What is a Digital Twin? Definition and Industrial Use Cases

What is a Digital Twin?

Consider a wind turbine operating in a remote location. The turbine has sensors monitoring blade vibration, bearing temperature, generator output, and electrical load. Every 10 seconds, this sensor data streams to a cloud system where a virtual model of the turbine—its geometry, its materials, its wear characteristics—runs in parallel.

That virtual model is the digital twin. It knows the design (from the engineering BOM), the current operational state (from sensors), and the as-manufactured configuration (which blade version, which bearing supplier, which generator firmware revision). When the twin's predicted bearing temperature deviates from the real bearing temperature, that is a signal: something is changing about the asset. When the twin predicts that the bearing will fail in 12 days based on the current degradation rate, maintenance is scheduled before the turbine goes dark.

The turbine owner avoids catastrophic failure, schedules maintenance at a convenient time, and orders parts with 12 days' lead time instead of emergency procurement. That is the value of a digital twin.

Digital Twin vs. CAD Model

The confusion starts here. Many engineers think a digital twin is just a nice 3D CAD model they can rotate on screen.

It is not.

A CAD model is static. It represents what the designer intended to build. It answers the question "what should this look like?"

A digital twin is dynamic. It is a living model synchronized with real-world sensor data, configuration history, and operational feedback. It answers the question "how is this really performing right now, and what will happen next?"

Example: An aircraft engine has a CAD model. The CAD model shows the geometry of the turbine blades, the materials specified, the tolerances. The digital twin of the same engine ingests:

• Real-time sensor data: RPM, fuel consumption, exhaust temperature, vibration signature, oil analysis results
• Operational history: flight hours, cycles, maintenance events, parts replaced
• Configuration: which blade revision is installed, which firmware version, which seals, which coatings

The CAD model is correct or incorrect when the engine is built. The digital twin is correct or incorrect in the moment, based on the most recent sensor reading. The value of the twin is that it can predict: "At the current degradation rate, this blade will fail in 500 more flight hours" or "This vibration signature indicates the bearing is spalling—schedule overhaul in the next maintenance window."

How Digital Twins Work

The architecture has three layers:

1. Physical Asset with embedded sensors

   • The wind turbine, aircraft engine, or pump in the real world
   • Sensors collect operational data: temperature, pressure, vibration, performance metrics
   • Data flows continuously to the edge or cloud

2. Digital Twin Model running in software

   • A virtual model of the asset that mirrors the physical one
   • Receives sensor data continuously and updates its internal state
   • Runs simulations: "If this degradation continues, when will failure occur?"
   • Learns from deviation: "The real temperature is 5°C higher than predicted; what changed?"

3. Decisions and Feedback Loop

   • Based on the twin's predictions, maintenance is scheduled, design changes are prioritized, or field updates are issued
   • Feedback flows back: "You predicted failure in 12 days; we did maintenance yesterday. Did we catch it in time?"
   • The twin learns and becomes more accurate over time

The strength of the loop is that every prediction the twin makes can be validated against ground truth. If the twin predicted a bearing would fail and you replaced it, you can inspect it and learn whether the prediction was correct. This feedback loop is what makes digital twins increasingly accurate over time.

What Data Feeds a Digital Twin?

A high-quality digital twin needs four streams of data:

1. Configuration Data (from PLM)

   • Which version of the design is this unit?
   • Which parts, which revisions, which software are installed?
   • This comes from the engineering BOM and configuration management system

2. Manufacturing Data (from MES)

   • How and when was this unit built?
   • Which operator? Which shift? Which tools? Which quality checks?
   • Did anything deviate from the design during manufacturing?

3. Operational Data (from Sensors and IoT)

   • What is the asset doing right now?
   • Temperature, pressure, vibration, fuel consumption, duty cycle?
   • These are real-time measurements from the physical asset

4. Service and Maintenance Data (from Service Systems)

   • What has broken? When was it repaired? What was replaced?
   • What design changes have been issued to this unit?
   • This feedback loop is critical for the twin to remain accurate

All four of these data streams should flow through the digital thread. If the thread is dirty (stale configuration, inconsistent operational records, missed maintenance events), the twin's predictions become less accurate.

Use Cases for Digital Twins

Predictive Maintenance

Instead of replacing parts on a calendar (every 5,000 hours) or reacting to failure, you predict when failure will occur and schedule maintenance at the exact right time. This reduces downtime, reduces waste (you don't replace parts that still have life left), and prevents catastrophic failures.

Example: A turbine bearing is monitored continuously. When vibration signature and temperature trends indicate spalling has started, maintenance is scheduled for the next available window—potentially weeks away—instead of waiting for catastrophic failure.

Performance Optimization

The digital twin allows you to compare predicted behavior (what the design said it should do) to actual behavior (what the sensors show it is doing). When there is a significant deviation, that is a signal to investigate.

Example: A wind turbine is producing 5% less power than the CAD model predicted. The digital twin identifies that this is correlated with blade surface roughness (erosion over time). This leads to a design change to protective coatings on new units.

Design Validation and Simulation

Instead of building expensive physical prototypes and field-testing them, new designs can be validated against the actual operational envelope of the current fleet's digital twins.

Example: An aircraft engine manufacturer wants to test a new compressor blade design. Instead of building a prototype and ground-testing it, they simulate the new blade design running on the digital twins of their existing engines under the actual flight envelopes those engines have experienced. If the simulation shows the new blade will improve efficiency without increasing wear, they can reduce the physical testing burden.

Fleet Management and Prognostics

When you have digital twins of hundreds of assets in the field, you can predict trends across the fleet and issue proactive field bulletins or recalls before failures occur at scale.

Example: A medical device manufacturer notices that digital twins of insulin pumps from a particular manufacturing batch are showing earlier-than-expected battery degradation. They issue a proactive software update to the fleet to reduce power consumption and extend battery life by 15%, preventing thousands of patients from experiencing device failure.

The Digital Twin Depends on the Digital Thread

This is critical: A digital twin is only as trustworthy as the data feeding it.

If the configuration data in the digital twin is stale (the system thinks version 3 firmware is installed but version 4 was actually installed in the field), the twin's predictions will be wrong. If the sensor data has gaps or is inconsistent, the twin's state diverges from reality. If the operational history is incomplete (maintenance events missing, parts replacements not recorded), the twin cannot learn accurately.

This is why the digital thread is the prerequisite for digital twins. The thread is the governed, consistent data backbone. The twin is the simulation running on top of it.

Why Digital Twins Matter More Now

Three things are converging:

1. Sensor costs are plummeting. IoT sensors that cost $1000 ten years ago cost $10 today. Streaming data from remote assets is now economically viable.

2. AI/ML makes prediction more accurate. Digital twins can now learn from huge fleets of operational data, making predictions more reliable than physics-based models alone.

3. Downtime is expensive and has externalities. An aircraft can't be delayed; a wind turbine offline costs $10,000/day; a production line stopped is a supply chain disaster. Predictive maintenance to avoid downtime has moved from nice-to-have to strategic imperative.

Result: Digital twins are moving from niche (aerospace, energy) to mainstream (automotive, industrial equipment, consumer appliances).

Next Steps

• To understand the data infrastructure that feeds digital twins, see What is a Digital Thread?
• To understand how digital twins fit into PLM architecture, see What is PLM?
• For a longer treatment of digital thread vs. digital twin, see Demystifying Digital Thread and Digital Twin