Who this is for: IT architects designing edge computing infrastructure, operations teams deploying cameras at remote or disconnected sites, and project managers planning drone-based inspection programs. If your cameras outnumber your inspectors, this is your deployment guide.
Read Time: 14-16 minutes
MVI Edge: Real-Time AI at the Source
When You Need Edge
MVI Mobile is for inspectors with cameras. MVI Edge is for cameras without inspectors.
MVI EDGE USE CASES
==================
Fixed Camera Inspection:
- Production line quality gate
- Facility perimeter monitoring
- Asset condition monitoring (hourly/daily capture)
- Safety compliance (PPE detection)
Remote/Disconnected Sites:
- Offshore platforms
- Remote substations
- Mining operations
- Pipeline right-of-way
Latency-Critical:
- Production line real-time reject
- Moving asset inspection (rail wheels)
- Automated sorting
Bandwidth-Constrained:
- Satellite-connected sites
- Sites with limited WAN
- High-resolution image volume
GigE Vision Camera Integration (MAS 9.0+):
- Basler cameras with Power over Ethernet
- Direct camera-to-MVI Edge pipeline
- Industrial-grade camera connectivityMVI Edge Architecture
MVI EDGE DEPLOYMENT
====================
┌─────────────────────────────────────┐
│ EDGE LOCATION │
│ (Factory Floor / Remote Site) │
│ │
│ Camera ──> Edge Device ──> Result │
│ (incl. │ │
│ GigE │ │
│ Vision)┌────┴────┐ │
│ │ MVI │ │
│ │ Edge │ │
│ │ Server │ │
│ │ │ │
│ │ GPU │ │
│ │ Models │ │
│ │ Queue │ │
│ └────┬────┘ │
│ │ │
│ Local Action │
│ (Alert / PLC / Display) │
│ │ │
│ MQTT Alert ──> Monitor │
│ Twilio SMS ──> Operator │
└──────────────┬──────────────────────┘
│
When Connected
│
┌──────────────┴──────────────────────┐
│ MVI SERVER (Central) │
│ - Model updates pushed to edge │
│ - Results synced for reporting │
│ - Retraining data collected │
│ - Centralized monitoring │
│ - Edge diagnostics dashboard │
│ (MAS 9.0+) │
└─────────────────────────────────────┘MQTT Alert Pipeline (MVI Edge to Monitor)
MAS 9.0 introduced a significantly enhanced alert pipeline from MVI Edge to Maximo Monitor.
MVI EDGE ALERT PIPELINE
========================
HOW IT WORKS:
────────────
1. MVI Edge detects object/defect in image
2. Detection triggers configured alert rule
3. MVI Edge sends MQTT alert message
4. Message received by Maximo Monitor
5. Monitor processes alert per rules
6. Actions triggered (WO, notification, escalation)
AUTO-CONFIGURATION:
──────────────────
MVI Edge auto-configures:
- Generic device type in Monitor
- Device gateway for MQTT communication
- No manual Monitor setup for basic alerts
v2 API INTEGRATION (MAS 9.0+):
──────────────────────────────
- MQTT messages sent via v2 APIs
- Faster alerting pipeline
- Enhanced message payload with image data
- Improved reliability and error handling
ALERT MESSAGE TEMPLATES (MAS 9.0+):
───────────────────────────────────
Admins can create MULTIPLE reusable templates
defining the structure of:
- MQTT messages (to Monitor)
- Twilio SMS messages (to operators)
Each template defines:
- Message format and fields
- Which detection data to include
- Routing information
- Severity mapping
RULE-BASED ALERTS:
─────────────────
Users configure rules PER INSPECTION:
- Rule 1: "If crack detected" → MQTT critical alert
- Rule 2: "If corrosion detected" → MQTT warning
- Rule 3: "If PPE missing" → Twilio SMS to supervisor
Different inspections can use different rules
and different templates.Edge Device Diagnostics (MAS 9.0+)
EDGE DIAGNOSTICS DASHBOARD
==========================
MAS 9.0 introduced centralized monitoring
of edge device health:
Per Device:
- Online/offline status
- Last communication timestamp
- GPU utilization
- Storage capacity
- Model versions deployed
- Inference throughput
- Error rate
Fleet View:
- All edge devices on single dashboard
- Alert when device goes offline
- Alert when storage running low
- Model version compliance across fleet
- Aggregate inference statisticsEdge Hardware Options
EDGE HARDWARE SELECTION
=======================
VERIFIED EDGE DEVICE:
────────────────────
NVIDIA Jetson Xavier NX
- Verified by IBM with nvidia-jetpack 4.5.1-b17
- This is the specifically documented edge device
- Other Jetson models work but are not formally
listed in IBM documentation
ENTRY LEVEL: NVIDIA Jetson Nano / Orin Nano
────────────────────────────────────────────
GPU: 128/1024 CUDA cores
Inference: 5-15 images/second
Power: 5-15W
Cost: $200-500
Use: Low-volume, single camera
Form factor: Small, fanless options available
MID RANGE: NVIDIA Jetson Xavier NX / Orin NX
────────────────────────────────────────────
GPU: 384/1024 CUDA cores
Inference: 15-30 images/second
Power: 10-25W
Cost: $400-800
Use: Multi-camera, moderate volume
Form factor: Small, industrial enclosures available
NOTE: Xavier NX is IBM-verified platform
HIGH PERFORMANCE: NVIDIA Jetson AGX Orin
────────────────────────────────────────
GPU: 2048 CUDA cores
Inference: 30-60 images/second
Power: 15-60W
Cost: $1,000-2,000
Use: High-volume, real-time production line
Form factor: Larger, active cooling
ENTERPRISE: GPU Server at Edge
──────────────────────────────
GPU: NVIDIA T4/A2 in server chassis
Inference: 50-200 images/second
Power: 200-400W
Cost: $5,000-15,000
Use: Multi-line factory, site-wide deployment
Form factor: Rack-mounted or industrial enclosure
EDGE LICENSING:
──────────────
MVI Edge has SEPARATE device-based licensing.
Each edge device requires its own license,
independent of MAS AppPoint user licensing.
DECISION FACTORS:
- Images per second required
- Number of cameras connected
- Environmental conditions (heat, dust, vibration)
- Power availability
- Physical space constraints
- Network connectivity pattern
- GigE Vision camera compatibility (MAS 9.0+)Edge Deployment Patterns
PATTERN 1: ALWAYS-CONNECTED EDGE
═════════════════════════════════
Camera ──> Edge Device ──> Local Inference ──> Result
│
┌──────┴──────┐
│ │
Local Action MQTT to Monitor
(real-time) (near real-time)
Use when: Reliable network exists but latency matters
Example: Factory floor with plant WiFi
Alert path: MQTT messages to Monitor via v2 APIs
PATTERN 2: OCCASIONALLY-CONNECTED EDGE
══════════════════════════════════════
Camera ──> Edge Device ──> Local Inference ──> Result
│
Local Storage
(queue results
+ alerts)
│
When Connected:
│
Batch Sync to Cloud
+ Queued MQTT alerts
Use when: Network intermittent
Example: Remote substation with satellite uplink
PATTERN 3: AIR-GAPPED EDGE
══════════════════════════
Camera ──> Edge Device ──> Local Inference ──> Result
│
Local Action ONLY
(alert, display, PLC)
Model updates: Physical media (USB)
Result export: Physical media or local system
Use when: No network connectivity at all
Example: Classified facility, deep mine, offshoreDrone Integration: Eyes in the Sky
Drones are the most transformative image source for MVI. They reach what inspectors cannot, cover area that fixed cameras miss, and generate the volume of imagery that makes AI worthwhile.
Drone Integration Patterns
PATTERN 1: BATCH UPLOAD (Simplest)
══════════════════════════════════
Step 1: Drone flies inspection route (automated or manual)
Step 2: Landing. SD card removed or WiFi transfer
Step 3: Images uploaded to MVI server
Step 4: MVI processes batch (minutes to hours)
Step 5: Results reviewed by inspector
Step 6: Work orders generated for findings
Timeline: Findings available 1-4 hours after flight
Best for: Scheduled inspections (monthly/quarterly)
Complexity: Low
PATTERN 2: EDGE STREAMING (Real-Time)
════════════════════════════════════
Step 1: Drone streams video to ground station
Step 2: MVI Edge at ground station processes frames
Step 3: Real-time detection overlay on operator display
Step 4: Operator marks critical findings
Step 5: Drone can re-inspect flagged areas immediately
Step 6: Results synced to MVI server post-flight
Timeline: Findings available during flight
Best for: Critical inspection, search operations
Complexity: High
PATTERN 3: HYBRID (Most Common)
═══════════════════════════════
Step 1: Drone captures high-resolution stills
(triggered at waypoints or interval)
Step 2: Lower-res preview streamed to ground station
Step 3: MVI Edge runs fast screening on preview
Step 4: Flagged areas trigger high-res capture
Step 5: Full-resolution batch processed post-flight
Step 6: Results merged and reviewed
Timeline: Critical alerts during flight,
full analysis 1-2 hours after
Best for: Large-area inspection with critical items
Complexity: MediumDrone Program Requirements
DRONE INSPECTION PROGRAM CHECKLIST
===================================
REGULATORY:
[ ] Part 107 (US) or equivalent certification
[ ] Airspace authorization for inspection areas
[ ] Insurance coverage for drone operations
[ ] Flight logging and record-keeping system
HARDWARE:
[ ] Drone: DJI Matrice 300/350 or equivalent
- Flight time: 40+ minutes
- Camera: 20+ MP, optical zoom
- RTK GPS for repeatable waypoints
- Wind resistance: 25+ mph
[ ] Ground Station:
- MVI Edge device (Jetson AGX or GPU server)
- High-brightness monitor for outdoor use
- Reliable power (generator or battery)
[ ] Storage:
- High-speed SD cards (V60+ rating)
- Portable SSD for transfers
- Cloud storage for long-term archive
FLIGHT OPERATIONS:
[ ] Flight routes defined per asset/site
[ ] Waypoints set for consistent capture angles
[ ] Altitude optimized for defect resolution
(Rule: defect must be 20+ pixels in image)
[ ] Overlap between images for full coverage
[ ] Weather minimums defined (wind, rain, light)
MVI INTEGRATION:
[ ] Models trained on drone imagery (not ground photos)
[ ] Camera settings consistent with training images
[ ] Batch upload pipeline tested
[ ] Naming convention for image-to-asset mapping
[ ] Georeferencing for defect location mapping
[ ] Alert rules configured for Edge streamingCamera-to-Resolution Math
DEFECT DETECTABILITY CALCULATOR
================================
Minimum defect size to detect: 20 pixels
Camera resolution: 20 MP (5472 x 3648)
Sensor FOV at distance:
Altitude Ground Width Pixel Size Min Defect
(meters) (meters) (mm/pixel) Detectable
───────── ──────────── ────────── ──────────
10 8.5 1.6 mm 32 mm
20 17.0 3.1 mm 62 mm
30 25.5 4.7 mm 94 mm
50 42.5 7.8 mm 156 mm
100 85.0 15.5 mm 310 mm
EXAMPLE: To detect a 50mm crack, you need to fly
at 15 meters or lower (with a 20MP camera).
