Machine Learning-Assisted Reliability-Aware Digital Twin Synchronization for Resource-Constrained IoT Sensor Networks

Introduction

Internet of Things (IoT) systems obtain physical-state information through distributed sensing devices and deliver these measurements to monitoring, control, and decision-support applications. In many deployments, sensing is inexpensive compared with radio communication. The sensor node may also have limited energy, memory, and processing capacity. When every measurement is forwarded, communication resources are spent even when the physical state changes slowly. The opposite choice, aggressive suppression, can leave the receiver with an outdated view of the monitored process.

DT technology changes this trade-off because the receiver can maintain a virtual state between physical updates. Recent surveys describe DTs as an increasingly important component of wireless and networked systems, with synchronization quality, model fidelity, and communication overhead treated as coupled design issues [1]-[4]. A DT is therefore not useful only because it contains a model. Its state must be corrected often enough to remain consistent with the physical entity, and those corrections must be delivered over the available network.

The synchronization problem becomes more difficult when link quality differs across devices. A sensor with a stable physical signal may require few updates, while another sensor may need frequent corrections but have a weak direct path to the collection point. Recent work has addressed DT synchronization through continual reinforcement learning, hybrid inverse reinforcement learning, semantic communication, wireless resource allocation, and explicit synchronization budgets [5]-[9]. These methods show that selecting when and what to synchronize is a resource-management problem. Most, however, evaluate abstract communication resources or optimized scheduling models rather than the measured asymmetric link structure of a real environmental sensor deployment.

Freshness adds a second dimension to reconstruction accuracy. The 2025 review by Loubany et al. [10] summarizes synchronization metrics and introduces AoI as a DT-oriented extension of freshness measures. Recent papers also use AoI, statistical AoI, state staleness, or related fidelity measures when deciding when a DT should migrate, refresh, or synchronize [11]-[15]. These metrics are relevant because a numerically plausible estimate may still be based on an old physical update. In a resource-constrained network, the scheduler must therefore consider both the error expected from skipping an update and the age of the last successful observation.

The present paper considers this problem using the Intel Berkeley Research Lab sensor deployment. Recent papers continue to reuse this trace for communication-efficient sensing, as discussed in Section 2.4. Unlike papers that use only the temperature series, the present evaluation also uses the released aggregate connectivity probabilities. This permits a trace-driven question that is usually abstracted away: if a node has a poor direct link, should the system spend scarce radio resources on repeated direct attempts, retain the DT estimate, or select a two-hop path through a better-connected node?

To answer this question, a lightweight DT is combined with a reliability-aware update scheduler. Each sensor has an autoregressive virtual-state predictor. At every 30-second slot, the scheduler ranks candidate updates from their prediction behavior, current AoI, route reliability, and radio cost. Direct and two-hop paths are compared using the recorded connectivity values, and relaying is used only when the expected delivery probability per radio transmission improves. The approach is intentionally compact so that the contribution of each component can be isolated through ablation rather than hidden inside a large learned policy.

The major contributions of this paper are summarized below:

• The synchronization policy assigns each sensor an update score based on its prediction behavior, AoI, recorded path reliability, and transmission cost. Updates are then selected subject to a fixed radio budget in each time slot.

• For every scheduled sensor, the direct route is compared with its best available two-hop alternative using the asymmetric Intel connectivity matrix. A relay is selected only when the two-hop delivery probability, after accounting for the cost of two radio transmissions, is better than direct delivery.

• The method is evaluated on a real five-day temperature trace containing 52 sensor nodes. Training, validation, and testing are performed in chronological order, and the held-out evaluation is repeated over 30 matched wireless realizations.

• At 9.62% of full radio traffic, the proposed method obtains 0.2076 ± 0.0042 RMSE, compared with 0.3921 ± 0.0330 for the strongest direct-link DT ablation. Mean AoI is reduced from 32.01 to 12.00 slots. The proposed method also retains its accuracy advantage across the tested 3.85%-28.85% traffic range.

