HoldFieldHoldField · Physical AI Quality Automation
Quality decisions should move with proof.
HoldField transforms inspection stations, finishing lines, electronics builds, and precision-part workflows into controlled evidence systems — starting with edge AI inspection and scaling into PLC-connected, motion-assisted, and robot-assisted quality automation.
AI detects. Rules decide. Evidence proves.
- Capture
- Coverage
- Rules
- Evidence
Visual QMS demo
Watch factory evidence become a controlled quality record.
A 95% visual walkthrough for six regulated industries. Every record is prepared for a person — nothing is auto-disposed, nothing claims certification.
- 01Signal detectedinspection · PLC · sensor
- 02Evidence sealedhash-bound · replayable
- 03QMS record preparedNCR · CAPA · deviation
- 04Human decision requiredQA reviewer approves
- 05Signature recordednamed · timestamped
- 06Audit-ready outputexportable packet
The system prepares the record. A human approves. Every time.
Physical AI Quality Automation
From manual inspection to controlled Physical AI evidence.
HoldField transforms inspection stations, finishing lines, electronics builds, and precision-part workflows into controlled evidence systems — starting with edge AI inspection and scaling into PLC-connected, motion-assisted, and robot-assisted quality automation.
- Inspection stationsCamera · lighting · evidence
- Finishing linesSurface defects · coating proof
- Electronics buildsConnector · label · OCR checks
- Precision partsBurrs · scratches · thread damage
One controlled evidence chain
- Capture
- Detect
- Measure
- Coverage
- Rules
- Route
- Seal
Edge AIVision on the station.Inspection runs on the station itself — every part, every cycle, no network round-trip.
HoldField Pod control-stack theater
Vision, PLC, Modbus, EtherCAT, robot, and QMS evidence working as one cell.
The Pod prepares evidence and bounded intent. PLC and safety remain final authority.
- 01Load5-axis bracket cell with deburr camera nest
- 02Indexservo settle window, indexed nest, camera trigger, I/O latch
- 03Capturehole-edge burr, chatter band, witness mark
- 04Evaluateserial image, measurement overlay, supplier lot, operator note
- 05PLC ackstation state, serial recipe, review route, PLC acknowledgement
- 06HandleMRB review cage
- 07QMSNCR + FAI evidence packet prepared for MRB
- serial frame
- FAI overlay
- supplier lot
- PLC ack
- MRB queue
- hash
HoldField — Physical AI Quality Automation
Multi-Station Thread Checking / Chasing Cell
Controlled motion. Verified depth. Sealed evidence.
- EtherCAT — real-time controlservo drives, remote I/O, clamp / gage / spindle, PLC cycle logic
- Modbus-TCP — supervisoryHoldField status, command intent, receipts, evidence association, health
Central control cabinet
- PLC CPUcompact DIN-rail controller — final real-time authority
- Safety / STOsafety relay or safety PLC, hardware-wired STO + E-stop
- Servo drives1–1.5 kW spindle/servo amplifiers for checking + chasing
- EtherCAT masterreal-time station + drive bus
- HMIoperator panel
- Edge gatewayHoldField EdgePod industrial PC (supervisory)
- Power protection24 VDC, disconnect, fusing, UPS
Vertical station module (×4, scales to 8–12)
- Servo spindlevertical thread tool, rotary-encoder depth
- Fixture / clamppart nest + pneumatic/servo clamp
- Thread gage probeL1 standoff / go–no-go verification
- Camera + lightingevidence frames, coverage map
- Local I/OEtherCAT remote I/O coupler
- Status towerready / pass / reject / review
PLC / Servo / Safety
- Real-time motion + servo depth
- Torque / load monitoring
- Homing, limits, interlocks
- Clamp sequence + cycle timing
- STO, E-stop, fault handling
- Pass / reject / review routing
HoldField EdgePod
- Camera + gage evidence
- Coverage proof (no coverage, no PASS)
- Rule evaluation + measurement
- Command-intent ledger + receipts
- Sealed cycle capsule + traceability
- Audit export + fleet analytics
Operator
- Mode selection
- Fault + reject review
- Retake approval
- Station reset
- Permitted manual override
Latency separated by layer
HoldField supervisory latency never affects servo depth control — depth is computed and bounded locally by the PLC/servo.
