The concept of a digital twin is deceptively simple: create a virtual model of a physical system that is kept in sync with the physical counterpart through continuous data exchange. In practice, building a useful digital twin is one of the more technically demanding challenges in industrial engineering.

The term was coined by Michael Grieves at the University of Michigan in 2002, in the context of product lifecycle management. For two decades it remained largely theoretical — the sensor technology, connectivity, and compute required to keep a complex physical system’s virtual model updated in real time simply did not exist at practical cost.

That changed around 2020. And in 2026, digital twins are transitioning from competitive advantage to operational standard in advanced manufacturing.

What Makes a Digital Twin Different from a Simulation

This distinction matters, and it is commonly confused.

A simulation model is a mathematical representation of a system built before the physical system exists. It uses nominal parameters — design specifications, material properties from datasheets — and produces outputs that represent predicted behaviour. You build the simulation to inform design decisions.

A digital twin is continuously updated with data from the physical system. Its parameters are not nominal design values; they are inferred from measurements of the actual system. The digital twin represents current behaviour, including manufacturing variation, wear, and operational history.

The practical implication: a digital twin can tell you how your specific machine — the one on the factory floor, with its particular bearing wear pattern and its particular motor winding temperature characteristic — will behave under a new operating condition. A simulation cannot.

The Architecture of an Industrial Digital Twin

A production-quality industrial digital twin has five components:

1. Physical System Instrumentation

The physical system must be measured. This means sensors — temperature, vibration, current, pressure, position, flow — distributed throughout the system and sampled at rates appropriate to the dynamics of interest.

For a BLDC motor drive system, the instrumentation might include:

  • 3-phase current sensors (100 kHz sampling for current ripple analysis)
  • Bearing vibration accelerometers (10 kHz sampling)
  • Motor winding thermistors
  • Shaft position encoder (high-resolution, for torque ripple analysis)
  • DC bus voltage sensor

The instrumentation strategy is constrained by cost and by the diagnostic value of the measurements. A systematic measurement value analysis (MVA) should precede instrumentation design.

2. Data Transport and Edge Processing

Raw sensor data from a single motor drive can exceed 1 GB/hour at full sampling rates. Transmitting this volume to a cloud backend continuously is neither practical nor necessary.

Edge processing performs three functions:

  • Compression: FFT of vibration signals, statistical moments of current signals, rather than raw time-series
  • Anomaly detection: Real-time comparison against baseline models, flagging deviations for transmission
  • Buffering: Local storage for high-resolution data capture during flagged events

OPC UA is the standard protocol for industrial edge-to-cloud communication, providing semantic data models that allow different equipment from different vendors to be described consistently.

3. The Model

The core of the digital twin is a mathematical model that predicts system behaviour from inputs. For a motor drive system, this typically includes:

Physics-based model: The motor’s electromagnetic equations, thermal model, and mechanical model. Parameters identified from measurement: winding resistance at current temperature, magnetisation curve, rotor inertia.

Data-driven model overlaid on physics: A machine learning model trained on historical operating data that captures nonlinearities and degradation effects that the physics model does not represent — bearing friction increase with wear, insulation resistance degradation with thermal cycling.

The hybrid physics + data model is standard practice in 2026. Pure physics models have unknown error in parameter identification. Pure data models require enormous amounts of training data and generalise poorly. The hybrid captures the strengths of both.

4. State Estimation and Synchronisation

Keeping the digital twin’s state in sync with the physical system requires continuous state estimation. The Kalman filter and its nonlinear variants (Extended Kalman Filter, Unscented Kalman Filter, Particle Filter) are the standard tools.

For a motor thermal digital twin, the state vector might include:

  • Winding temperature (directly measured)
  • Core temperature (estimated from model + ambient temperature measurement)
  • Bearing temperature (estimated from power loss model + bearing friction model)
  • Remaining insulation life (accumulated thermal stress integral)

The Kalman filter fuses the model predictions with the actual measurements, continuously correcting model states to match reality.

5. Applications: What the Digital Twin Enables

Remaining Useful Life (RUL) Prediction: The digital twin tracks accumulated damage (thermal cycles, mechanical fatigue, partial discharge events in insulation) against known failure thresholds. This is far more accurate than time-based maintenance schedules.

Virtual Commissioning: Before a new machine or production line is physically assembled, the control software is tested against the digital twin. Bugs in the PLC code, controller tuning issues, and integration problems are found in simulation rather than on the factory floor. Siemens reports 50–70% reduction in physical commissioning time for customers using virtual commissioning.

Process Optimisation: Operators and engineers use the digital twin to safely explore the operating parameter space. “What happens if I increase conveyor speed by 15%?” The digital twin answers in seconds, without risk to the physical system.

Operator Training: New operators can be trained on the digital twin without any risk to production equipment. Procedure errors, emergency responses, and abnormal condition handling can all be practiced in simulation.

Digital Twins at Entlar: Our Roadmap

We are building digital twin capability for Entlar fans deployed in commercial installations. Our target architecture:

Phase 1 (in development): A thermal digital twin running in our cloud backend. Inputs: BLE-reported motor current (sampled at 10 Hz via fan control board) and ambient temperature. Model: a simplified R-C thermal network identified from controlled bench tests. Output: estimated winding temperature and bearing temperature, updated every 30 seconds.

Phase 2: Vibration monitoring added via a low-cost MEMS accelerometer integrated into the fan housing. Edge processing on the fan control board computes vibration spectrum. Digital twin detects bearing defect frequency emergence.

Phase 3: Full predictive maintenance dashboard for facility managers. Every Entlar fan in a hotel or office building visible in one view, with health scores, predicted service intervals, and anomaly alerts.

The technology is not novel — it is well-established in industrial applications. The challenge is implementing it cost-effectively at the scale of a consumer/commercial ceiling fan. The hardware constraints (a Cortex-M4 with 512 KB flash and 128 KB RAM) force creative decisions about what models can run on-device versus in the cloud.

Conclusion

Digital twins are not hype. They are a genuine, practical technology with measurable ROI in industrial applications. The costs of implementation have fallen dramatically, and the benefits — reduced unplanned downtime, shorter commissioning cycles, safer process exploration — are quantifiable.

For engineers, understanding how to build, calibrate, and use digital twin models is becoming a core competency. The tools are mature. The methods are established. The value is real.

The physical world is gaining a virtual shadow. Engineering is going to change significantly as a result.