The remainder of the paper is organized as follows. Section 2 reviews recent work on DT synchronization, freshness-aware state maintenance, edge-resource management, and adaptive sensing. Section 3 presents the system model and the proposed synchronization method. Section 4 describes the dataset, experimental protocol, baselines, and modeling assumptions. Section 5 presents the results and limitations. Section 6 concludes the paper and highlight future work.

Complete Article

The complete article, including all figures, tables, equations and algorithms, is available in the official publication PDF.

View Full PDF

Conclusion

This paper presented a reliability-aware DT synchronization method for resource-constrained IoT sensor networks. The method combines an AR(2) virtual-state predictor with AoI, measured link reliability, and radio-transmission cost. Direct delivery is retained for most sensors. A two-hop relay is selected only when its expected success probability per radio transmission exceeds the direct alternative. The design therefore targets weak links without converting the full network to multi-hop forwarding.

The method was evaluated with a real five-day Intel Berkeley temperature trace and the corresponding aggregate connectivity measurements. At five transmissions per 30-second slot, equal to 9.62% of full telemetry, the proposed method obtains 0.2076 ± 0.0042 RMSE. The direct-link-aware DT obtains 0.3921 ± 0.0330 under the same budget, giving a 47.07% relative RMSE reduction. Mean AoI decreases from 32.01 to 12.00 slots, and the proposed approach remains the lowest-RMSE restricted method across the tested 3.85%-28.85% traffic range.

The current work is limited to one temperature trace and an aggregate-probability channel model. Future work will extend the evaluation to multiple sensing modalities, time-varying packet traces, and an embedded or emulated network. These experiments are needed before making claims about other radio technologies or deployment environments.

Data Availability

The Intel Berkeley Research Lab sensor measurements, node locations, and connectivity data are publicly available from the Intel Lab Data repository.

