TY  - JOUR
AU  - Samal, Ajit Kumar
AU  - Mishra, Brojo Kishore
PY  - 2026
DA  - 2026/09/02
TI  - GenAI-Enhanced Federated Digital Twins for Predictive Resilience in IoT-Enabled Sustainable Smart Infrastructure
JO  - Next-Generation Computing Systems and Technologies
T2  - Next-Generation Computing Systems and Technologies
JF  - Next-Generation Computing Systems and Technologies
VL  - 2
IS  - 3
SP  - 76
EP  - 88
DO  - 10.62762/NGCST.2026.306047
UR  - https://www.icck.org/article/abs/NGCST.2026.306047
KW  - digital twins
KW  - federated learning
KW  - generative AI
KW  - IoT infrastructure
KW  - predictive resilience
KW  - sustainable smart cities
AB  - This study introduces GenAI-FDT (GenFedTwin Resilience), a decentralized architecture designed to address the growing vulnerability of IoT-enabled smart infrastructures to climate change, particularly in the context of data silos, strict privacy regulations, and the scarcity of ground-truth data for rare catastrophic events such as floods in Maharashtra. Unlike traditional centralized digital twins, which struggle with data heterogeneity and excessive communication costs, the proposed framework integrates quantized diffusion-based Generative AI, semi-supervised federated learning (FedProx), and lightweight LSTM-CNN digital twins deployed on resource-constrained edge devices like Raspberry Pi using TinyML. Local IoT nodes generate privacy-preserving failure traces from limited real-world data while consuming less than 1mW of inference power, and their model parameters are aggregated via Eclipse Ditto into a global resilience twin, enabling real-time ``what-if'' simulations without reliance on energy-intensive data centers. Empirical evaluations on augmented public datasets with 25% non-IID heterogeneity demonstrate clear advantages over centralized baselines, including a 10% improvement in anomaly detection (F1-score of 92.3) and a marked improvement in predicting infrastructure downtime during disasters. Moreover, the edge-native design yields substantial sustainability gains, reducing energy consumption by 21% compared to vanilla federated learning baselines, with projected CO$_2$ emissions of 12--18g per communication round at 100-node scale. By delivering a robust, low-carbon framework that upholds data sovereignty and supports scalable IoT-Federated Digital Twins for Society~5.0, GenAI-FDT directly advances UN SDG~11, promoting fair and lasting urban resilience while filling critical gaps in the current literature.
SN  - 3070-3328
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Samal2026GenAIEnhan,
  author = {Ajit Kumar Samal and Brojo Kishore Mishra},
  title = {GenAI-Enhanced Federated Digital Twins for Predictive Resilience in IoT-Enabled Sustainable Smart Infrastructure},
  journal = {Next-Generation Computing Systems and Technologies},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {76-88},
  doi = {10.62762/NGCST.2026.306047},
  url = {https://www.icck.org/article/abs/NGCST.2026.306047},
  abstract = {This study introduces GenAI-FDT (GenFedTwin Resilience), a decentralized architecture designed to address the growing vulnerability of IoT-enabled smart infrastructures to climate change, particularly in the context of data silos, strict privacy regulations, and the scarcity of ground-truth data for rare catastrophic events such as floods in Maharashtra. Unlike traditional centralized digital twins, which struggle with data heterogeneity and excessive communication costs, the proposed framework integrates quantized diffusion-based Generative AI, semi-supervised federated learning (FedProx), and lightweight LSTM-CNN digital twins deployed on resource-constrained edge devices like Raspberry Pi using TinyML. Local IoT nodes generate privacy-preserving failure traces from limited real-world data while consuming less than 1mW of inference power, and their model parameters are aggregated via Eclipse Ditto into a global resilience twin, enabling real-time ``what-if'' simulations without reliance on energy-intensive data centers. Empirical evaluations on augmented public datasets with 25\% non-IID heterogeneity demonstrate clear advantages over centralized baselines, including a 10\% improvement in anomaly detection (F1-score of 92.3) and a marked improvement in predicting infrastructure downtime during disasters. Moreover, the edge-native design yields substantial sustainability gains, reducing energy consumption by 21\% compared to vanilla federated learning baselines, with projected CO\$\_2\$ emissions of 12--18g per communication round at 100-node scale. By delivering a robust, low-carbon framework that upholds data sovereignty and supports scalable IoT-Federated Digital Twins for Society~5.0, GenAI-FDT directly advances UN SDG~11, promoting fair and lasting urban resilience while filling critical gaps in the current literature.},
  keywords = {digital twins, federated learning, generative AI, IoT infrastructure, predictive resilience, sustainable smart cities},
  issn = {3070-3328},
  publisher = {Institute of Central Computation and Knowledge}
}