources, I.R. and D.M.; data curation, I.R. and D.M.; writing—original draft preparation, I.R., D.M., E.D., J.C., and M.K.; writing—review and editing, I.R., D.M., E.D., J.C., and M.K.; visualization, D.M.; supervision, E.D. and I.R.; project administration, I.R.; funding acquisition, E.D. and I.R. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported in part by a grant to maintain the research potential of Kazimierz Wielki University (grant of Ministry of Science and Higher Education, 2025) and grant no. 0613/SBAD/4940 for Poznan University of Technology.

Data Availability Statement

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
AIArtificial intelligence
CNNConvolutional Neural Network
CoAPConstrained Application Protocol
CPUCentral processing unit
DLDeep learning
DTDigital twin
GDPRGeneral Data Protection Regulation
GPUGraphics processing unit
IoTInternet of Things
LiDARLight detection and ranging
LIMELocal Interpretable Model-agnostic Explanation
LoRaWANLong Range Wide Area Network
LSTMLong short-term memory
MLMachine learning
MQTTMessage Queuing Telemetry Transport
OPC UAOpen Platform Communications Unified Architecture
PRISMAPreferred Reporting Items for Systematic Reviews and Meta-Analyses
RAMRandom access memory
SHAPSHapley Additive exPlanation
SMESmall- and medium -sized enterprise
WoSWeb of Science

References

  1. Bellavista, P.; Bicocchi, N.; Fogli, M.; Giannelli, C.; Mamei, M.; Picone, M. Requirements and design patterns for adaptive, autonomous, and context-aware digital twins in industry 4.0 digital factories. Comput. Ind. 2023, 149, 103918. [Google Scholar] [CrossRef]
  2. Awouda, A.; Traini, E.; Bruno, G.; Chiabert, P. IoT-Based Framework for Digital Twins in the Industry 5.0 Era. Sensors 2024, 24, 594. [Google Scholar] [CrossRef]
  3. Ogunsakin, R.; Mehandjiev, N.; Marín, C.A. Towards adaptive digital twins architecture. Comput. Ind. 2023, 149, 103920. [Google Scholar] [CrossRef]
  4. Botín-Sanabria, D.M.; Mihaita, A.-S.; Peimbert-García, R.E.; Ramírez-Moreno, M.A.; Ramírez-Mendoza, R.A.; Lozoya-Santos, J.D.J. Digital twin technology challenges and applications: A comprehensive review. Remote Sens. 2022, 14, 1335. [Google Scholar] [CrossRef]
  5. Fuller, A.; Fan, Z.; Day, C.; Barlow, C. Digital Twin: Enabling Technologies, Challenges and Open Research. IEEE Access 2020, 8, 108952–108971. [Google Scholar] [CrossRef]
  6. Lv, Z. Digital Twins in Industry 5.0. Research 2023, 6, 0071. [Google Scholar] [CrossRef]
  7. Rasheed, A.; San, O.; Kvamsdal, T. Digital twin: Values, challenges and enablers. arXiv 2019, arXiv:1910.01719. [Google Scholar] [CrossRef]
  8. Rojek, I.; Macko, M.; Mikołajewski, D.; Saga, M.; Burczynski, T. Modern methods in the field of machine modeling and simulation as a research and practical issue related to Industry 4.0. Bull. Pol. Acad. Sci. Tech. Sci. 2021, 69, e136719. [Google Scholar] [CrossRef]
  9. Al-Ali, A.-R.; Gupta, R.; Batool, T.Z.; Landolsi, T.; Aloul, F.; Al Nabulsi, A. Digital twin conceptual model within the context of internet of things. Future Internet 2020, 12, 163. [Google Scholar] [CrossRef]
  10. Ye, W.; Zhao, X.; Wang, Y.; Chen, H.; Lin, L. Digital Twin-Assisted Online Monitoring of Elevator Abnormal Vibration Using Improved ShuffleNetV2 and Domain Adaptation. IEEE Access 2025, 13, 46897–46908. [Google Scholar] [CrossRef]
  11. Zhu, P.; Deng, L.; Tang, B.; Yang, Q.; Li, Q. Digital twin-enabled entropy regularized wavelet attention domain adaptation network for gearboxes fault diagnosis without fault data. Adv. Eng. Inform. 2025, 64, 103055. [Google Scholar] [CrossRef]
  12. Tong, X.; Liu, Q.; Zhou, Y.; Sun, P. A digital twin-driven cutting force adaptive control approach for milling process. J. Intell. Manuf. 2025, 36, 551–568. [Google Scholar] [CrossRef]
  13. Tao, Y.; Lei, B.; Shi, H.; Chen, J.; Zhang, X. Adaptive multi-layer deployment for a digital-twin-empowered satellite-terrestrial integrated network. Front. Inf. Technol. Electron. Eng. 2025, 26, 246–259. [Google Scholar] [CrossRef]
  14. Pratt, S.; Kosmal, T.; Williams, C. Adaptively sampled distance functions: A unifying digital twin representation for advanced manufacturing. Robot. Comput. Integr. Manuf. 2025, 92, 102877. [Google Scholar] [CrossRef]
  15. Azari, M.S.; Santini, S.; Edrisi, F.; Flammini, F. Self-adaptive fault diagnosis for unseen working conditions based on digital twins and domain generalization. Reliab. Eng. Syst. Saf. 2025, 254, 110560. [Google Scholar] [CrossRef]
  16. Jia, S.; Sun, D.; Noman, K.; Wang, X.; Li, Y. Lifting wavelet-informed hierarchical domain adaptation network: An interpretable digital twin-driven gearbox fault diagnosis method. Reliab. Eng. Syst. Saf. 2025, 254, 110660. [Google Scholar] [CrossRef]
  17. Mou, F.; Lou, J.; Tang, Z.; Wu, Y.; Jia, W.; Zhang, Y.; Zhao, W. Adaptive Digital Twin Migration in Vehicular Edge Computing and Networks. IEEE Trans. Veh. Technol. 2025, 74, 4839–4854. [Google Scholar] [CrossRef]
  18. Bin Hakim, S.; Adil, M.; Velasquez, A.; Song, H.H. ANSR-DT: An Adaptive Neuro-Symbolic Learning and Reasoning Framework for Digital Twins. arXiv 2025, arXiv:2501.08561. [Google Scholar]
  19. Rojek, I.; Mikołajewski, D.; Dostatni, E.; Kopowski, J. Specificity of 3D Printing and AI-Based Optimization of Medical Devices Using the Example of a Group of Exoskeletons. Appl. Sci. 2023, 13, 1060. [Google Scholar] [CrossRef]
  20. Ayepah-Mensah, D.; Sun, G.; Pang, Y.; Jiang, W. Adaptive Digital Twin and Communication-Efficient Federated Learning Network Slicing for 5G-enabled Internet of Things. arXiv 2024, arXiv:2407.10987. [Google Scholar]
  21. Walton, R.; Ciarallo, F.W.; Champagne, L.E. A unified digital twin approach incorporating virtual, physical, and prescriptive analytical components to support adaptive real-time decision-making. Comput. Ind. Eng. 2024, 193, 110241. [Google Scholar] [CrossRef]
  22. Almeida Costa, E.; Menezes Rebello, C.; Schnitman, L.; Loureiro, J.M.; Ribeiro, A.M.; Idelfonso, B.R. NogueiraAdaptive digital twin for pressure swing adsorption systems: Integrating a novel feedback tracking system, online learning and uncertainty assessment for enhanced performance. Eng. Appl. Artif. Intell. 2024, 127, 107364. [Google Scholar] [CrossRef]
  23. Wang, D.; Ma, H.; Qiao, J. Multilayer adaptive critic design with digital twin for data-driven optimal tracking control and industrial applications. Eng. Appl. Artif. Intell. 2024, 133, 108228. [Google Scholar] [CrossRef]
  24. Gupta, R.; Tian, B.; Wang, Y.; Nahrstedt, K. TWIN-ADAPT: Continuous Learning for Digital Twin-Enabled Online Anomaly Classification in IoT-Driven Smart Labs. Future Internet 2024, 16, 239. [Google Scholar] [CrossRef]
  25. Li, Z.; Ding, X.; Song, Z.; Wang, L.; Qin, B.; Huang, W. Digital twin-assisted dual transfer: A novel information-model adaptation method for rolling bearing fault diagnosis. Inf. Fusion 2024, 106, 102271. [Google Scholar] [CrossRef]
  26. Dasgupta, S.; Rahman, M.; Harnessing, S.J. Digital Twin Technology for Adaptive Traffic Signal Control: Improving Signalized Intersection Performance and User Satisfaction. IEEE Internet Things J. 2024, 11, 36596–36618. [Google Scholar] [CrossRef]
  27. Hu, S.; Li, M.; Gao, J.; Zhou, C.; Shen, X. Adaptive Device-Edge Collaboration on DNN Inference in AIoT: A Digital-Twin-Assisted Approach. IEEE Internet Things J. 2024, 11, 12893–12908. [Google Scholar] [CrossRef]
  28. Kamal, H.; Yánez, W.; Hassan, S.; Sobhy, D. Digital-Twin-Based Deep Reinforcement Learning Approach for Adaptive Traffic Signal Control. IEEE Internet Things J. 2024, 11, 21946–21953. [Google Scholar] [CrossRef]
  29. Zhang, Y.; Zhang, H.; Lu, Y.; Sun, W.; Wei, L.; Zhang, Y.; Wang, B. Adaptive Digital Twin Placement and Transfer in Wireless Computing Power Network. IEEE Internet Things J. 2024, 11, 10924–10936. [Google Scholar] [CrossRef]
  30. Feng, H.; Cao, K.; Huang, G.; Liu, H. ABWOA: Adaptive boundary whale optimization algorithm for large-scale digital twin network construction. J. Cloud Comput. 2024, 13, 110. [Google Scholar] [CrossRef]
  31. Yang, W.; Yang, Y.; Xiang, W.; Yuan, L.; Yu, K.; Alonso, A.H.; Ureña, J.; Pang, Z. Adaptive optimization federated learning enabled digital twins in industrial IoT. J. Ind. Inf. Integr. 2024, 41, 100645. [Google Scholar] [CrossRef]
  32. Rojek, I.; Studziński, J. Comparison of different types of neuronal nets for failures location within water-supply networks. Maint. Reliab. 2014, 16, 42–47. [Google Scholar]
  33. Qu, W.; Li, J.; Zhang, R.; Liu, S.; Bao, J. Adaptive planning of human—robot collaborative disassembly for end-of-life lithium-ion batteries based on digital twin. J. Intell. Manuf. 2024, 35, 2021–2043. [Google Scholar] [CrossRef]
  34. Ma, Y.; Song, F.; Pau, G.; You, I.; Zhang, H. Adaptive Service Provisioning for Dynamic Resource Allocation in Network Digital Twin. IEEE Netw. 2024, 38, 61–68. [Google Scholar] [CrossRef]
  35. Tao, F.; Cheng, J.; Qi, Q.; Zhang, M.; Zhang, H.; Sui, F. Digital twin-driven product design, manufacturing and service with big data. Int. J. Adv. Manuf. Technol. 2018, 94, 3563–3576. [Google Scholar] [CrossRef]
  36. Cui, l.; Xiao, Y.; Liu, D.; Han, H. Digital twin-driven graph domain adaptation neural network for remaining useful life prediction of rolling bearing. Reliab. Eng. Syst. Saf. 2024, 245, 109991. [Google Scholar] [CrossRef]
  37. Rojek, I.; Jagodziński, M. Hybrid Artificial Intelligence System in Constraint Based Scheduling of Integrated Manufacturing ERP Systems. In Hybrid Artificial Intelligent Systems, Proceedings of the 7th International Conference, HAIS 2012, Salamanca, Spain, 28–30 March 2012; Corchado, E., Snášel, V., Abraham, A., Woźniak, M., Graña, M., Cho, S.B., Eds.; Lecture Notes in Computer Science Series; Springer: Berlin/Heidelberg, Germany, 2012; pp. 229–240. [Google Scholar]