| Layer | Owner | Budget | Purpose |
|---|---|---|---|
| Safety | Safety relay / safety PLC / drive STO | Hardware path | Stop dangerous motion |
| Servo motion | Servo drive + PLC | Low-millisecond | Depth, torque, speed, chase protection |
| Station control | PLC over EtherCAT | 1–10 ms | Sequence, clamp, spindle, gage, I/O |
| HoldField supervisory | EdgePod | 50–250 ms | Verify, record, inspect, explain, trace |
Depth from rotation
- Rotationservo rotary-encoder counts from home/reference
- Pitch / leadthread pitch (1 / TPI) or screw lead ÷ gear ratio
- Calculated depthgated by homing + torque + gage + tolerance
Rotary-encoder depth is the default and is validated by homing, torque signature, gage and a tolerance gate — never trusted alone. A linear/depth encoder stays optional: the wiring and the evidence field are reserved, fitted only if validation proves it is needed.
Two modes, one machine
One machine, two modes: thread checking (low force, gage + camera verification) and thread chasing (1–1.5 kW, torque-curve and tool-break protection, then re-verification). PLC/safety remains final authority in both.
Universal machine contract
One machine contract. Any station.
PLC/safety remains final authorityHoldField normalizes PLCs, robots, gages, cameras, motion status, safety state, and evidence into one governed station contract. The PLC keeps real-time authority. HoldField verifies, records, and blocks unsafe or unproven cycles.
- MitsubishiMC / Modbus-TCP
- SiemensS7 / OPC UA
- RockwellEtherNet/IP
- BeckhoffTwinCAT / EtherCAT
- OmronSysmac / FINS
- Robot cellROS 2 handshake
- GenericOPC UA read model
- Any PLC / robot / stationMitsubishi · Siemens · Rockwell · Beckhoff · Omron · robot controller
- MachineAdapterone vendor-neutral seam; the product never talks to a vendor SDK directly
- StationSnapshotnormalized machine / station / safety / motion state — conservative by default
- AuthorityGatepure, fail-closed; blocks unsafe, unverified, or out-of-capability intents
- Evidence capsulesealed, hashed receipt — no coverage, no PASS
- Operator review / PLC routethe PLC performs the physical routing; HoldField proves the decision
Impossible by design
- No safety write
- No direct servo control
- No forced PASS
- No clearing a safety fault
- No remote STO reset
Enforced at type and runtime level — no adapter can claim them.
Real PLC connection, governed by evidence
HoldField connects to PLC stations through a normalized machine contract. Modbus-TCP v1 supports conservative digital I/O supervision: station status, safety state, approved intent, acknowledgment, and receipt binding. Real-time motion stays inside the PLC and servo drive.
- PLC digital I/O
- ModbusTcpMachineAdapter
- StationSnapshot
- AuthorityGate
- Intent ledger
- PLC request (pulsed)
- Receipt
- Evidence capsule
From PLC signal to sealed evidence
HoldField now normalizes live PLC station status into one machine contract. Every approved station request passes through safety state, adapter health, command intent, PLC acknowledgment, and receipt binding before it can become part of a quality decision.
- PLC / station I/O
- ModbusTcpMachineAdapter
- StationSnapshot
- AuthorityGate
- Intent ledger
- Receipt
- Evidence capsule
- Operator review
From machine state to operator action
HoldField turns live PLC station status into a simple operator screen and a deep engineer console. Operators see safe next actions. Engineers see adapter health, signal mapping, latency, receipts, and blocked reasons. The machine still controls motion; HoldField proves and governs every request.
Operator sees
- Ready
- Running
- Review Required
- Retake Required
- PLC Offline
- Safety Not OK
- Receipt Unresolved
- Faulted
Engineer sees
- Adapter health
- Signal mapping
- Latency metrics
- Authority decisions
- Receipt recovery
- Blocked reasons
From station status to the right screen. HoldField gives operators a safe station view and engineers a diagnostics console from the same machine contract. Operators see what to do next. Engineers see the signal map, receipts, latency, and blocked reasons. The PLC still controls the machine; HoldField governs the evidence path.
Physical AI
Automation ladder
Evidence moves through the station — from capture to a customer go/no-go.
- Edge AI inspectionCapture defects with local evidence.
- PLC-connected routingRoute results without bypassing safety.
- Motion-assisted coverageProve every required view.
- Robot-assisted handlingBounded, commissioned skills only.
- Controlled deploymentRun a supervised deployment before promotion.
Physical AI · Controlled deployment
Prove it in a supervised deployment
Run a supervised deployment before any station is promoted.
- Shadow deployment
- Uncertain case queue
- False-reject review
- Escape candidate review
- Drift signal
- Go / no-go packet
The problem
Quality risk is expensive because it stays hidden too long.
