Module 05  ·  AI Digital Twin

The Living Digital Twin

The continuously calibrated AI Digital Twin — a living digital representation of the refinery. Twin keeps every module synchronized with reality at all times through automated calibration, soft sensors, ML anomaly detection, and a natural-language interface backed by the process model.

05.1 · Overview

Refinium Twin

Refinium Twin is the foundational intelligence engine of the platform. It maintains a continuously updated, AI-calibrated digital representation of the complete refinery — keeping every module (Plan, Finance, Process, RTO, PC&S) synchronized with reality at all times. Without Twin, the other modules would gradually drift from the plant; with Twin, they never do.

05.2 · Why It Matters

Key Capabilities at a Glance

01
Continuous Calibration
AI-driven calibration runs continuously against live DCS, historian, and analyzer data. Reactor severity, fouling coefficients, column tray efficiency, and catalyst activity are all kept current — automatically.
02
Soft Sensors That Replace Hardware
Octane, cetane, sulfur content, catalyst activity, fouling factor, column flooding margin, blend quality — predicted in real time from the calibrated model with confidence intervals.
03
ML Anomaly Detection
Detects divergence between model-predicted and plant-measured behavior. Identifies root causes, quantifies margin impact, and recommends corrective action.
04
Natural Language Interface
A conversational AI assistant grounded in the process model. Ask plain-language questions, get model-backed answers with the underlying simulation as evidence.
05
Full Audit Trail & Governance
Every calibration, every model update, every parameter change is versioned, audited, and reversible. Role-based access and approval workflow built in.
06
Model Health Scoring
Calibration age, data quality indicators, parameter bounds, and drift alerts give operators a continuous view of the digital twin’s health.
05.3 · Technical Depth

Full Capability Set

Every capability included in Refinium Twin.

Data Reconciliation
Mass balance closureEnergy balance closureComponent balance closureGross error detectionSensor validationRedundancy analysisData quality flagsOutlier removalCross-instrument validationHistorical reconciliation
AI Calibration Engine
Continuous model calibrationReactor severity tuningFouling coefficient estimationColumn tray efficiencyCatalyst activity trackingML anomaly detectionModel health scoringCalibration age monitoringParameter bounds enforcementModel drift alerts
Soft Sensors & Predictive Analytics
Product octane predictionCetane soft sensorSulfur content predictionCatalyst activityFouling factorColumn flooding marginHeat recovery lossBlend quality predictionConfidence intervalsData quality indicators
Natural Language Interface & Governance
Conversational AI assistantPlain-language queriesModel-backed explanationsPredictive constraint alertsUpload guidanceFull audit trailApprove/revert workflowVersion controlRole-based accessDaily performance reports
05.4 · Use Cases

Refinium Twin in Action

Scenario
FCC gasoline yield has dropped — why?
Twin detects a divergence between model-predicted and actual FCC gasoline yield. The AI layer identifies catalyst activity decline as the root cause, quantifies the margin impact, and generates a recommended corrective action with full economic context from Finance.
Scenario
What is the actual octane of the gasoline pool right now?
Twin runs the soft-sensor stack against live plant data and returns the predicted octane with a confidence interval — replacing the need for continuous lab analysis.
Scenario
Ask the AI: "Why did our hydrogen demand spike last hour?"
The natural-language assistant traces the spike to the simulator, identifies the FCC feed shift as the cause, and presents the model-backed evidence — complete with the underlying mass balance.

See Refinium Twin on Your Refinery

Request a personalized demo. We will walk through Refinium Twin on your refinery's actual configuration.

Request Demo