  38. Cui, L.; Wang, X.; Liu, D.; Wang, H. An Adaptive Sparse Graph Learning Method Based on Digital Twin Dictionary for Remaining Useful Life Prediction of Rolling Element Bearings. IEEE Trans. Ind. Inform. 2024, 20, 10892–10900. [Google Scholar] [CrossRef]
  39. Marah, H.; Challenger, M. Adaptive hybrid reasoning for agent-based digital twins of distributed multi-robot systems. Simulation 2024, 100, 931–957. [Google Scholar] [CrossRef]
  40. Rojek, I.; Dostatni, E. Machine learning methods for optimal compatibility of materials in ecodesign. Bull. Pol. Acad. Sci. Tech. Sci. 2020, 68, 199–206. [Google Scholar] [CrossRef]
  41. Li, T.; Luan, Y.; Pang, Z.; Zhang, W. Structural Digital Twin Modeling and Adaptive Pretrain-Finetune Learning for Dynamic Impact Identification on Wind Turbine Blades. IEEE Trans. Ind. Inform. 2024, 20, 10292–10303. [Google Scholar] [CrossRef]
  42. Zhao, L.; Li, T.; Zhang, E.; Lin, Y.; Wan, S.; Hawbani, A.; Guizani, M. Adaptive Swarm Intelligent Offloading Based on Digital Twin-assisted Prediction in VEC. IEEE Trans. Mob. Comput. 2024, 23, 8158–8174. [Google Scholar] [CrossRef]
  43. Qiao, D.; Li, M.; Guo, S.; Zhao, J.; Xiao, B. Resources-Efficient Adaptive Federated Learning for Digital Twin-Enabled IIoT. IEEE Trans. Netw. Sci. Eng. 2024, 11, 3639–3652. [Google Scholar] [CrossRef]
  44. Ye, J.; Qin, S.; Xiao, Q.; Jiang, W.; Tang, W.; Li, X. Adaptive Bitrate Algorithms via Deep Reinforcement Learning with Digital Twins Assisted Trajectory. IEEE Trans. Netw. Sci. Eng. 2024, 11, 3522–3535. [Google Scholar] [CrossRef]
  45. Tan, X.; Wang, M.; Wang, T.; Zheng, Q.; Wu, J.; Yang, J. Adaptive Task Scheduling in Digital Twin Empowered Cloud-Native Vehicular Networks. IEEE Trans. Veh. Technol. 2024, 73, 8973–8987. [Google Scholar] [CrossRef]
  46. Zhao, l.; Li, S.; Guan, Y.; Wan, S.; Hawbani, A.; Bi, Y.; Guizani, M. Adaptive Multi-UAV Trajectory Planning Leveraging Digital Twin Technology for Urban IIoT Applications. IEEE Trans. Netw. Sci. Eng. 2024, 11, 5349–5363. [Google Scholar] [CrossRef]
  47. Flammini, F. Data-Driven Anomaly Detection in Smart-Railways through Self-Adaptation, Process Mining, and Digital Twins. In Proceedings of the 2024 IEEE International Conference on Big Data (BigData), Washington, DC, USA, 15–18 December 2024; pp. 8653–8657. [Google Scholar]
  48. Wang, G.; Li, H.; Ye, S.; Zhao, H.; Ding, H.; Xie, S. RFWNet: A Multiscale Remote Sensing Forest Wildfire Detection Network with Digital Twinning, Adaptive Spatial Aggregation, and Dynamic Sparse Features. IEEE Trans. Geosci. Remote. Sens. 2024, 62, 1–23. [Google Scholar] [CrossRef]
  49. Xie, J.; Yang, Z.; Hu, C.C.; Yang, H.C.; Lu, Y.; Zhao, Y.Y.F. Investigation on domain adaptation of additive manufacturing monitoring systems to enhance digital twin reusability. In Proceedings of the 2024 IEEE 20th International Conference on Automation Science and Engineering, Bari, Italy, 28 August–1 September 2024; pp. 690–697. [Google Scholar]
  50. Herath, M.; Alvi, M.; Minerva, R.; Dutta, H.; Crespi, N.; Raza, S.M. Smart City Digital Twins: A Modular and Adaptive Architecture for Real-Time Data-Driven Urban Management. In Proceedings of the 2024 20th International Conference on Network and Service Management, Prague, Czech Republic, 28–31 October 2024; pp. 1–7. [Google Scholar]
  51. Cetinsaya, B.; Neumann, C.; Bruder, G.; Reiners, D.; Cruz-Neira, C. Camera-based Adaptive Line Formation and Dynamic Leader-Following Optimization (CALF-DLFO) for Drone Swarms in Real-time Updated Digital Twins. In Proceedings of the 2024 10th International Conference on Control, Decision and Information Technologies, Vallette, Malta, 1–4 July 2024; pp. 2769–2774. [Google Scholar]
  52. Häfner, P.; Bergmann, V.; Häfner, V.; Longge Michels, F.; Grethler, M.; Karande, A. An Adaptive Learning Environment for Industry 4.0 Competencies Based on a Learning Factory and Its Immersive Digital Twin. In Proceedings of the 16th International Conference on Computer Supported Education, Angers, France, 2–4 May 2024; pp. 720–731. [Google Scholar]
  53. Dittler, D.; Stauss, D.; Rentschler, P.; Stümpfle, J.; Jazdi, N.; Weyrich, M. Flexible Co-Simulation Approach for Model Adaption in Digital Twins of Power-to-X Platforms. In Proceedings of the 2024 IEEE 29th International Conference on Emerging Technologies and Factory Automation, Padova, Italy, 10–13 September 2024; pp. 1–4. [Google Scholar]
  54. Marasigan, J.A.L.; Wong, J.H. Adaptive Robotics: Integrating Robotic Simulation, AI, Image Analysis, and Cloud-Based Digital Twin Simulation for Dynamic Task Completion. In Artificial Intelligence in HCI, Proceedings of the 5th International Conference, AI-HCI 2024, Held as Part of the 26th HCI International Conference, HCII 2024, Washington, DC, USA, 29 June–4 July 2024; Lecture Notes in Computer Science Series; Springer Nature: Cham, Switzerland, 2024; Volume 53, pp. 262–271. [Google Scholar]