Missed defect
Today
A defect slips the visual check and ships to the customer.
HoldField
Controlled capture, measurement, and a rule gate catch it at the station — before the part moves.
Escape rateThe layer
One controlled layer, from capture to evidence.
Perception, measurement, coverage, rules, station health, human review, and evidence work together before a decision moves forward.
01 · Capture
Controlled frames under a repeatable lighting recipe — comparable across every shift.
AI is a signal source, not the unchecked final judge.
Proof Explorer
See exactly what HoldField proves in your world.
Pick a use case. Watch one part travel the real proof chain — end to end.
Final visual inspection proof for regulated device and component lines.
Part on line
Finished devices, machined implants, and molded components before release.
- Surface flaws and burrs
- Missing or misaligned features
- Contamination or residue
- Label / UDI mismatch
Station capture
Fixed station cameras and lighting capture the part on the line — no cloud round-trip, evidence stays local.
- signal
Edge AI signal
Edge AI flags candidate defects as a signal
- deterministic
Rules decide
Deterministic rules apply the criteria and thresholds
- no coverage · no PASS
Coverage gate
Coverage must be complete before a PASS is allowed
Sealed evidence capsule
- Raw frame + overlay + defect crop
- Coverage map and measurements
- Rule and model versions
- Operator + station + timestamp
- Hash-sealed evidence capsule
Audit-ready proof
A replayable, tamper-evident record per unit — evidence-ready for QMS and traceability reviews (HoldField never claims certification).
Worlds
See the proof chain for your world.
Public visual examples only. Customer data stays private. No named-company claims.
Evidence
Every decision becomes replayable evidence.
Raw frames, overlays, defect crops, coverage maps, measurements, thresholds, rule and model versions, station IDs, timestamps, reviewer actions, and hashes stay together — in one sealed capsule.
The HoldField Proof Spine
Proving industrial readiness without moving station authority.
- Station stays local.
- Evidence becomes reviewable.
- Customers see readiness.
- Deployment expands only after proof.
Capture
Camera, gage, PLC state, and operator action — at the station.
- Camera
- Gage
- PLC state
- Operator action
Prove
Coverage, rule result, receipt, hash, review packet.
Govern
Role-scoped view, redacted evidence, audit trail, no station command.
Deploy
Preflight, rollback, readiness packet, operational acceptance.
Expand
Site comparison, safety delta, measured value, next station gate.
Every surface records and recommends. The PLC controls, safety protects, and a person decides.
Trust
Trust boundaries
No AI makes the final call alone. No data integration controls hardware. No robot skill runs outside the safety gate. No simulator-only proof is sold as hardware readiness.
- PLC/safety remains final authority.
- Edge AI is a signal source, not the safety authority.
- Rules decide against the active criteria pack.
- PASS requires evidence, coverage, and rule compliance.
- Missing coverage becomes REVIEW, RETAKE, or BLOCK — not silent PASS.
- UNS/DataOps publish data only.
- Robot skills are bounded and commissioned.
- Rotary motion is safety-gated and coverage-bound.
- Quantum/PQC is security-policy-oriented and never controls hardware.
- Simulator-only proof is never sold as hardware readiness.
Trust
Built as governed industrial infrastructure.
- AI signal
- Defect candidate located on the part face.
- Coverage
- All critical regions inspected — no coverage, no pass.
- Rules
- Measured 1.4 mm exceeds the 1.0 mm critical-region threshold.
- Decision
- REJECT — made by the deterministic rule pack, not the model.
- Evidence
- Raw + overlay + crop + coverage + measurement + hash, sealed.
- Fallback
- On ambiguity: review, retake, or block the station — never a silent pass.
No AI makes the final call alone. See the full governance model.
Select a control to see how it is enforced.
Edge-first deployment — Inspection runs on the station at the edge — a decision never waits on a network round-trip.
Deployment
Start with one painful inspection point.
A Station Fit Report maps the part, defect classes, lighting, evidence needs, integration path, and KPIs before a single station goes live.
- Part & stationThe one part family and the single painful inspection point to start at.
- Defect classesThe failure modes that actually matter — and their accept/reject thresholds.
- Lighting recipeThe repeatable condition that reveals each defect family at the station.
- Evidence needsWhat a sealed decision must carry to satisfy your audit and customers.
- Integration pathHow the governed result reaches the PLC and MES — HoldField never bypasses control.
- KPIs & gateThe numbers — escape rate, coverage, dispute time — a deployment must move to convert.