Skip to main content

Manufacturing Industry Processes

How does the manufacturing industry actually run — line by line, recipe by recipe, batch by batch?

This page describes the workflows that span the industry. The function-level twin — how a single business runs its own production internally — sits at Functional Specification and MRP / Resource Planning.

Standard vs Point-of-Difference​

Every industry-level workflow sits somewhere on a continuum. Standard workflows are commoditised — same playbook every factory runs. Point-of-Difference (PoD) workflows are where one factory beats another. AI accelerates the commoditisation of Standard; it sharpens the moat around PoD.

StepLabelWhy
Real-time OEE calculation + displaySTANDARDEvery factory should have this; Wave-2 SaaS commoditised the capability.
Predictive maintenance on rotating equipmentSTANDARDAcoustic + vibration AI is mature; vendor choice matters more than technique.
Vision QC on visual defectsSTANDARDCognex + Landing AI matured this; transfer learning across plants is the lever.
Standard work compliance + electronic work instructionSTANDARDTulip + competitors made this commodity.
Energy metering + per-batch attributionSTANDARDSub-metering hardware is cheap; the practice is what's new.
MES execution + dispatchSTANDARDMature; vendor choice; not where you win.
Custom recipe design + parameter optimisationPODProcess IP moat. AI augments engineers; engineers still make the call.
Novel product introduction (NPI) + scale-upPODCompressing weeks-of-debugging into days via simulation + AI is the differentiator.
Supply-chain resilience strategyPODDual-source + sub-tier visibility + crisis playbook; AI-assisted but judgment-led.
Multi-site production allocationPODTax + tariff + capability + carbon optimisation; the planner's craft.
Cross-functional kaizen + structured problem solvingPODCoach + operator skill; AI-assisted via root-cause analytics but driven by humans.
Strategic make-vs-buyPODCapital + capability + risk + IP fusion; pure judgment.
In-plant DePIN / DPP integration designPODEmerging discipline; today's lever is who learns first, not which vendor wins.

The pattern: engineering judgment + cross-functional craft + bet sizing survive AI commoditisation. Real-time telemetry + first-pass detection + electronic work instruction do not.

Industry-Wide Workflow Matrix​

Twelve recurring industry workflows. Each has a trigger, a sequence, an outcome, and the AI-augmentation pattern that is changing how it runs.

WI1. Inbound order → production scheduling​

  • Trigger: Customer order arrives; demand signal updates.
  • Sequence: Order receipt → demand explosion → MRP run → capacity check → schedule creation → release to floor.
  • Outcome: Production schedule that fits demand, capacity, and constraints.
  • AI pattern: AI scheduling reads constraints + due dates + changeover matrices + energy prices and proposes a schedule that beats the human-spreadsheet baseline on aggregate cost. Human approves and tweaks. The schedule becomes the contract the line executes against.

WI2. Real-time OEE measurement + alerting​

  • Trigger: Machine running; cycle counters incrementing; alarm conditions changing.
  • Sequence: Sensor → edge gateway → OEE composite calculated → operator screen + supervisor dashboard → alarm if threshold breached.
  • Outcome: Plant runs against its OEE composite in real time, not weekly.
  • AI pattern: Anomaly detection across availability + performance + quality channels; root-cause narrowing; auto-generated kaizen suggestions. Wave-2 SaaS (Factbird, Tulip, MachineMetrics) is the standard delivery vehicle.

WI3. Predictive maintenance loop​

  • Trigger: Vibration / acoustic / thermal signal crosses anomaly threshold.
  • Sequence: Sensor → edge inference → PdM platform → work-order generation → spare-parts allocation → maintenance dispatch → repair → close-out → MTBF + MTTR update.
  • Outcome: Unplanned downtime → planned downtime; mean-time-to-repair compresses.
  • AI pattern: Acoustic + vibration AI (Augury-class) trained on rotating-equipment failure libraries detects bearings, gears, motor windings before catastrophic failure. Transferable across plants — the model improves across the network.