  55. Zehnder, E.; Torgersen, L.N.S.; Ask, T.F.; Knox, B.J.; Morgenstern, H.; Gaiser, J.; Naudet, J.; Garcia-Perez, A.; Stahl, C. Digital Twins and Extended Reality for Tailoring Better Adapted Cybersecurity Trainings in Critical Infrastructures. In Augmented Cognition, Proceedings of the 18th International Conference, AC 2024, Held as Part of the 26th HCI International Conference, HCII 2024, Washington, DC, USA, 29 June–4 July 2024; Lecture Notes in Computer Science Series; Springer Nature: Cham, Switzerland, 2024; Volume 11, pp. 233–252. [Google Scholar]
  56. Verda Çakir, L.; Duran, K.; Thomson, C.; Broadbent, M.; Canberk, B. AI in Energy Digital Twining: A Reinforcement Learning-Based Adaptive Digital Twin Model for Green Cities. In Proceedings of the ICC 2024—IEEE International Conference on Communications, Denver, CO, USA, 9–13 June 2024; pp. 4767–4772. [Google Scholar]
  57. Chen, Y.; Yu, H.; Guo, Q.; Zhao, S.; Taleb, T. Dynamic Edge AI Service Management and Adaptation Via Off-Policy Meta-Reinforcement Learning and Digital Twin. In Proceedings of the ICC 2024—IEEE International Conference on Communications, Denver, CO, USA, 9–13 June 2024; pp. 867–872. [Google Scholar]
  58. Liu, S.; Wu, W.; Li, S.; Luan, T.H.; Zhang, N. Digital Twin-Assisted Adaptive Preloading for Short Video Streaming. In Proceedings of the ICC 2024—IEEE International Conference on Communications, Denver, CO, USA, 9–13 June 2024; pp. 1431–1436. [Google Scholar]
  59. Liang, H.; Wang, Y.; Wei, C.; Yu, S.; Fang, F. An Adaptive Scheduling Method for Infrastructure Construction Progress Based on Digital Twin. In Proceedings of the 2024 IEEE 22nd International Conference on Industrial Informatics, Beijing, China, 17–20 August 2024; pp. 1–6. [Google Scholar]
  60. Fu, K.; Song, R.; Pant, P.; Hamacher, T.; Peric, V.S. A Self-adaptive Digital Twin with Broad Learning System: An Example of Heat Pump. In Proceedings of the 2024 IEEE PES Innovative Smart Grid Technologies Europe, Dubrovnik, Croatia, 14–17 October 2024; pp. 1–5. [Google Scholar]
  61. Artioli, E.; Tashtarian, F.; Timmerer, C. DIGITWISE: Digital Twin-based Modeling of Adaptive Video Streaming Engagement. In Proceedings of the 15th ACM Multimedia Systems Conference, Bari, Italy, 15–18 April 2024; pp. 78–88. [Google Scholar]
  62. D’Ingillo, R.; D’Amico, A.; Ambrosone, R.; Virgillito, E.; Gatto, V.; Straullu, S.; Aquilino, F.; Curri, V. Deep Learning Gain and Tilt Adaptive Digital Twin Modeling of Optical Line Systems for Accurate OSNR Predictions. In Proceedings of the 2024 International Conference on Optical Network Design and Modeling, Madrid, Spain, 6–9 May 2024; pp. 1–3. [Google Scholar]
  63. Taheri, E.; Bruel, P.; Prakash, P.; Rattihalli, G.; Hogade, N.; Dhakal, A.; Hong Enriquez, R.P.; Wilde, T.; Popokh, L.; Milojicic, D.J.; et al. DyTwin: Federated Adaptive Digital Twins for Data Centers—Visualization and Anomaly Detection. In Proceedings of the SC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis, Atlanta, GA, USA, 17–22 November 2024; pp. 847–852. [Google Scholar]
  64. Xu, S.; Pirbhulal, S.; Abie, H. An Architecture of Adaptive Cognitive Digital Twins for Resilient Healthcare Infrastructures and Services. In Secure and Resilient Digital Transformation of Healthcare, Proceedings of the Second International Workshop, SUNRISE 2024, Bergen, Norway, 25 November 2024; Communications in Computer and Information Science; Springer Nature: Cham, Switzerland, 2024; pp. 3–22. [Google Scholar]
  65. Barat, S.; Yadav, A.; Thogaru, H.; Kulkarni, V.; Bhattacharya, K. Imparting Adaptiveness and Resilience to Parcel Delivery Networks: A Digital Twin Centric Simulation Based Approach. In Proceedings of the 2024 Winter Simulation Conference, Orlando, FL, USA, 15–18 December 2024; pp. 2999–3010. [Google Scholar]
  66. Jaber, M.; Karakra, A.; Alsadeh, A.S.; Taweel, A. Digital Twin Adapted Agile Software Development Life Cycle. In Agile Processes in Software Engineering and Extreme Programming—Workshops; Springer: Berlin, Germany, 2024; pp. 196–202. [Google Scholar]
  67. Li, T.; Long, Q.; Chai, H.; Zhang, S.; Jiang, F.; Liu, H.; Huang, W.; Jin, D.; Li, Y. Generative AI Empowered Network Digital Twins: Architecture, Technologies, and Applications. ACM Comput. Surv. 2025, 57, 157. [Google Scholar] [CrossRef]
  68. Allamaa, J.P.; Patrinos, P.; Van der Auweraer, H.; Son, T.D. Learning Based NMPC Adaptation for Autonomous Driving using Parallelized Digital Twin. In Proceedings of the ICC 2024—IEEE International Conference on Communications, Denver, CO, USA, 9–13 June 2024. [Google Scholar]
  69. Jia, P.; Wang, X.; Shen, X. Hierarchical Digital Twin for Efficient 6G Network Orchestration via Adaptive Attribute Selection and Scalable Network Modeling. arXiv 2024, arXiv:2403.12398. [Google Scholar] [CrossRef]
