Review. Optimise. Perform.
Real-time intelligence for centralised chilled-water plants.
Measure the plant. Model the operating space. Admit the best safe state.
Chillers, chilled-water pumps, condenser-water pumps, cooling towers, building load, weather and controls are one coupled system. vChill optimises the whole plant, not isolated equipment.
A handsome chiller kW/RT with the towers wide open and the CHW pumps on the wrong side of the curve is not optimisation. It is component vanity.
The physics only pays whole-plant.
SIMULATED DEMO PLANT A
Know the plant before you touch it.
Connect → Measure → Understand
PLC, BMS, SCADA, meters, sensors, historian.
High-frequency physical telemetry, with quality on every point.
Plant state, whole-plant kW/RT, load, anomalies, diagnostics.
Measurement is not passive. It is the beginning of control. An uninstrumented plant and a measured plant are not operationally equivalent. Once operators or algorithms act on measured performance, instrumentation has entered the feedback loop. Shadow Mode lets vChill evaluate what it would do before it is permitted to write.
Missing is a state of knowledge. It is not a value.
| EQ | STATE | POWER |
|---|---|---|
| CT-1 | RUN | 22.0 kW |
| CT-2 | RUN | 21.8 kW |
| CT-3 | OFF | 0.0 kW |
| CT-4 | OFFLINE | — |
The same surface the operator uses. Simulated Demo Plant A. CT-4 is offline — power is —, not 0.0 kW.
Search the operating space.
Predict → Optimise → Control
Expected whole-plant response under alternative states.
Search the constrained operating space. Select a whole-plant state.
Supervisory writes only where authorised, through the Safety Gate.
vChill combines deterministic engineering, a digital twin, statistical models and a constrained optimiser. Optional advanced learners are plug-ins. The optimiser does not bypass engineering constraints. Autonomous control is never the default.
OBSERVE → LEARN → SHADOW → ADVISE → SUPERVISED → AUTONOMOUS
Multiple candidate control actions can be evaluated against live telemetry while the plant remains under native control. A shadow strategy proposes a state, records predicted versus actual kW, and does not write. In Shadow Mode, wrote = false.
Simulated traces for illustration. Shadow evaluations in the product set wrote = false.
Sensor quality, equipment limits, service requirements, authorisation, operating envelope, communications health, and fallback. vChill is a supervisory layer. It never replaces OEM protections, safety PLCs, or hardwired interlocks. Return to native BMS/PLC control is always explicit.
Measure what resulted.
Quantify → Evidence
Actual versus expected energy use, with uncertainty.
Traceable record of what changed and what resulted.
vChill quantifies operational performance in engineering terms and translates it into energy and cost consequences. It does not operate conservation registries or issue certificates.
Source telemetry, model version, approval, control action, measured response, baseline, uncertainty, and a SHA-256 evidence package. Tamper-evident application provenance — not a blockchain.
Simulated Demo Plant A. Condenser heat rejection is approximately evaporator load plus compressor work; auxiliaries treated separately.
Simulated baseline vs actual with an intervention after 14:20. Uncertainty shown as a band, not false precision.
The product is the intelligence beneath it: plant connectivity, engineering models, optimisation, control logic, diagnostics, and evidence.
Connect to plant systems. Survive WAN loss. Enforce local safety.
Calculate, diagnose, twin, optimise. Versioned, replaceable strategies.
Historian, fleet permissions, evidence, licensing.
HMI, alarms, decisions, supervisory control.
vChill has evolved through chilled-water optimisation work and live plant data, not laboratory-only simulation. Historical project data is available inside the platform for replay and analysis. Public client logos and savings percentages are not shown here.
A centralised chilled-water plant is a coupled thermodynamic system and a control search. Cooling is delivered as Q = ṁ cp ΔT. That load, together with compressor work, is ultimately rejected through the condenser loop toward the wet-bulb sink. Staging, CHWS, condenser-water temperature, flow, differential pressure and tower speed create a vast constrained operating space too large for fixed sequencing to explore intelligently.
At any instant the plant occupies one operating point. That is occupancy, not knowledge. Unmeasured coordinates remain uncertain. Candidate writes remain many. ∞ → 1 describes the collapse of the control search, not a claim that the plant was ever physically in several states at once.
Most control wanders. A handsome chiller kW/RT with the towers wide open and the CHW pumps on the wrong side of the curve is not optimisation. It is component vanity. The physics only pays whole-plant.
Call the unmeasured machine Schrödinger’s chiller. OFFLINE is not OFF. Writing 0 kW into an empty measurement is not measurement. It is a fiction that makes the spreadsheet consistent and the optimiser dangerous. The box stays closed until a sensor opens it. Missing is a state of knowledge. It is not a value.
Measurement is not passive. It is the beginning of control. Once operators or algorithms respond to measured performance, instrumentation has entered the feedback loop. Shadow Mode lets vChill calculate candidate actions before it is permitted to write.
vChill is informationally negentropic. Measurement reduces uncertainty. The plant model removes implausible states. Constraints remove inadmissible states. The Safety Gate removes unsafe states. The second law remains untouched. We do not defeat entropy. We stop adding operational disorder.
Infinity is the search. Measurement, model, constraints and authority reduce the candidate set toward one admitted control action: minimise whole-plant energy while maintaining service.
That is the singularity.
We use the language of physics because the plant is already doing the physics.
x(t) measured plant state
u candidate control action
J(u) total plant energy objective
g(x,u) operating + safety constraints
u* admitted optimum
min_u J(u)
s.t. Q delivered
T_CHWS within service band
quality != missing
SafetyGate(x,u) = pass
authorisation = true
∞ candidates → one admissible u*Review. Optimise. Perform.
Review the plant. Optimise the state. Prove the performance.