Real-time synchronization: physical to digital twin guide

An IoT monitoring dashboard displays current and historical sensor readings — a visualisation layer on top of sensor data. A digital twin is a behavioural model of a physical asset that is continuously calibrated against sensor data — it represents not just current measurements but the inferred state of the asset, enabling simulation of future states and "what-if" scenario testing that raw sensor dashboards cannot support. The distinction is between observation (dashboards) and understanding (twins). In practice, the boundary is blurry — many "digital twins" are primarily sophisticated monitoring dashboards, and some "monitoring platforms" include model-based analytics. The defining characteristic of a genuine twin is the presence of a computational model that simulates asset behaviour, not just records it.

Resilient twin synchronisation requires buffering at the edge — local storage that accumulates sensor readings during connectivity interruptions and replays them when connectivity is restored. MQTT's QoS level 1 (at-least-once) and level 2 (exactly-once) delivery guarantees handle short interruptions. For extended outages, edge compute nodes should maintain a local twin state that continues to operate autonomously, with the cloud twin reconciling the buffered event stream when connectivity restores. The reconciliation strategy (time-ordered replay vs. last-write-wins for conflicting updates) must be defined based on the specific consistency requirements of each twin property.

OPC Unified Architecture (OPC UA) is an industrial communication protocol standard providing a platform-independent, service-oriented architecture for industrial data exchange. It defines both a communication protocol and a semantic data modelling framework — assets expose their data through standardised address spaces with defined data types, access control, and subscription mechanisms. OPC UA is the dominant standard for industrial equipment data access because it provides vendor-independent access to PLCs, CNC machines, robots, and SCADA systems from any OEM — a critical requirement for enterprise digital twins that must integrate data from heterogeneous equipment across a production facility. The OPC Foundation's companion specifications extend OPC UA for specific industries (Process, Machinery, Robotics) with standardised information models that accelerate integration across vendor boundaries.

Storage requirements depend heavily on the number of monitored data points and retention requirements. A representative industrial digital twin monitoring 10,000 data points at 1-second granularity generates approximately 2–5 TB of raw time-series data per year. Compression and aggregation reduce this significantly: time-series databases like InfluxDB and TimescaleDB achieve 10–50× compression on typical sensor data, and tiered storage (high-resolution recent data, aggregated historical data) further reduces costs. Define retention and resolution tiers early: 30 days at full resolution for operational use, 1 year at 1-minute aggregation for trend analysis, 5+ years at 1-hour aggregation for long-term planning — typical tiered retention achieves 95%+ storage cost reduction versus storing all data at full resolution indefinitely.

Yes — combining real-time synchronisation with simulation is the highest-value digital twin capability. Real-time data continuously calibrates the simulation model, keeping it aligned with actual asset condition rather than degrading against the original design specification. This enables "predictive digital twins" that simulate future asset states based on current conditions: a gearbox showing early-stage vibration anomalies can be simulated through its projected degradation trajectory, predicting time-to-failure with quantified uncertainty. The architecture requires separating the state twin (real-time synchronised current state) from the simulation twin (physics or ML model that accepts current state as initial conditions) — both components backed by the same synchronisation pipeline but serving different purposes.

Digital twin synchronisation pipelines create bidirectional connections between IT (cloud platform) and OT (operational technology — PLCs, sensors, industrial networks) that are high-priority attack targets. Key security requirements: OT network segmentation preventing direct internet-accessible exposure of industrial systems, with twin synchronisation flowing through DMZ-resident edge gateways; TLS encryption for all telemetry transmission; certificate-based device authentication rather than shared secrets; and principle of least privilege for all pipeline service accounts. The twin platform itself should be treated as critical infrastructure — access to a twin that can both read asset state and send commands back to assets provides significant attack surface. Read-only twins that cannot command physical assets represent meaningfully lower risk than bidirectional twins that support remote operation.

Accurate timestamps are foundational to digital twin quality — event ordering, latency measurement, and correlation across multiple data sources all depend on timestamp accuracy. Industrial networks should implement IEEE 1588 Precision Time Protocol (PTP) or GPS-disciplined NTP to achieve sub-millisecond clock synchronisation across all data sources. Cloud platforms typically use NTP with ±1ms accuracy — adequate for most analytics use cases but insufficient for control-loop applications requiring microsecond event ordering. Timestamp inaccuracy manifests as event ordering anomalies, spurious correlations between physically unrelated events, and incorrectly calculated process durations — often appearing as mysterious data quality issues that are difficult to diagnose without explicit clock synchronisation monitoring.

Graph-based digital twins model assets and their relationships as nodes and edges in a knowledge graph, enabling queries like "find all pumps whose upstream valve is in a degraded state" or "trace all downstream systems affected by this sensor failure." Azure Digital Twins and DTDL use a graph model as the core data structure. Graph-based twins are most valuable for complex interconnected systems — building infrastructure (HVAC dependencies, power distribution topologies), supply chain networks, and industrial process systems where understanding relationship-dependent impacts is as important as individual asset state. For simpler asset monitoring use cases without complex interdependencies, time-series databases with simpler data models are more efficient and easier to operate.