  70. Huang, H.; Xu, X. Advancing Digital Twin Implementation using Edge Adapters based on Containerization. In Proceedings of the 2023 IEEE 19th International Conference on Automation Science and Engineering, Auckland, New Zealand, 26–30 August 2023; pp. 1–6. [Google Scholar]
  71. Dhakal, A.; Ran, X.; Wang, Y.; Chen, J.; Ramakrishnan, K. SLAM-share: Visual simultaneous localization and mapping for real-time multi-user augmented reality. In Proceedings of the 18th ACM CoNEXT, Rome, Italy, 6–9 December 2022; pp. 293–306. [Google Scholar]
  72. Dittler, D.; Stegmaier, V.; Jazdi, N.; Weyrich, M. Illustrating the benefits of efficient creation and adaption of behavior models in intelligent Digital Twins over the machine life cycle. J. Manuf. Syst. 2024, 76, 520–539. [Google Scholar] [CrossRef]
  73. Jimmy, S.; Berhane, K.; Muhammad, K. User Digital Twin-Driven Video Streaming for Customized Preferences and Adaptive Transcoding. arXiv 2024, arXiv:2407.09766. [Google Scholar] [CrossRef]
  74. Tezzele, M.; Carr, S.; Topcu, U.; Willcox, K.E. Adaptive planning for risk-aware predictive digital twins. arXiv 2024, arXiv:2407.20490. [Google Scholar] [CrossRef]
  75. Duong, T.Q.; Nguyen, L.D.; Bui, T.T.; Pham, K.D.; Karagiannidis, G.K. Machine learning-aided real-time optimized multibeam for 6G integrated satellite-terrestrial networks: Global coverage for mobile services. IEEE Netw. 2023, 37, 86–93. [Google Scholar] [CrossRef]
  76. Zhang, M.; Marklund, H.; Dhawan, N.; Gupta, A.; Levine, S.; Finn, C. Adaptive risk minimization: Learning to adapt to domain shift. Adv. Neural Inf. Process. Syst. 2021, 34, 23664–23678. [Google Scholar]
  77. Shafiei Dizaji, M. Deep learning-based Visual Measurement Extraction within an Adaptive Digital Twin Framework from Limited Data Using Transfer Learning. arXiv 2024, arXiv:2410.05403. [Google Scholar] [CrossRef]
  78. He, M.; Wu, H.; Zhou, C.; Hu, S.; Tang, Z.; Zhuang, W. Digital Twin-Assisted Robust and Adaptive Resource Slicing in LEO Satellite Networks. In Proceedings of the GLOBECOM 2024–2024 IEEE Global Communications Conference, Cape Town, South Africa, 8–12 December 2024. [Google Scholar]
  79. Wang, Y.; Wang, S.; Yang, W.; Shen, C.; Li, J. A digital-twin-based adaptive multi-objective Harris Hawks Optimizer for dynamic hybrid flow green scheduling problem with dynamic events. Appl. Soft Comput. 2023, 143, 110274. [Google Scholar] [CrossRef]
  80. Tao, Y.; Wu, J.; Lin, X.; Yang, W. DRL-Driven Digital Twin Function Virtualization for Adaptive Service Response in 6G Networks. IEEE Netw. Lett. 2023, 5, 125–129. [Google Scholar] [CrossRef]
  81. Fang, W.; Zhang, H.; Qian, W.; Guo, Y.; Li, S.; Liu, Z.; Liu, C.; Hong, D. An Adaptive Job Shop Scheduling Mechanism for Disturbances by Running Reinforcement Learning in Digital Twin Environment. J. Comput. Inf. Sci. Eng. 2023, 23, 051013. [Google Scholar] [CrossRef]
  82. Xu, H.; Wu, J.; Pan, Q.; Liu, X.; Verikoukis, C.V. Digital Twin and Meta RL Empowered Fast-Adaptation of Joint User Scheduling and Task Offloading for Mobile Industrial IoT. IEEE J. Sel. Areas Commun. 2023, 41, 3254–3266. [Google Scholar] [CrossRef]
  83. Wang, J.; Zhang, Z.; Liu, Z.; Han, B.; Bao, H.; Ji, S. Digital twin aided adversarial transfer learning method for domain adaptation fault diagnosis. Reliab. Eng. Syst. Saf. 2023, 234, 109152. [Google Scholar] [CrossRef]
  84. Yan, S.; Zhong, X.; Shao, H.; Ming, Y.; Liu, C.; Liu, B. Digital twin-assisted imbalanced fault diagnosis framework using subdomain adaptive mechanism and margin-aware regularization. Reliab. Eng. Syst. Saf. 2023, 239, 109522. [Google Scholar] [CrossRef]
  85. Zhang, Y.; Ji, J.; Ren, Z.; Ni, Q.; Gu, F.; Feng, K.; Yu, K.; Ge, J.; Lei, Z.; Liu, Z. Digital twin-driven partial domain adaptation network for intelligent fault diagnosis of rolling bearing. Reliab. Eng. Syst. Saf. 2023, 234, 109186. [Google Scholar] [CrossRef]
  86. Kaewunruen, S.; AbdelHadi, M.; Kongpuang, M.; Pansuk, W.; Remennikov, A.M. Digital Twins for Managing Railway Bridge Maintenance, Resilience, and Climate Change Adaptation. Sensors 2023, 23, 252. [Google Scholar] [CrossRef] [PubMed]
  87. Jaber, M.; Karakra, A.; Awadallah, R.; Barghouthi, H.; Taweel, A. Adapting the Software Development Life Cycle for Digital Twin Development in Healthcare. In Proceedings of the 2023 20th ACS/IEEE International Conference on Computer Systems and Applications, Giza, Egypt, 4–7 December 2023; pp. 1–6. [Google Scholar]
  88. Köhler-Bussmeier, M.; Rölke, H. Petri-Nets@Run.Time: Handling Uncertainty during Run-Time Adaptation using Digital Twins. In Proceedings of the 2023 International Workshop on Petri Nets and Software Engineering, Geneva, Switzerland, 24–25 June 2024; pp. 34–52. [Google Scholar]
