APC Model-Plant Mismatch Index (APMMI)

GMI-PI-003Process IndustriesIntegration-RequiredPattern O-2: Model Drift / Currency

Definition

Average normalized mismatch between the manipulated-variable moves the APC's internal model predicts should be required and the moves actually required to hold the controlled-variable targets: reported alongside Control Authority Remaining (CAR) to show how much disturbance-rejection headroom the controller has consumed compensating for that mismatch.

The Measurement Gap

Closed-loop process control has a failure mode that no legacy controller metric detects: the model degrades, the controller compensates by pushing manipulated variables harder, the targets are still held, every dashboard says the controller is performing well, and the only thing changing is the room the controller has left to handle a disturbance. Controller uptime, controller service factor, and target-hit percentage all measure whether the APC is online and meeting setpoints: none of them measures whether the moves required to hit those setpoints are growing. Operator override rate is a downstream signal that arrives only after operators have lost confidence. Manipulated variable saturation alarms register the cliff edge, not the slope toward it. There is no continuous, normalized measurement of how much the actual control effort has diverged from the model-predicted control effort, and no measurement of how much move capacity remains before the controller saturates on the next upset.

Decision Supported:

Is the APC model still a representative description of the plant, or has it drifted enough that the controller is working harder than the model predicts, and how much control authority is left for handling a real disturbance?

Formula

Plain language: For each manipulated variable, compare the move the model says is needed against the move the controller actually makes, normalize by the variable's range, then average across all manipulated variables. CAR is the smallest remaining move room across all MVs.

Notation:

APMMI = ( 1 / M ) × Σ | MV_actual_m − MV_model_predicted_m | / MV_range_m

CAR_m = min( MV_max_m − MV_actual_m, MV_actual_m − MV_min_m ) / MV_range_m
CAR = min( CAR_m ) across all MVs

Variables:

Variable Definition Unit
M Number of manipulated variables in the APC application count
MV_actual_m Actual value of manipulated variable m needed to hold controlled-variable targets engineering unit
MV_model_predicted_m Value the APC model predicts should be needed for the same targets (steady-state gain matrix prediction) engineering unit
MV_range_m Operational range of manipulated variable m (MV_max_m − MV_min_m) engineering unit
MV_max_m, MV_min_m Hard or operating limits on manipulated variable m engineering unit

Proposed formula, pending validation pilots.

At a Glance

Unit
Index (0 to 1)
Trend Direction
Mixed; Decreasing (APMMI); CAR approaching 1.0
Reporting Frequency
Weekly during steady-state operation
Category
AI Process Optimization Integrity
Constraint Boundary Time-at-Risk (CBTR) →

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