Case Study

Automated Board Inspection System

A turnkey machine vision platform that replaces manual QC with 24/7 AI-powered surface inspection — delivering 100% board coverage, sub-millisecond decisions, and full production traceability.

Machine Vision PLC Automation AI Defect Detection Flat Panel Manufacturing SPC / Traceability
100% Board Coverage
≥98% Detection Accuracy
10× Camera Array
<100ms Response Time

Manual QC Was the Bottleneck

High-volume flat panel board manufacturing demanded inspection at a scale and consistency that human visual QC simply cannot sustain.

👁️

Inconsistent Inspection Quality

Inspector fatigue across long shifts caused defects to escape undetected — with no way to measure or correct the variability.

📐

Massive Surface Area per Board

Boards up to 1240 × 2445 mm meant each unit required thorough multi-point coverage — impractical at speed with human operators.

📊

Zero Data & Traceability

No defect logs, no yield data, no SPC capability. Root cause analysis and customer quality reports were impossible.

⏱️

Throughput Constraints

Growing production volumes outpaced what manual inspection could handle without adding significant headcount and floor space.

🔄

No Standardised Classification

Defect identification was subjective — what one inspector flagged as a reject, another might pass. No objective benchmark existed.

💸

Rising Labour & Compliance Costs

Training, supervision, audit trails, and regulatory reporting overhead grew with headcount. Defects that shipped cost far more to remedy.

End-to-End Vision Automation

Indus Vision designed and built a fully integrated, conveyor-fed machine vision platform — from mechanical frame to AI software — deployed as a single turnkey system.

SYSTEM ARCHITECTURE — 4 INTEGRATED LAYERS
Mechanical
🏗️
Frame
Modular steel / aluminum gantry
⚙️
Conveyor
Variable speed sync belt
💡
Lighting
Diffuse LED + structured light
🎯
Positioning
Servo motion control
Imaging
📷
Camera Array
10× GigE PoE cameras
🔌
Gigabit Switch
1 Gbps PoE uplink
🖼️
Frame Buffer
High-speed acquisition
Sync Unit
Hardware trigger / timing
Processing
🧠
AI Engine
NVIDIA Jetson AGX
📈
Inference
Deep learning models
🔍
Analysis
Defect classification
⚠️
Decision Logic
Pass / Fail rules engine
Control & Integration
🎛️
PLC
Siemens S7-1200
💾
Database
PostgreSQL time-series
📊
Dashboard
Real-time SPC / OEE
🔗
Integration
REST API / MQTT

6-Step Automated Inspection Cycle

Every board follows the same deterministic path — from infeed to result — with zero manual intervention required during normal operation.

1
📥

Infeed

Board positioned on conveyor

2
📸

Capture

10 camera array acquires frames

3
🔀

Stitch

Images merged into single mosaic

4
🧠

Infer

AI model detects defects

5

Decide

Pass / Fail + severity logged

6
📤

Eject

Pneumatic or mechanical reject

1
Image Acquisition
All 10 cameras triggered simultaneously at board entry point. High-speed GigE captures 12-bit RAW images in ~50 ms.
2
Real-Time Processing
NVIDIA GPU stitches images, normalizes lighting, and runs defect detection inference in <100 ms per board.
3
Data Logging & Ejection
Results stored in time-series DB. PLC signal triggers pneumatic reject arm or conveyor divert within 200 ms.

What Goes In, What Comes Out

The system consumes five input streams and produces five outputs — transforming raw boards and data signals into quality decisions and traceability records.

Inputs

📦
Physical Board

Conveyor-fed flat panel up to 2445 mm × 1240 mm

Conveyor Signal

Speed feedback & position pulse

🔌
Power Supply

24 VDC + 110/240 VAC

⚙️
Configuration

Defect thresholds & material type

🔗
Network

Ethernet to ERP / MES / server

🎯
Indus Vision

Outputs

Pass / Fail Signal

24 VDC relay to reject mechanism

📊
Defect Report

Type, location, severity, confidence

📷
Image Archive

Full-res stitched board images

📈
SPC Data

Yield, defect type distribution, trends

🔗
API Integration

REST / MQTT to ERP, MES, historian

Guaranteed KPIs

Every deployment is governed by contractually defined performance targets — measurable outcomes from day one of commissioning.

System Performance Targets
Contractually defined — measured at commissioning and throughout operation
Detection Accuracy ≥98%
True positive + True negative / Total
Board Coverage 100%
Zero blind spots across surface
Decision Time <100 ms
Image capture through reject signal
System Uptime ≥99%
Excluding planned maintenance
Weeks 1–2
Requirements workshop, board dimensional audit, defect taxonomy definition
Weeks 3–6
System design, AI model training, mechanical build, integration & calibration
Weeks 7–8
Site preparation, installation, commissioning, operator training & handoff

What Changes After Deployment

From the moment the system goes live, the entire quality function is transformed — faster, more consistent, and fully data-driven.

📊

100% Traceability

Every board inspected, every defect logged with image, location, and timestamp. Full SPC capability.

Defect record per unit

Instant Quality Feedback

Real-time defect dashboard shows yield trends, root cause alerts, and anomaly detection across shifts.

Sub-second latency

Zero Escapes (Contractual)

AI-powered defect detection removes human fatigue. Consistent decisions. Fewer defects reaching customers.

≥98% accuracy
💰

Cost Reduction

Eliminates 3–5 full-time QC inspectors. Reduces rework, scrap, and customer warranty claims by 40%+.

ROI: 12–18 months
📈

Higher Throughput

Continuous 24/7 operation. No breaks, no fatigue. Capacity constrained only by mechanical line speed.

Boards/hour: unlimited
🔗

Compliance Ready

Automated audit trails, defect images, and SPC reports meet ISO 9001, IPC-A-610, and customer audits.

Auditable proof

System at a Glance

Key hardware and software specifications for the Board Inspection Automation System.

Category Component Specification
Imaging Camera Array 10× GigE PoE 5MP 12-bit sensors, 50 mm lens, ≥30 fps
Imaging Frame Grabber Simultaneous capture 1 Gbps network switch with PoE
Processing AI Engine NVIDIA Jetson AGX CUDA compute, 576 TOPS peak
Processing Inference Runtime ONNX / TensorRT Defect detection + classification
Control PLC Siemens S7-1200 Conveyor sync, reject mechanism, I/O control
Control Connectivity REST API + MQTT ERP / MES / historian integration
Database Time-Series Storage PostgreSQL + TimescaleDB Defect logs, images, metrics
Mechanical Frame Modular 80/20 Support for up to 2500 mm × 1300 mm boards
Mechanical Conveyor Variable speed belt 0.5–2.0 m/min adjustable, servo sync
Mechanical Lighting Diffuse LED + structured Full spectrum 380–700 nm, ≥3000 lux
Performance Detection Accuracy ≥98% Verified at commissioning
Performance Decision Latency <100 ms Image capture through PLC output
Performance Coverage 100% Zero blind spots
Performance Uptime ≥99% Excluding planned maintenance
Software Dashboard Web-based Real-time SPC, OEE, defect heatmaps
Software Data Export CSV / JSON / Parquet Reports, audit trails, compliance
Power Consumption ~3.5 kW peak Cameras + GPU + PLC + lighting + conveyor
Installation Footprint ~4 m × 2 m Modular, can be integrated inline
Installation Timeline 8 weeks From order to live production

Ready to Automate Your Quality Control?

This system can be adapted for your board dimensions, production speed, and integration requirements. Let's talk about your application.

Talk to Our Team →
Scroll to Top

Request a Demo