WI4. Vision quality control + line stop​

  • Trigger: Unit moves under camera at line speed.
  • Sequence: Camera capture → vision model inference → pass/fail → if fail, line stop + reject + alert → escalation to engineer.
  • Outcome: Defects caught before pallet seal; scrap localised to the source line, not the customer return.
  • AI pattern: Continuous learning — every operator-confirmed defect becomes a training example; vision models drift-correct against the line's actual product mix. Landing AI / Cognex ViDi are the typical platforms.

WI5. Standard work + electronic work instruction​

  • Trigger: Operator logs into station for a shift; new product run starts.
  • Sequence: Operator authentication → work instruction served to screen → step-by-step compliance capture → torque + scan + cycle confirmation per step → digital traveller record.
  • Outcome: Right product built right way; per-unit traceability into the DPP record.
  • AI pattern: Tulip-class low-code apps let engineers compose work instructions without coding. AI generates first-draft work instructions from CAD + BOM + similar-product history; engineer reviews + approves.

WI6. Energy + utility metering and optimisation​

  • Trigger: Continuous metering of electrical + thermal + compressed-air consumption.
  • Sequence: Sub-meter → edge → energy platform → attribution per machine + per batch → optimisation recommendations (load shedding, peak avoidance, idle suppression).
  • Outcome: Energy enters cost-of-goods; optimisation pays back in months on rotating equipment + HVAC.
  • AI pattern: Demand-response models match production to grid-price + renewable availability; HVAC optimisation runs predictively against weather + occupancy + production load.

WI7. Quality + SPC + CAPA loop​

  • Trigger: Statistical process control chart shows out-of-control event (≥1 point outside ±3σ, or Western Electric rule).
  • Sequence: Detection → containment → root-cause investigation → corrective + preventive action (CAPA) → effectiveness verification → close-out.
  • Outcome: Defect classes prevented at source, not caught downstream.
  • AI pattern: Multivariate SPC across process tags surfaces correlations a human would miss. Root-cause narrowing via process-trace AI compresses investigation time. The CAPA itself remains human-led; AI augments evidence gathering.

WI8. Changeover + setup reduction (SMED)​

  • Trigger: Schedule transitions from product A to product B on the same line.
  • Sequence: Last-piece A → internal-setup + external-setup activities → first-good-piece B → ramp-to-cycle-time.
  • Outcome: Changeover time reduced; small-batch economics improved; flexibility increased.
  • AI pattern: Video-process-analysis AI (a Factbird capability among others) compares actual changeover to ideal-state sequence; flags lost minutes; trains the next-shift video walkthrough.

WI9. Supplier qualification + incoming inspection​

  • Trigger: New supplier proposed, or new lot arrives from existing supplier.
  • Sequence: Initial qualification (audit, samples, capability study) → ongoing incoming inspection → SCAR (supplier corrective action request) if defects.
  • Outcome: Material entering the plant meets spec; downstream defects traceable to inbound source.
  • AI pattern: Supplier-data-portal AI matches inbound certificates of analysis to spec; flags discrepancies; supplier-risk scoring across the tier-N supply chain.

WI10. Digital product passport + traceability​

  • Trigger: Unit produced; serialised; ready for DPP record.
  • Sequence: Machine signs unit data → DPP record created (BOM, source, conditions, operator, machine, energy, carbon) → on-chain attestation → DPP token attached to unit → QR / NFC printed on label.
  • Outcome: Every unit carries verifiable birth-certificate from production through customer through end-of-life.
  • AI pattern: AI doesn't dominate this workflow — cryptography does. AI assists in summarising the DPP record into customer-readable narrative.

