A Data-Driven Paradigm: Digital Twins, IoT and AI for Predictive Management of High-Voltage T&D Assets

HVDC assets. Conventional time-based and reactive maintenance strategies are increasingly inadequate for non-linear, multi-physics degradation processes [1, 6, 13, 16].

This paper presents a data-driven predictive asset management paradigm integrating Digital

Twins (DTs), IoT-based condition monitoring and physics-informed artificial intelligence (AI) to enable condition-aware maintenance prioritisation decisions across HV transmission systems

[2, 4]. The approach shifts asset supervision from static threshold-based alarms toward probabilistic Asset Health Index and Remaining Useful Life estimation, supporting early fault detection, maintenance prioritisation and optimized allocation of capital and operational resources [3, 10].

The technical architecture is structured into four coupled layers. At the physical layer, nonintrusive online sensing provides continuous multi-physics data streams, including dissolved gas analysis (DGA), vibration signatures, partial discharge indicators, thermal measurements, conductor sag and SF₆ density for transformers, overhead lines, circuit breakers, GIS [6, 7]. At the communication layer, deterministic and interoperable data exchange is achieved using IEC 61850-90-3, enabling vendor-agnostic integration of condition monitoring devices, while precise time alignment is ensured through IEEE 1588 Precision Time Protocol [8, 9]. At the virtual layer, high-fidelity DTs embed governing physical laws—such as heat transfer, electromagnetic behaviour and mechanical dynamics—using numerical modelling and PhysicsInformed Neural Networks (PINNs). By constraining machine-learning models with partial differential equations, PINN-based DTs maintain physical consistency under extrapolative operating regimes, addressing limitations of purely data-driven approaches [4, 5]. At the cognitive layer, AI-based prognostics and health management algorithms, including ensemble learning, Random Forests and Long Short-Term Memory (LSTM) networks, transform synchronized data streams into actionable intelligence as failure probabilities, health indices and prescriptive maintenance recommendations.

The framework’s effectiveness is demonstrated through field deployments and simulationbased case studies spanning multiple asset classes. Performance indicators are derived from peer-reviewed field deployments, pilot-scale utility implementations and simulation studies, as referenced in the cited literature. DT-based Dynamic Line Rating enables continuous recalculation of conductor ampacity using real-time meteorological and operational data, unlocking capacity gains of 5–35% under typical conditions and up to 55% during favourable night-time cooling without compromising thermal safety margins [13, 16]. For transformer fleets, hybrid cloud–edge DTs integrating online DGA and AI diagnostics achieve F1-scores exceeding 0.92 in early fault classification, enabling outage deferral and optimized maintenance planning [7, 10]. For switchgear and GIS, vibration-based DTs and LSTM-driven SF₆ leakage prediction enable early identification of mechanical degradation and environmentally critical gas losses [6, 8]. At system level, federated DTs support predictive assessment of frequency stability and inertia margins under high renewable penetration, enabling proactive operational decision-making [2, 15].

By demonstrating scalability from component-level diagnostics to system-level resilience, this paper establishes DT-enabled asset management as a foundational enabler for modern transmission networks.