  89. Soykan, B.; Bany Abdelnabi, A.A.; Rabadi, G. Adaptive Scheduling in Dynamic Environments: A Data-Driven Approach With Digital Twins And Deep Reinforcement Learning. In Proceedings of the 2023 IEEE 3rd International Conference on Digital Twins and Parallel Intelligence (DTPI), Orlando, FL, USA, 7–9 November 2023; pp. 1–6. [Google Scholar]
  90. Emre Demir, O.; Colledani, M.; Teicher, U.; Ben Achour, A.; Seidel, A.; Hänel, A.; Teferi Yemane, A.; Borzi, G.P.; Uhlenkamp, J.-F.; Eberlein, S.; et al. Vertically Integrated Digital Twins for Rapid Adaptation Of Manufacturing Value Chains. In Proceedings of the International Conference on Modelling and Simulation 2023, Florence, Italy, 20–23 June 2023; pp. 435–441. [Google Scholar]
  91. Dittler, D.; Lierhammer, P.; Braun, D.; Müller, T.; Jazdi, N.; Weyrich, M. A Novel Model Adaption Approach for intelligent Digital Twins of Modular Production Systems. In Proceedings of the 2023 IEEE 28th International Conference on Emerging Technologies and Factory Automation, Sinaia, Romania, 12–15 September 2023; pp. 1–8. [Google Scholar]
  92. Burattini, S. Supporting Adaptive Multi-Agent Systems with Digital Twins Environments. In Multi-Agent Systems, Proceedings of the 20th European Conference, EUMAS 2023, Naples, Italy, 14–15 September 2023; Lecture Notes in Computer Science; Springer Nature: Cham, Switzerland, 2024; pp. 445–451. [Google Scholar]
  93. Zhang, H.; Luo, T.; Wang, Q. Adaptive Digital Twin Server Deployment for Dynamic Edge Networks in IoT System. In Proceedings of the 2023 IEEE/CIC International Conference on Communications in China (ICCC), Dalian, China, 10–12 August 2023; pp. 1–6. [Google Scholar]
  94. Mourtzis, D.; Varsamis, A.; Zygomalas, S.; Angelopoulos, J. Adaptive Control of 3D Printer Based on Digital Twin Concept. In Innovative Intelligent Industrial Production and Logistics, Proceedings of the 4th International Conference, IN4PL 2023, Rome, Italy, 15–17 November 2023; Communications in Computer and Information Science; Springer Nature: Cham, Switzerland, 2023; pp. 3–17. [Google Scholar]
  95. Spaney, P.; Becker, S.; Ströbel, R.; Fleischer, J.; Zenhari, S.; Möhring, H.-C.; Splettstößer, A.-K.; Wortmann, A. A Model-Driven Digital Twin for Manufacturing Process Adaptation. In Proceedings of the 2023 ACM/IEEE International Conference on Model Driven Engineering Languages and Systems Companion (MODELS-C), Västerås, Sweden, 1–6 October 2023; pp. 465–469. [Google Scholar]
  96. Antunes, A. Designing a Digital Twin for Adaptive Serious Games-based Therapy. In Proceedings of the 22nd International Conference on Mobile and Ubiquitous Multimedia, Vienna, Austria, 3–6 December 2023; pp. 568–570. [Google Scholar]
  97. Alsayed Kassem, J.; Zhong, L.; Taal, A.; Grosso, P. Adaptive Services Function Chain Orchestration For Digital Health Twin Use Cases: Heuristic-boosted Q-Learning Approach. In Proceedings of the 2023 IEEE 9th International Conference on Network Softwarization, Madrid, Spain, 19–23 June 2023; pp. 187–191. [Google Scholar]
  98. Houé, A.; Seguin, C.; Frizon de Lamotte, F.; Julien, N.; Allègre, W. Modular and Distributed Architecture Using an Embedded Digital Twin for Adaptation of Assistive Technologies. In Service Oriented, Holonic and Multi-Agent Manufacturing Systems for Industry of the Future; Studies in Computational Intelligence; Springer Nature: Cham, Switzerland, 2023; pp. 26–37. [Google Scholar]
  99. Wei, P.; Zhou, W.; Fu, L.; Wang, H.; Li, X.; Huang, X.; Li, J.; Ding, Y.; Cao, Q. Study on Digital Twin Technology to Drilling Performance Improvement and Risk Reduction. In Proceedings of the SPE Conference at Oman Petroleum & Energy Show, Muscat, Oman, 12–14 May 2025. [Google Scholar] [CrossRef]
  100. Zhang, N.; Bahsoon, R.; Tziritas, N.; Theodoropoulos, G. A Digital Twin Approach for Adaptive Compliance in Cyber-Physical Systems: Case of Smart Warehouse Logistics. arXiv 2023, arXiv:2310.07116. [Google Scholar] [CrossRef]
  101. Menezes Rebello, C.; Jäschke, J.; Nogueira, I.B.R. Digital Twin Framework for Optimal and Autonomous Decision-Making in Cyber-Physical Systems: Enhancing Reliability and Adaptability in the Oil and Gas Industry. arXiv 2023, arXiv:2311.12755. [Google Scholar]
  102. Zhang, C.; Zhou, G.; Jing, Y.; Wang, R.; Chang, F. A Digital Twin-Based Automatic Programming Method for Adaptive Control of Manufacturing Cells. IEEE Access 2022, 10, 80784–80793. [Google Scholar] [CrossRef]
  103. Dai, Y.; Zhang, Y. Adaptive Digital Twin for Vehicular Edge Computing and Networks. J. Commun. Inf. Netw. 2022, 7, 48–59. [Google Scholar] [CrossRef]
  104. Liu, S.; Lu, Y.; Zheng, P.; Shen, H.; Bao, J. Adaptive reconstruction of digital twins for machining systems: A transfer learning approach. Robot. Comput.-Integr. Manuf. 2022, 78, 102390. [Google Scholar] [CrossRef]
Applsci 15 08525 g001
Figure 1. PRISMA flow diagram of the review process.
Figure 1. PRISMA flow diagram of the review process.
Applsci 15 08525 g001
Applsci 15 08525 g002
Figure 2. Publications by category according to WoS.
Figure 2. Publications by category according to WoS.
Applsci 15 08525 g002
Applsci 15 08525 g003
Figure 3. Publications by micro-topic.
Figure 3. Publications by micro-topic.
Applsci 15 08525 g003
Applsci 15 08525 g004
Figure 4. Publications by SDG.
Figure 4. Publications by SDG.
Applsci 15 08525 g004
Applsci 15 08525 g005
Figure 5. Publications by year.
Figure 5. Publications by year.
Applsci 15 08525 g005
Applsci 15 08525 g006
Figure 6. Publications by type.
Figure 6. Publications by type.
Applsci 15 08525 g006
Applsci 15 08525 g007
Figure 7. Publications by subject area.
Figure 7. Publications by subject area.
Applsci 15 08525 g007
Applsci 15 08525 g008
Figure 8. Publications by country.
Figure 8. Publications by country.
Applsci 15 08525 g008
Applsci 15 08525 g009
Figure 9. Publications by funding.
Figure 9. Publications by funding.
Applsci 15 08525 g009
Table 1. Bibliometric analysis procedure used in this study (own approach).
Table 1. Bibliometric analysis procedure used in this study (own approach).
Stage NameTasks
Defining research objectivesDefining goals of the bibliometric analysis
Selecting databases and data collectionsChoosing appropriate dataset(s) and developing research queries according to the study goals
Data preprocessingCleaning the collected data to remove duplicates and irrelevant records
Bibliometric software selectionChoosing a suitable bibliometric software tool for analysis
Data analysisDescription, Author, Journal, Area/Topics, Institution/Country, etc.
Visualization (where possible)Visualizing the analysis results to present insights
Interpretation and discussionInterpreting findings in the context of the research goals
Table 2. Detailed search queries over databases.
Table 2. Detailed search queries over databases.
ParameterDescription
Inclusion criteriaArticles (original, reviews), books, and chapters up to ten years after publication, including conference proceedings, in English
Exclusion criteriaBooks older than 10 years, letters, communication, editorials, conference abstracts without full text, and other languages than English
Keywords usedartificial intelligence, AI, machine learning, ML, deep learning, DL, digital twin, personalization, adaptation
Used field codes (WoS)“Subject” field (consisting of title, abstract, keyword plus, and other keywords)
Used field codes (Scopus)article title, abstract, and keywords
Used field codes (dblp)Manually
Boolean operators usedYes, e.g., “digital twin” AND (“AI” OR “ML”) AND adaptation
Applied filtersResults refined by publication year, document type (e.g., articles, reviews), and subject area (e.g., industry, engineering)
Iteration and validation optionsQuery run iteratively, refinement based on the results, and validation by ensuring relevant articles appear among the top hits
Leverage truncation and wildcards usedUsed symbols like * for word variations (e.g., “digital twin*”) and ? for alternative spellings (e.g., “personali?ation”)
Table 3. General summary of results of bibliographic analysis (WoS, Scopus, dblp).
Table 3. General summary of results of bibliographic analysis (WoS, Scopus, dblp).
Parameter/FeatureValue
Leading types of publicationArticle (44.00%), conference paper (36.00%), review article (16.00%)
Leading areas of scienceComputer science information systems, computer science artificial intelligence
Leading topicsIndustrial: Design and Manufacturing (28.40%), Telecommunication (8.23%)
Leading countriesGermany, Spain, USA, China
Leading scientistsYi C, Gao Y
Leading affiliationsChinese Academy of Sciences
Leading funders (where information available)European Union
Sustainable development goals (SDGs)Industry Innovation and Infrastructure, Responsible Consumption and Production, Sustainable Cities and Communities, Good Health and Wellbeing
Table 4. A comparison showing the most commonly used ways of personalizing DTs to optimize adaptation in simple DTs versus more complex AI-based DTs (own version).
Table 4. A comparison showing the most commonly used ways of personalizing DTs to optimize adaptation in simple DTs versus more complex AI-based DTs (own version).
Personalization MethodSimple DTsComplex AI-Based DTs
User-defined parametersManual input of parameters by usersAutomated adjustment of parameters based on AI models
Sensor(s) data integrationBasic real-time data feedsAdvanced multi-sensor fusion with AI-driven predictions
Rule-based adaptationPredefined if–then rulesAI-driven dynamic rule generation and self-learning
ML personalizationNot typically usedUses, e.g., DL for adaptation
Predictive analyticsSimple trend analysisAI-powered predictive maintenance and optimization
Simulation capabilitiesLimited, predefined scenariosDynamic, real-time simulations based on AI models
Decision-making supportStatic reports and alertsAI-driven decision support with automated recommendations
Context awareness Basic environment recognition Deep contextual adaptation using AI and IoT data
Real-time adaptationLimited, often requires manual interventionContinuous and autonomous real-time adaptation
Human interaction adaptationManual adjustments with user feedbackAI-driven adjustments based on behavioral patterns
Table 5. Results of the review: the most frequently used ways to develop personalized DTs to optimize adaptation from simple DTs toward more complex AI-based DTs (own version).
