Twin Calibration Drift Index (TCDI)

GMI-PI-010Process IndustriesIntegration-RequiredPattern O-3: Twin Fidelity

Definition

Mean absolute deviation between a digital twin's predictions and matched plant measurements across key process variables, normalized by each variable's operating range: detecting whether the twin is modeling the current plant or a historical version that no longer exists.

The Measurement Gap

Digital twins are used to evaluate process change decisions, optimization scenarios, and capacity assessments. The twin reports an answer in engineering units; the recipient assumes the twin reflects the plant. The failure mode is silent divergence: fouling accumulates, catalyst deactivates, mechanical wear changes hydraulic resistance, and the twin (calibrated to last quarter's plant) answers questions about a plant that no longer exists. Model deployment date, twin calibration date, and even average prediction accuracy on individual variables fail to register this gap because they do not normalize, do not aggregate, and do not match comparison conditions. The twin can be ninety-eight percent accurate on yield prediction at last quarter's operating point and twenty percent wrong on the current operating point, and no existing metric catches the difference. Without a continuous, condition-matched, normalized comparison across the variables that matter, every twin-based decision carries unquantified error.

Decision Supported:

Can the twin's "what if" analyses be trusted for process change decisions, or has the twin drifted far enough from reality that its recommendations are based on a plant that no longer exists?

Formula

Plain language: Match the twin and the plant at steady-state operating conditions, compare predicted to actual on each key variable, normalize each deviation by that variable's operating range, average across variables, multiply by one hundred.

Notation:

TCDI = ( 1 / V ) × Σ | Predicted_v − Actual_v | / Range_v × 100

Variables:

Variable Definition Unit
V Number of key process variables in the comparison count
Predicted_v Twin's predicted value for variable v under current operating conditions engineering unit
Actual_v Measured plant value for variable v at matched conditions engineering unit
Range_v Normal operating range of variable v (max − min) engineering unit
Matched conditions Steady-state operation with explicit equivalence criteria across feed quality, ambient conditions, and operating mode n/a

Report TCDI per unit operation and per variable to identify where drift is concentrated.

Proposed formula, pending validation pilots.

At a Glance

Unit
Percentage
Trend Direction
Lower; Decreasing toward zero
Reporting Frequency
Monthly, per unit operation
Category
Process Twin and Simulation Reliability
← Mechanism Attribution Accuracy (MAA)Blend Give-Away Distribution Index (BGDI) →

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