Test the move. Measure the improvement.

DRIVE treats the physics twin as an action gate. The proposal must make physical sense in the current operating state and remain inside the configured process, asset, controller, and policy boundaries.

• Current-state confidence. Freshness, completeness, units, plausibility, operating mode, and source quality determine whether the current state is trustworthy enough to evaluate.
• Candidate counterfactual. The twin estimates how the process and asset should respond to a bounded change, including the intended outcome, assumptions, and relevant uncertainty.
• Configured operating envelope. Process limits, OEM constraints, motor and VFD limits, controller state, minimum/maximum demand, rates of change, and customer policy define the eligible region.
• Reversibility and rollback. An approved action class needs a known reversal or safe step-down path. Unexpected controller or physical response freezes the sequence and records the exception.
• Explainable refusal. Bad data, an invalid model state, a limit breach, missing approval, an unsafe operating state, or a controller rejection produces a reason—not a silent retry.
• Controller authority. DRIVE remains supervisory. The PLC, DCS, VFD protections, safety PLC, and SIS keep their configured roles and can reject or override a proposal.

Authority is promoted with evidence, not elapsed time. Every step has a named owner, eligible operating states, action range, validation threshold, and revocation rule.

• Observe. Read supported signals, establish data quality, characterize operating states, and build a baseline without proposing or writing a change.
• Advise. Show a bounded recommendation, expected outcome, assumptions, constraints, and refusal reasons for an operator or engineer to evaluate.
• Shadow. Generate time-aligned proposals without sending them to the controller. Compare predicted outcomes with what the physical system actually did.
• Supervised. Allow an authorized operator to approve a certified action class within its configured range, time window, operating state, and controller path.
• Bounded. Permit only the separately approved action classes and operating envelopes that have passed the customer’s evidence and governance gates.
• Freeze, revoke, or step down. Failed evidence, stale data, unexpected response, limit breach, policy change, model invalidation, or controller rejection returns the action path to a safer state.

A DRIVE outcome should be inspectable by an operator, controls engineer, energy manager, or auditor. Modeled projections remain labeled until the physical result is measured and normalized.

• Baseline and context. Record the comparison window, demand, production or batch context, operating mode, data quality, and normalization factors.
• Proposal and prediction. Record the candidate action, allowed range, twin version, assumptions, expected physical response, intended KPI change, and confidence.
• Guardrail decision. Record every quality, physics, limit, action-class, time-window, approval, and controller-state check with its pass, refusal, or exception reason.
• Approval and command receipt. Attribute the human or policy approval, signed command scope, idempotency or replay checks, controller response, and effective setpoint.
• Measured outcome. Compare the real response with the prediction using the same process and asset context. Preserve deviations, alarms, overrides, and rollbacks.
• Normalized value. Express energy as useful work where possible and adjust for demand, production, weather, batch, or another approved driver before calling a change an improvement.

TwinEdge connects your data, assets, and workflows in a complete package. Compare subscriptions on the pricing page.

• 1. Fit and signal review. Review the asset, VFD/PLC, available signals, sample rate, process objective, operating constraints, baseline KPI, controller interface, and change-management requirements.
• 2. Read-only baseline. Validate signal meaning and quality, calculate physics-aware KPIs, characterize operating states, and agree on the baseline and normalization method.
• 3. Shadow validation. Run candidates without live action, compare predictions with observed system behavior, challenge refusals, and tune the approved envelope.
• 4. Supervised action class. If approved, enable a narrow reversible action through the existing control path with explicit human approval, stop conditions, and rollback.
• 5. Evidence review. Review prediction accuracy, controller acceptance, physical outcome, normalized value, exceptions, and operator feedback before any promotion.
• 6. Bounded expansion. Add authority, action classes, operating states, assets, or sites only when each addition passes its own engineering and governance gate.

The exact set is deployment-specific. DRIVE begins with a signal and control contract rather than assuming every driver, tag, or asset has the same semantics.

• VFD speed
• Frequency
• Current
• Power
• Torque or load
• Drive status
• PLC/DCS mode
• Flow
• Pressure
• Temperature
• Process demand
• Production context
• Operating envelope
• OEM limits
• Interlocks
• Approval policy
• Command receipt
• Outcome KPI