WI11. Carbon attestation per unit produced​

  • Trigger: Continuous metering of energy + materials + processes; periodic supplier scope-3 ingestion.
  • Sequence: Scope-1 (on-site fuel + process emissions) + scope-2 (purchased energy) + scope-3 (supplier-attested upstream + customer-attested downstream) → unit allocation → on-chain attestation → DPP record update.
  • Outcome: Buyer can verify scope-1+2+3 per unit; procurement-grade evidence not vendor self-report.
  • AI pattern: AI assists in allocation across multi-product lines + reconciling supplier-attested data against expected ranges. The attestation itself is cryptographic.

WI12. AGV / cobot fleet coordination​

  • Trigger: Material movement need detected (line-side replenishment, finished-goods to warehouse, scrap removal).
  • Sequence: Job appears in fleet manager → AGV bid / assignment → path planning → execution → completion confirmation → next job.
  • Outcome: Material flow without human transport; line-side never starves; warehouse never bottlenecks.
  • AI pattern: Reinforcement-learning fleet coordination; cm-precision positioning via RTK (GEODNET-class); multi-vendor fleet coordination via open standards (VDA 5050).

Workflow Cascade — Where the Loops Connect​

WI1 Order → schedule
   ↓
WI5 Standard work executes the schedule
   ↓
WI2 Real-time OEE measures execution
   ├─► WI3 PdM catches degradation early
   ├─► WI4 Vision QC catches defects in flight
   ├─► WI7 SPC catches process drift
   ├─► WI6 Energy attribution per batch
   ├─► WI12 AGV moves material in support
   └─► WI8 Changeover compresses non-value time
   ↓
WI9 Supplier quality feeds back upstream
   ↓
WI10 DPP + WI11 carbon attestation crystallise the proof
   ↓
Outcome: order fulfilled with verifiable evidence

The compound loop: better data → better OEE → better PdM → less downtime → more throughput → more data → better models. The factory that closes this loop fastest compounds; the factory that breaks it at any link manages chaos.

What These Workflows Require​

RequirementWithout it, what breaks
Open data plane (OPC UA + MQTT + UNS)Every workflow stalls behind a vendor SDK negotiation
Real-time OEE composite on operator screenWI2 dies; downstream WI3 + WI4 + WI7 lose their trigger
Standardised event vocabulary (ISA-95, EPCIS, MTConnect)WI9 + WI10 + WI11 can't reconcile across plants and suppliers
Edge compute at the lineWI3 + WI4 stall on cloud round-trip latency
Machine identity (DID per asset)WI10 + WI11 cannot produce evidence regulators or buyers will accept
Cross-functional kaizen disciplineWI7 generates evidence; no one acts on it

What to Skip​

Don't run two PdM platforms in parallel. Pick one acoustic + vibration platform per equipment class and commit. Comparison-shopping for two years burns capex and confuses the operations team.

Don't ship work instructions on paper for one product while running Tulip on another. Mixed-mode operations destroy the standard-work signal. Commit to electronic or commit to paper; mixed mode produces the worst of both.

Don't try to optimise WI6 (energy) before you've stabilised WI2 (OEE). Energy variance follows production variance. Fix the production loop first.

Don't build a custom DPP implementation. The ESPR mandate spawns open implementations (GS1 Digital Link, EBSI, EPCIS-on-chain). Adopt the open standard early; rebuild later only if the standard is inadequate to your category.

Don't run kaizen sessions without OEE data. Kaizen without measurement is opinion exchange. The Wave-2 OEE platforms make this a non-issue; the limit is operator discipline, not tooling.

Context​

Questions​

  • Of the twelve workflows, which one does your plant run worst — and which is the single highest-leverage move to improve it?
  • The Standard vs PoD split says AI commoditises Standard. Which of your current PoD workflows is about to slip into Standard?
  • WI10 (DPP) and WI11 (carbon) are regulatory-pulled. Which one bites your buyer first?
  • Look at the workflow cascade diagram. Which link in your plant breaks first when one input upstream changes (new product, new supplier, new operator)?
  • The compound loop requires every workflow to feed the next. Which workflow in your plant is the orphan — fed by nothing, feeding nothing?