Table 5. Results of the review: the most frequently used ways to develop personalized DTs to optimize adaptation from simple DTs toward more complex AI-based DTs (own version).
Way of DevelopmentDetails
User-defined parametersIn the initial stages, simple DTs rely on user-defined parameters, where manual input is required to configure the model for basic adaptation
Sensor data integrationAs DTs evolve, IoT sensor integration enables real-time data collection, enabling dynamic adaptations based on environmental changes
Rule-based adaptationBasic rule-based systems using predefined logic (e.g., if–then conditions) help DTs respond to changes, although they remain rigid and require manual updates
Cloud integrationStoring and processing DT data in the cloud increases scalability and connectivity, making it easier to implement AI-driven adaptations
Edge computing for faster responseTo improve real-time adaptability, edge computing processes data closer to the source, reducing latency and enabling faster decision making
ML modelsIncorporating ML algorithms enables DTs to identify patterns in data, optimizing performance by learning from past experience
Predictive analyticsAdvanced DTs use predictive analytics to anticipate potential issues, enabling proactive maintenance and decision making
AI-driven optimizationAI-driven DTs leverage deep learning to dynamically optimize processes, continually improving their behavior based on incoming data
Generative AI for scenario simulationAI-driven DTs, including generative AI models, create synthetic scenarios to test and optimize responses before applying them to the real world
Autonomous decision makingAs complexity increases, DTs are moving toward AI-driven autonomy, making data-driven decisions with minimal human intervention
Behavioral and contextual adaptationAI-based DTs can analyze user behavior and environmental contexts, adapting their capabilities to specific needs
Integration of natural language processing (NLP)Generative AI empowers DTs by enabling human-like conversational interactions, improving usability and adaptability
Multi-agent AI systemsAdvanced AI-based twins feature multiple AI agents that work together to optimize adaptation and decision making across the system
Self-learning and evolutionThe most complex DTs leverage reinforcement learning, enabling continuous self-improvement without predefined rules
Human–AI collaborationUltimately, DTs evolve into AI-assisted systems where human and AI experts work together to provide optimal adaptation and innovation
Table 6. Emerging trends, research gaps, and dominant themes related to the personalization of AI-based DTs in engineering.
Table 6. Emerging trends, research gaps, and dominant themes related to the personalization of AI-based DTs in engineering.
Feature/AreaEmerging TrendsResearch GapsDominant Themes
Signal-level dataIntegration of multimodal sensor data for high-fidelity digital replicasStandardized preprocessing pipelines for noise reduction and synchronizationHigh-resolution, real-time signal acquisition and interpretation
IoT integrationSeamless data flow between physical assets and digital twins via connected devicesData heterogeneity, latency issues, and lack of interoperability among IoT platformsContinuous monitoring, remote diagnostics, and asset connectivity
Edge computingLocal (close-to-the-source) processing for latency-sensitive applications and real-time personalizationScalability of models across distributed edge nodesReal-time inference, privacy-preserving analytics, decentralized intelligence
AI/MLAdaptive AI models that learn user/asset-specific behavior over timeExplainability, transfer learning for limited-data environmentsPredictive maintenance, anomaly detection, intelligent decision support
Human-in-the-loopInteractive interfaces for user feedback and real-time customizationModeling human behavior and integrating user intent into system logicUser-centric design, collaborative control, intuitive interfaces
Cybersecurity and privacyPersonalized access control and encrypted communication protocolsRobustness against adversarial attacks in real-time AI-based control systemsTrustworthiness, secure data exchange, compliance with General Data Protection Regulation (GDPR)/industry standards
Standardization and InteroperabilityPush for open architectures and common modeling frameworksAbsence of widely adopted standards for AI-driven personalization componentsModel sharing, cross-platform compatibility, modular twin architecture
Domain-specific DTsCustomization for aerospace, manufacturing, healthcare, etc.Lack of domain-specific datasets and validation benchmarksApplication-tailored modeling, fine-tuned simulations, task-specific optimization
Simulation and feedbackBi-directional learning between real and virtual systemsClosed-loop validation of adaptive models remains underexploredContinuous learning, simulation-informed control strategies
Sustainability and energy efficiencyGreen AI techniques and energy-aware modelingBalancing personalization complexity with energy/resource constraintsLow-power design, eco-conscious computation, sustainable system integration
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Rojek, I.; Mikołajewski, D.; Dostatni, E.; Cybulski, J.; Kozielski, M. Personalization of AI-Based Digital Twins to Optimize Adaptation in Industrial Design and Manufacturing—Review. Appl. Sci. 2025, 15, 8525. https://doi.org/10.3390/app15158525

AMA Style

Rojek I, Mikołajewski D, Dostatni E, Cybulski J, Kozielski M. Personalization of AI-Based Digital Twins to Optimize Adaptation in Industrial Design and Manufacturing—Review. Applied Sciences. 2025; 15(15):8525. https://doi.org/10.3390/app15158525

Chicago/Turabian Style

Rojek, Izabela, Dariusz Mikołajewski, Ewa Dostatni, Jan Cybulski, and Mirosław Kozielski. 2025. "Personalization of AI-Based Digital Twins to Optimize Adaptation in Industrial Design and Manufacturing—Review" Applied Sciences 15, no. 15: 8525. https://doi.org/10.3390/app15158525

APA Style

Rojek, I., Mikołajewski, D., Dostatni, E., Cybulski, J., & Kozielski, M. (2025). Personalization of AI-Based Digital Twins to Optimize Adaptation in Industrial Design and Manufacturing—Review. Applied Sciences, 15(15), 8525. https://doi.org/10.3390/app15158525

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop