PRODUCTS
Five capability modules, licensed and deployed as one system
Each module runs standalone or as part of the connected loop. NOVYRON ships the software: licences, deployment, integration and lifecycle services. Robots, cameras and PLCs remain your hardware or your vendor's — the platform stays vendor-neutral.
- SOFTWARE LICENCE
- EDGE / CLOUD DEPLOYMENT
- INTEGRATION SERVICES
- LIFECYCLE SUPPORT
01 · MODEL
Simulation & Digital Twin
The problem. Physical trial-and-error on a running line is slow, expensive and risky. Layout changes, new products and automation decisions get made on intuition because testing them for real would stop production.
WORKFLOW
- S1Model the process, assets and constraints from layouts, cycle data and PLC signals
- S2Run baseline and candidate scenarios against recorded or synthetic demand
- S3Compare throughput, utilisation and bottlenecks side by side with stated assumptions
- S4Commission the chosen change virtually, then hand validated parameters to the floor
COMMON USE CASES
- CELL & LINE DESIGN
- VIRTUAL COMMISSIONING
- SCENARIO COMPARISON
- SYNTHETIC DATA
- TRAINING ENVIRONMENTS
- BASELINE · current layout142 u/h
- SCENARIO B · added buffer + re-route178 u/h
- SCENARIO C · second gripper171 u/h
ASSUMPTIONS
demand profile W34 · changeover 4.5 min · 2 operators · run 10×8h
Representative data, not a customer system. The assumptions line is the point: a scenario comparison is only usable if you can check what it assumed.
- INPUTS
- CAD / layout files, cycle and takt data, PLC signal logs, demand profiles
- OUTPUTS
- Scenario comparisons, bottleneck analysis, validated parameters, synthetic datasets
- INTEGRATES
- CAD/PLM exports, historian databases, OPC UA, the Perception and Guidance modules
02 · PERCEIVE
AI Perception
The problem. Manual inspection and monitoring miss defects, drift with fatigue and produce no structured record. Camera and sensor data exists but never becomes evidence anyone can act on.
WORKFLOW
- S1Connect camera streams and sensors; validate coverage, lighting and data quality
- S2Train or configure detection models — with synthetic data from the twin where labelled samples are scarce
- S3Run inspection, counting and anomaly detection live; every detection carries source frame, timestamp, confidence and model version
- S4Route low-confidence detections to human review queues; feed verdicts back into evaluation
COMMON USE CASES
- VISUAL QUALITY INSPECTION
- ANOMALY DETECTION
- COUNTING & TRACKING
- SAFETY MONITORING
- CLASS
- seal_gap (defect)
- CONFIDENCE
- 0.81 · threshold 0.85 → queued
- MODEL
- inspect-seal v3.2.1
- REVIEW
- awaiting QA verdict
Representative data, not a customer system. Confidence 0.81 sits below the 0.85 threshold, so the detection queues for a human verdict instead of being auto-rejected.
- INPUTS
- Camera streams (GigE/RTSP), industrial sensors, part specifications, labelled or synthetic samples
- OUTPUTS
- Detections with evidence, review queues, quality reports, structured event streams
- INTEGRATES
- Vision hardware vendors, MES/QMS, the Guidance module's approval gates, Events & Alerts
03 · DECIDE
Guidance & Decision Support
The problem. AI output without governance is either ignored or blindly trusted. Operations teams need recommendations they can interrogate — and a record of who approved what, when and why.
WORKFLOW
- S1Define rules, thresholds and approval gates per event type and role
- S2Receive events from Perception, the twin or external systems; generate recommendations with confidence and evidence attached
- S3A permitted human approves, modifies or rejects — consequential actions never auto-execute past a gate
- S4Every decision lands in the audit trail with actor, evidence snapshot and downstream effect
COMMON USE CASES
- APPROVAL WORKFLOWS
- ESCALATION ROUTING
- DECISION AUDIT
- SHIFT GUIDANCE
Divert lane 2 output to rework buffer
- trigger
- D-88412 seal_gap · conf 0.81 · cam-07 22:14:03 GST
- impact
- 14 units re-routed · est. 6 min · reversible
role required: line supervisor · the approving account, the action and its evidence are all recorded
Representative data, not a customer system. The recommendation cannot execute on its own — it waits for a named role, and the approval is recorded with its evidence.
- INPUTS
- Perception events, twin scenarios, external system events, rules and role policies
- OUTPUTS
- Governed decisions, approval records, escalations, full decision history
- INTEGRATES
- Identity providers (SSO/RBAC), messaging/ticketing, the Orchestration module
04 · ORCHESTRATE
Robotics Orchestration
The problem. Robots, AGVs and software services each speak their vendor's language. Coordinating them into one workflow usually means brittle point-to-point integrations that fail silently.
WORKFLOW
- S1Connect robots, systems and services through vendor-neutral adapters
- S2Compose workflows from tasks, states and handover rules — approved changes only
- S3Execute with live state per participant; exceptions pause into defined safe states
- S4Operators resolve exceptions with manual controls; every intervention is logged
COMMON USE CASES
- MIXED-FLEET COORDINATION
- LINE ↔ WMS HANDOVERS
- EXCEPTION MANAGEMENT
- TASK DISPATCH
- picking · t-3081arm-A2 · kukapicking · t-3081
- en route · dock 3amr-07 · miren route · dock 3
- degraded · retrying 2/3label-svc · apidegraded · retrying 2/3
- paused · awaiting upstreamwrap-cell · abbpaused · awaiting upstream
EXCEPTION RULE
if label-svc fails 3× → hold dispatch, notify shift lead, safe-state wrap-cell
Representative data, not a customer system. One service is degraded; the exception rule below is what stops that from quietly stalling the line.
- INPUTS
- Robot/AGV APIs, PLC and MES signals, WMS/ERP tasks, approved Guidance decisions
- OUTPUTS
- Coordinated task execution, live workflow state, exception logs, intervention records
- INTEGRATES
- KUKA, ABB, UR, MiR and other vendor APIs · OPC UA, MQTT, REST · WMS/MES/ERP
05 · OPTIMISE
Edge & Cloud Operations
The problem. AI systems degrade quietly: models drift, edge devices fall offline, versions diverge across sites. Without operational tooling, yesterday's working deployment is next quarter's incident.
WORKFLOW
- S1Package modules and models into signed, versioned deployment units
- S2Deploy to edge and cloud targets with staged rollout and health checks
- S3Monitor telemetry, model performance and data freshness; degraded states surface as first-class alerts
- S4Roll back in one governed step; every version change lands in the audit trail
COMMON USE CASES
- FLEET-WIDE ROLLOUTS
- MODEL LIFECYCLE
- SITE TELEMETRY
- AUDIT & COMPLIANCE
| TARGET | VERSION | HEALTH |
|---|---|---|
| edge · cell-04v2.4.1 | v2.4.1 | healthy |
| edge · line-2v2.4.1 | v2.4.1 | healthy |
| edge · dock-gwv2.3.9 ← v2.4.1 | v2.3.9 ← v2.4.1 | rolled backaudit #4471 |
| cloud · analyticsv2.4.1 | v2.4.1 | healthy |
Representative data, not a customer system. One target failed and was rolled back in a single step — the audit reference beside it is what this module exists for.
- INPUTS
- Deployment units, target inventory, telemetry streams, model registries
- OUTPUTS
- Staged rollouts, health dashboards, rollback actions, complete audit history
- INTEGRATES
- Kubernetes/edge runtimes, observability stacks, SIEM, enterprise identity
APPLIED SYSTEMS
Complete systems composed from the modules
A module is a capability. An applied system is those capabilities configured for one industrial process — its operating states, its recipes, its instruments and its acceptance tests. The two below are packaged and field-adaptable for liquid process plants, where the automation already exists but the operating picture is spread across separate screens.
Demineralisation Automation & AI
One operating layer over ion-exchange and hybrid RO–demineralisation trains: live process view, regeneration recipes, cycle analytics and an auditable event record.
