Digital Twin Fidelity Index (DTFI)

GMI-AR-019Asset ReliabilityInfrastructure-Dependent

Definition

Digital Twin Fidelity Index (DTFI) measures how closely a twin's predicted values match actual measured values across key performance parameters.

The Measurement Gap

A pharmaceutical manufacturer invested $2.4 million in digital twins for six critical reactor vessels. The twins modeled thermal stress, agitation wear, seal degradation, and vessel wall thinning. The value proposition: simulate reliability impact of operating decisions before executing them.

Eighteen months after commissioning, the production manager used Reactor 3's twin to evaluate running at 15% higher throughput for two weeks. The twin predicted a 4.2% increase in seal degradation with no impact on vessel wall integrity. The manager approved the increase.

Ten days in, the agitator shaft seal failed. Batch loss: $180,000 in materials plus 72 hours of downtime. The investigation found the twin's agitation model hadn't been updated to reflect a seal material change made during maintenance six months earlier. The new seal material had different thermal expansion properties. The twin was modeling a component that no longer existed.

Nobody measured whether the twin matched reality. It had been commissioned, validated once, and left to run. In 18 months, the reactor had undergone component changes, process parameter adjustments, and material changes the twin reflected none of.

OEE captured the batch loss. Asset Availability dropped. Neither could trace the failure to a decision based on an inaccurate twin.

There is no metric that measures the gap between a twin's predictions and actual asset behavior.

Formula

Digital Twin Fidelity Index (DTFI) measures how closely a twin's predicted values match actual measured values across key performance parameters.

In plain language:

How close is the twin to reality, and is the gap growing?

The Formula

DTFI = 1 - (1/P) x Sum_p [ (1/M_p) x Sum_m |T_pm - R_pm| / R_pm ]

Where P is the number of key performance parameters tracked, M_p is the number of matched measurement points for parameter p, T_pm is the twin's predicted value, and R_pm is the actual measured value.

Worked Example

A compressor twin tracks four parameters over 30 days:

| Parameter | Mean Twin Value | Mean Actual Value | Mean Abs % Error |
| Discharge temp (C) | 187.3 | 191.8 | 3.1% |
| Vibration (mm/s RMS) | 4.21 | 4.87 | 14.2% |
| Oil return temp (C) | 72.6 | 73.1 | 1.8% |
| Bearing wear rate (um/1000hr) | 0.34 | 0.52 | 38.7% |

Average error = 14.45%. DTFI = 1 - 0.1445 = 0.856

Below the 0.90 threshold. Temperature modeling is good (less than 5% error). Bearing wear rate, the critical parameter, is off by 39%. The twin is usable for thermal decisions but unreliable for wear predictions.

Sign Convention

DTFI of 1.0 is perfect. A 0.90 threshold is proposed for operational reliance. Below 0.80, twin outputs are misleading. These thresholds are proposed, not proven.

Cross-References

Downstream: Degradation Projection Validity Rate suffers when DTFI is low. An inaccurate twin produces unreliable projections. Health Score Contextual Accuracy is affected when health scores derive from twin simulations. Strategy Override Safety Index is affected when strategy recommendations use twin-simulated coverage.

DTFI has no upstream dependencies. It depends on engineering model quality and configuration management.

The dangerous scenario: high DTFI for easy parameters (temperature, pressure) and low DTFI for critical ones (wear rates, fatigue). Per-parameter reporting is essential.

Proposed formula, pending validation pilots.

At a Glance

Category
AI Data and Source Trustworthiness

Decision Supported

Can I trust this twin's simulation output for operational decisions, or has the twin diverged from reality to the point where its recommendations are unreliable?

Failure Code Fidelity Rate (FCFR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment