Digital twin
A live model of your assets, lines and operating conditions, so AI agents work from the plant as it is.
Manufacturing Copilot
Physical AI for plant operations
A governed Physical AI system for manufacturing and plant operations that improves performance, reliability and asset utilisation, and reduces downtime, through bounded supervisory autonomy.
01/What it does
Four parts, one system
Plant ontology
Defines asset identities, units, variables, data sources and relationships, so AI agents get machine-specific context and the definitions stay protected.
AI agents
Find faults and process inefficiencies, then propose changes to setpoints, parameters or production conditions within your operating constraints.
Unified knowledge layer
Operators ask in plain language and see the source behind every answer.
02/Bounded autonomy
Agents propose. The control plane decides.
The Data Nucleus control plane is the independent layer that decides what each agent may do.
Lower consequence
Within authority
Proceeds automatically.
Higher consequence
Refer
Goes to an authorised person, with supporting evidence.
Never crossed
Hard limits
Machine-specific operating constraints that no agent and no approval can override.
- A signed, versioned Agent Risk Passport defines what each agent is approved to do.
- Runtime assessment weighs consequence, likelihood, uncertainty and current operating context.
- Agents cannot change governance, protected ontology definitions or their own permissions.
- Any expansion of authority needs risk assessment and human approval.
03/Intervention evaluation
Tested before it touches the line
Candidate interventions are evaluated on controlled plant-state snapshots with approved predictive models, including ML surrogate models linked to the digital twin.
- 01
Snapshot
Controlled plant-state snapshots.
- 02
Evaluate
Expected performance, reliability, uncertainty, model applicability and operating limits.
- 03
Coordinate
Shared resources, interventions already authorised and cumulative effects.
04/Safe execution
Controlled from proposal to the machine
Controlled execution
Changes reach the machine through an independently governed path; an agent holds no write access.
Bounded authority
Authorisation covers the target, parameters, conditions, sequence and validity period of the whole intervention.
Verified before execution
Asset identity, engineering units, control ownership and machine mode are checked immediately before execution.
Local control
Local controllers keep real-time control, and protective functions and interlocks stay independent of agent optimisation.
Runtime monitoring
Predicted and observed behaviour are compared, with a process-specific fallback if they diverge.
Everything recorded
Every intervention, approval, risk assessment, model version and outcome is recorded.
05/Plant systems
Connected in ISA-95 order
- IoT edgeMQTT
- PLC
- SCADA / HMIOPC UA
- Historian
- MES
- MRP
06/Pilot cohort
Join the pilot cohort
Closed-loop sensing, assessment, authorisation and physical intervention, tested and optimised on real manufacturing machines in sandboxed pilots. A simulated factory environment supports development, validation and scaling.
You get
- A fully operational unified knowledge layer
- Deployment on one line or cell with your own data and documents
- Direct access to the engineering/forward deployment team
- A say in the roadmap
You bring
- One workflow with a measurable baseline
- Representative data and approved interfaces
- A named engineering sponsor
- Feedback on a fixed cadence