OPTICAL ZOOM CHANGES THE MATH:
30x zoom at 100m altitude = equivalent of 3.3m
Detectable defect: ~10mm
RULE: Test with your actual drone + camera + altitude
on known defects before committing to flight plans.Field Deployment Patterns: What Works
Pattern: The Morning Sync Model
MORNING SYNC PATTERN
====================
(Best for mobile crews with intermittent connectivity)
06:00 - Crew arrives at dispatch office
- Sync iOS devices: latest Core ML models
- Download offline assignments
06:30 - Deploy to field
07:00 - Begin inspections
to - MVI Mobile runs offline on iPhones/iPads
15:00 - All results stored locally on device
- Camera images + MVI analysis + inspector notes
15:30 - Return to dispatch office
- Auto-sync results to MVI server
- Work orders generated from day's findings
- Override data captured for retraining
16:00 - Review flagged findings with supervisor
- Confirm/reject MVI detections
- Priority work orders escalated
OVERNIGHT:
- MVI server processes any additional analysis
- Dashboard updates for management review
- Retraining data incorporated weeklyPattern: The Continuous Monitoring Model
CONTINUOUS MONITORING PATTERN
=============================
(Best for fixed cameras with edge devices)
SETUP:
- Fixed cameras at inspection points
(including GigE Vision cameras on MAS 9.0+)
- MVI Edge device per camera cluster (1-8 cameras)
- Network connection to MVI server (can be intermittent)
- MQTT alert rules configured per inspection
OPERATION:
- Camera captures image every N minutes
(N = 1 to 60 depending on criticality)
- MVI Edge analyzes immediately
- Results:
- Normal: Log locally, batch sync hourly
- Anomaly: MQTT alert to Monitor immediately
- Critical: MQTT alert + Twilio SMS to operator
+ auto-create work order
EXAMPLE: Cooling Tower Monitoring
- 4 cameras (2 GigE Vision + 2 IP cameras)
- Image capture every 15 minutes
- MVI Edge on Jetson Xavier NX
- Detects: Excessive drift, structural damage,
fill degradation, fan blade issues
- Normal: 384 images/day, all processed locally
- Alert threshold: 2-3 MQTT alerts per week average
- Human review: Morning dashboard check
- Edge diagnostics: Monitored from central dashboardKey Takeaways
- MVI Edge enables fixed-camera and real-time inspection -- Edge devices process images locally with sub-100ms latency. MQTT alert pipelines to Monitor (enhanced with v2 APIs in MAS 9.0) provide automated alerting with configurable alert templates and rule-based routing.
- Offline-first design is mandatory for edge -- Network failures are when, not if. Edge deployments must function without connectivity, queue results locally, and sync gracefully. Edge devices queue MQTT alerts for delivery when connected.
- Drone integration transforms large-area inspection -- Batch upload for scheduled inspections, edge streaming for real-time critical detection. The camera-to-resolution math determines what defects you can detect at what altitude. Do the math before you fly.
- Edge hardware selection matches inference requirements -- NVIDIA Jetson Xavier NX is the IBM-verified edge platform. Jetson Nano for single-camera low-volume. Jetson AGX Orin for multi-camera real-time. GPU server for enterprise-scale edge. MAS 9.0 adds GigE Vision camera support and edge diagnostics dashboards.
- Three edge patterns cover all connectivity scenarios -- Always-connected for factory floors, occasionally-connected for remote sites, and air-gapped for classified or deep-field operations. Choose based on your network reality.
- Each edge device requires separate licensing -- MVI Edge has device-based licensing independent of MAS AppPoint user licensing. Factor this into your edge deployment budget.
What Comes Next
You have models running everywhere: server, mobile (iOS), edge, drones. In Part 9, we connect everything back to the MAS ecosystem. How inspection findings become work orders in Manage, condition scores in Health, alerts in Monitor (via MQTT), and training data for Predict. The closed loop that makes visual inspection operational.
Previous: Part 7 - MVI Mobile: AI-Powered Inspection on iOS
Next: Part 9 - Integration with MAS Applications
Series: MAS VISUAL INSPECTION | Part 8 of 12
TheMaximoGuys | Enterprise Maximo. No fluff. Just results.
The Maximo Guys