References

1. M. Bhatia and R. Kumar, “Digital Twin Technology in Wireless Communication: A Comprehensive Engineering Survey, ” IEEE Internet of Things Journal, vol. 12, no. 12, pp. 19144-19166, 2025, doi: 10.1109/JIOT.2025.3565843
2. D.-H. Tran et al., “Network Digital Twin for 6G and Beyond: An End-to-End View Across Multi-Domain Network Ecosystems, ” IEEE Open Journal of the Communications Society, vol. 6, pp. 6866-6911, 2025, doi: 10.1109/OJCOMS.2025.3599866
3. T. Li, Q. Long, H. Chai, S. Zhang, F. Jiang, H. Liu, W. Huang, D. Jin, and Y. Li, “Generative AI Empowered Network Digital Twins: Architecture, Technologies, and Applications, ” ACM Computing Surveys, vol. 57, no. 6, Art. no. 157, pp. 1-43, 2025, doi: 10.1145/3711682
4. Y. Sanjalawe, S. Fraihat, S. Al-E’mari, et al., “Bridging the gap: A comprehensive survey on AI-driven digital twin networks for future wireless systems, ” Journal of King Saud University - Computer and Information Sciences, vol. 38, Art. no. 113, 2026, doi: 10.1007/s44443-026-00522-y
5. H. Tong, M. Chen, J. Zhao, Y. Hu, Z. Yang, Y. Liu, and C. Yin, “Continual Reinforcement Learning for Digital Twin Synchronization Optimization, ” IEEE Transactions on Mobile Computing, vol. 24, no. 8, pp. 6843-6857, 2025, doi: 10.1109/TMC.2025.3546507
6. Q. Zhang, Y. Wang, and Z. Li, “Scheduling of Digital Twin Synchronization in Industrial Internet of Things: A Hybrid Inverse Reinforcement Learning Approach, ” IEEE Internet of Things Journal, vol. 12, no. 5, pp. 5137-5147, 2025, doi: 10.1109/JIOT.2024.3486125
7. B. Li, H. Cai, L. Liu, and Z. Fei, “Delay-Aware Digital Twin Synchronization in Mobile Edge Networks With Semantic Communications, ” IEEE Transactions on Vehicular Technology, vol. 74, no. 7, pp. 10974-10983, 2025, doi: 10.1109/TVT.2025.3548844
8. H. Yu, Y. Liu, Z. Yang, H. Sun, and M. Chen, “Optimizing Wireless Resource Management and Synchronization in Digital Twin Networks, ” IEEE Internet of Things Journal, vol. 12, no. 15, pp. 29152-29163, 2025, doi: 10.1109/JIOT.2025.3543126
9. Y. Li, W. Liang, Z. Xu, W. Xu, and X. Jia, “Budget-Constrained Digital Twin Synchronization and Its Application on Fidelity-Aware Queries in Edge Computing, ” IEEE Transactions on Mobile Computing, vol. 24, no. 1, pp. 165-182, 2025, doi: 10.1109/TMC.2024.3455357
10. A. Loubany, M. Itani, and S. Sharafeddine, “From Age of Information to Age of Digital Twin: A Review on Synchronization Metrics for IoT Networks, ” IEEE Access, vol. 13, pp. 131102-131119, 2025, doi: 10.1109/ACCESS.2025.3591589
11. K. Noroozi, T. D. Todd, D. Zhao, and G. Karakostas, “Age of Information in Digital Twin Migration, ” IEEE Transactions on Vehicular Technology, vol. 74, no. 10, pp. 16281-16294, 2025, doi: 10.1109/TVT.2025.3570505
12. Y. Xiao, Q. Du, and G. K. Karagiannidis, “Statistical Age of Information: A Risk-Aware Metric and Its Applications in Status Updates, ” IEEE Transactions on Wireless Communications, vol. 24, no. 3, pp. 2325-2340, 2025, doi: 10.1109/TWC.2024.3520324
13. Y. Wang, L. Tang, L. Wang, H. Zhang, W. Wang, D. Fang, and Q. Chen, “Digital Twin Information Synchronization Strategy for IIoT Based on Dual-Time-Scale Network Slicing Orchestration, ” IEEE Internet of Things Journal, vol. 13, no. 2, pp. 1919-1935, 2026, doi: 10.1109/JIOT.2025.3625762
14. J. Li, J. Wang, W. Liang, Q. Chen, S. K. Das, and X. Jia, “Digital Twin Freshness Maximization in Edge Computing, ” IEEE Transactions on Services Computing, vol. 19, no. 2, pp. 1107-1119, 2026, doi: 10.1109/TSC.2026.3651602
15. G. Khalaf, M. Itani, and S. Sharafeddine, “A UAV-Aided Digital Twin Framework for IoT Networks With High Accuracy and Synchronization, ” IEEE Transactions on Network and Service Management, vol. 23, pp. 3013-3025, 2026, doi: 10.1109/TNSM.2026.3670040
16. Y. Zhang, L. Wang, and W. Liang, “Deep Reinforcement Learning for Mobility-Aware Digital Twin Migrations in Edge Computing, ” IEEE Transactions on Services Computing, vol. 18, pp. 704-717, 2025, doi: 10.1109/TSC.2025.3528331