THE PROBLEM
Water quality gets read from a single conductivity number, while resin cycle behaviour, raw-water composition, regeneration recipe, rinse quality, chemical consumption and RO performance all move together. When those live on separate screens, the plant reacts after quality has already drifted or chemicals have already been spent.
WHAT IT DOES
- LIVE PROCESS VIEW
- REGENERATION STATE MACHINE
- CYCLE-END FORECAST
- BREAKTHROUGH TREND ALERTS
- CHEMICAL & WATER LEDGER
- RO ↔ ION-EXCHANGE CORRELATION
- SHIFT HANDOVER RECORD
SIGNALS IT READS
Raw-water flow, temperature, conductivity and pH; RO feed, permeate and concentrate flows with differential pressure; cation, anion and mixed-bed outlet conductivity; chemical tank levels and dosing flow; neutralisation tank level and pH; product tank level and quality.
MODULES COMPOSED
WHAT STAYS OUTSIDE
Equipment protection, overflow, dry-run, high pressure, chemical safety and quality isolation stay in the PLC/DCS. The analytics layer produces forecasts, deviation flags and explained recommendations; it never stands in for an interlock, and it does not resize a resin bed or correct a sampling arrangement that is wrong in the field.
Ceramic Ultrafiltration Automation & AI
A modular automation layer for crossflow ceramic ultrafiltration: process control, cleaning recipes, fouling analytics, energy visibility and operator decision screens, across a single skid or a multi-train plant.
THE PROBLEM
A durable membrane is not a self-optimising process. Crossflow velocity, transmembrane pressure, flux, temperature, concentration, cleaning and pump energy sit in constant trade-off, and a drift usually leaves a trace in the flux curve or the pump load long before an alarm sounds. Cleaning on a fixed calendar then cleans some cycles too early and others too late.
WHAT IT DOES
- MEMBRANE TRAIN STATUS
- CIP & BACKPULSE RECIPES
- TMP / FLUX OPERATING ENVELOPE
- FOULING INVESTIGATION CARDS
- CLEANING RECOVERY TRACKING
- SPECIFIC-ENERGY KPIS
- PER-CYCLE RECORD
SIGNALS IT READS
Feed flow, temperature and a solids or turbidity proxy; recirculation flow and pump speed; feed, retentate and permeate pressures; permeate flow and quality indicators; retentate concentration proxy; crossflow and auxiliary power; cleaning tank level, temperature, pH/conductivity and flow; valve command and feedback; module and train identity.
WHAT STAYS OUTSIDE
High pressure, pump protection, tank overflow, chemical safety, valve position verification and emergency stop stay in the PLC/DCS. The system does not move the membrane manufacturer’s pressure, pH, temperature or chemical-resistance limits, does not select a cleaning chemical on its own, and does not claim a fouling diagnosis without laboratory confirmation.
PROCESS INTELLIGENCE
Nine process domains, each with its own page
Beyond the packaged systems above, the same five modules are applied as a modelling and decision layer to nine industrial process domains — from drying and reactors to starch conversion and plant-wide optimisation. Each domain page states its signals, its models, its measurement examples and what stays outside its authority.
- D01AI-Enabled Industrial Drying Optimisation
- D02AI-Supported Reactor Intelligence
- D03Intelligent Ultrafiltration Line Operations
- D04AI-Guided Demineralisation Lines
- D05Intelligent Polishing Operations
- D06AI-Assisted Liquefaction Control
- D07AI-Assisted Saccharification Intelligence
- D08AI-Enabled Process Improvement Discovery
- D09Real-Time Process Optimisation
SOFTWARE / HARDWARE BOUNDARY
NOVYRON licenses and operates software. Robots, cameras, sensors and controllers are supplied by you or your hardware vendors; the platform connects to them through documented, vendor-neutral interfaces. Hardware procurement can be coordinated within a project, but it is never a hidden dependency.
Which module fits your environment?
Bring the process and the systems involved — we map them to the modules above and reply with an architecture sketch, not a sales deck.