Volume 23, Issue 59, January - June, 2026

HESTIA: Hybrid Edge-Supported Twin for Infrastructure Analysis using Graph and Temporal Networks

Melsiline Reeba Melkias¹, Subha Hency Jose Paul¹, Mano Jemila Manoharan Rethnabai², Lawrence Kebila Anns Subi³, Anantha Christu Raj Palayyan^4♦, Sweety Jose Paul⁵

¹Division of Biomedical Engineering, Karunya Institute of Technology and Science, Coimbatore, Tamilnadu, India
²Department of Electrical and Electronics Engineering, Dr.N.G.P. Institute of Technology, Coimbatore, Tamilnadu, India
³Department of Electronics and Communication Engineering, Ponjesly College of Engineering, Nagercoil, Tamilnadu, India
⁴Division of Robotics Engineering, Karunya Institute of Technology and Science, Coimbatore, Tamilnadu, India
⁵Department of Electrical and Electronics Engineering, PSG College of Technology, Coimbatore, Tamilnadu, India

^♦Corresponding Author
Anantha Christu Raj Palayyan, Division of Biomedical Engineering, Karunya Institute of Technology and Science, Coimbatore, Tamilnadu, India

ABSTRACT

For ensuring real-time structural health monitoring (SHM), there is a need of reliable framework that balances interpretability, computational efficacy, and adaptability. As an effort, this paper introduces “HESTIA”, a novel hybrid edgesupported twin for infrastructure analysis which integrates physics-based reduced-order-models with advanced edge-deployed deep learning. Importantly, the proposed methodology is suitable for monitoring progressive damage such as cracks and fatigue, which are critical indicators of structural weakening. Additionally, The HESTIA approach is highly applicable in areas like lab prototypes, scaled models, and campus SHM testbeds. The embedded digital twin offers state estimation and uncertainty quantification. On the other hand, two AI components – Graph Neural Network (GNN) is utilized for structural topology representation and Temporal Convolutional Network (TCN) aids in effective time-series analysis. This hybrid combination enables accurate modelling of both spatial dependencies & temporal evolution across sensor nodes and structural responses. Next, the outcome of digital twin and networks are fused via the Bayesian ensemble strategy for robust anomaly detection and early damage identification. The HESTIA architecture is optimized with Harris Hawks Optimization (HHO) that leverages feature reweighting and decision threshold tuning. Overall, the suggested model demonstrates its potential in nextgeneration SHM solution for pro-active infrastructure maintenance through attainment of 85 ms in inference time, 5% in computational overload, and competitive anomaly detection rate of 93%. Further, the approach inhibited a superior recital of specificity 96.2%, precision of 93.1%, and accuracy of 92.6% making it a reliable option for real-time SHM applications.

Keywords: Hybrid digital twin, graph neural network, temporal convolutional network, edge AI, Bayesian fusion, real-time structural analysis, civil infrastructure, digital twin, crack detection, fatigue identification, scaled models, campus SHM testbeds, and structural health monitoring.

Indian Journal of Engineering, 2026, 23(59), e2ije1708

PDF

DOI: https://doi.org/10.54905/disssi.v23i59.e2ije1708

Published: 21 February 2026