17. X. Ai, W. Liang, Y. Zhang, and W. Xu, “Fidelity-Aware Inference Services in DT-Assisted Edge Computing via Service Model Retraining, ” IEEE Transactions on Services Computing, vol. 18, no. 4, pp. 2089-2102, 2025, doi: 10.1109/TSC.2025.3586126
18. J. Li, J. Wang, W. Liang, X. Jia, and A. Y. Zomaya, “Inference Service Fidelity Maximization in DT-Assisted Edge Computing, ” IEEE Transactions on Mobile Computing, vol. 25, no. 1, pp. 1352-1366, 2026, doi: 10.1109/TMC.2025.3600390
19. X. Ai, W. Liang, and C. Liu, “Joint Optimization of Model Retraining and Inference Services in DT-Assisted Edge Computing, ” IEEE Transactions on Networking, vol. 34, pp. 1804-1819, 2026, doi: 10.1109/TON.2025.3632228
20. Y. Shi, L. Lao, X. Zheng, et al., “Real-time Spatiotemporal Synchronization for Human-robot Collaborative Systems: An Integrated Digital Twin Framework with Adaptive Data Processing, ” Journal of Intelligent & Robotic Systems, vol. 112, Art. no. 12, 2026, doi: 10.1007/s10846-025-02346-w
21. A. Ndikumana, K.-K. Nguyen, and M. Cheriet, “Digital Twin Backed Closed-Loops for Energy-Aware and Open RAN-Based Fixed Wireless Access Serving Rural Areas, ” IEEE Transactions on Mobile Computing, vol. 24, no. 3, pp. 1669-1683, 2025, doi: 10.1109/TMC.2024.3482985
22. M. A. Qadir, M. Naeem, and W. Ejaz, “Digital twin-assisted multi-layer networks for low-latency and energy-efficient communication, ” Computer Communications, vol. 241, Art. no. 108219, 2025, doi: 10.1016/j.comcom.2025.108219
23. S. Chu, J. Li, J. Wang, Y. Ni, K. Wei, W. Chen, and S. Jin, “Resource Efficient Asynchronous Federated Learning for Digital Twin Empowered IoT Network, ” IEEE Transactions on Green Communications and Networking, vol. 10, pp. 201-217, 2026, doi: 10.1109/TGCN.2025.3574143
24. Y. Zhou, Y. Fu, Z. Shi, H. H. Yang, K. Hung, and Y. Zhang, “Energy-Efficient Federated Learning and Migration in Digital Twin Edge Networks, ” IEEE Transactions on Cognitive Communications and Networking, vol. 12, pp. 743-758, 2026, doi: 10.1109/TCCN.2025.3567096
25. Y. Kwon, “Smart Manufacturing Innovation Through the Integration of Digital Twin and IoT Energy Management, ” International Journal of Control, Automation and Systems, vol. 23, pp. 3427-3439, 2025, doi: 10.1007/s12555-025-0046-1
26. L. Chang, X. Yang, K. Zhang, et al., “Blockchain-based synchronized data transmission with dynamic device management for digital twin in IIoT, ” Peer-to-Peer Networking and Applications, vol. 18, Art. no. 85, 2025, doi: 10.1007/s12083-025-01910-3
27. M. Yahya, M. Naeem, M. O. Butt, R. H. Alsisi, and W. Ejaz, “Robust Task Offloading in Digital Twin-Enabled UAV-IoT Networks Under Spatial Uncertainty, ” IEEE Open Journal of the Communications Society, vol. 7, pp. 1717-1728, 2026, doi: 10.1109/OJCOMS.2026.3663770
28. K. Hayawi, J. Sajid, A. W. Malik, and S. S. Mathew, “Digital Twin-Assisted Task Offloading for Workload Management at Fog Nodes, ” IEEE Internet of Things Journal, vol. 12, no. 13, pp. 23061-23072, 2025, doi: 10.1109/JIOT.2025.3550832
29. F. Alanazi and M. N. Alanazi, “Machine learning driven aggregation aware bitmap MAC protocol for energy efficient data transmission in WSNs, ” Scientific Reports, vol. 15, Art. no. 36893, 2025, doi: 10.1038/s41598-025-20835-8
30. G. Lu, Y. Sun, Y. Jiang, and B. Butler, “Adaptive Sampling for Spatiotemporal Anomaly Monitoring in Wireless Sensor Networks, ” arXiv:2607.15235, 2026. Accepted for IEEE ISSC 2026.
31. J. Kangethe et al., “Learning to Transmit: Volatility-Aware Predictive Communication for IoT Sensors, ” arXiv:2607.19590, 2026.
32. Intel Berkeley Research Lab, “Intel Lab Data, ” MIT Computer Science and Artificial Intelligence Laboratory. Accessed: Aug. 14, 2026. [Online]. Available: https://db.csail.mit.edu/labdata/labdata.html