Asep Irfan Setiawan
Back to all case studies
AI2026Automation & Software Engineer

AI/ML Quality Inspection: Computer Vision NG Defect Detection

Deep learning computer vision model (YOLOv8) deployed to automatically detect and classify NG (defective) components and surface flaws on high-volume production lines.

PLC (Mitsubishi, Omron, Siemens, LS)Modbus Protocol (RTU/TCP)Industrial Electrical Wiring & SafetyPythonComputer Vision (OpenCV, YOLO)FastAPITkinter (Desktop GUI)
Open Live System
AI/ML Quality Inspection: Computer Vision NG Defect Detection Cover
AI/ML Quality Inspection: Computer Vision NG Defect Detection Gallery Image 1
AI/ML Quality Inspection: Computer Vision NG Defect Detection Gallery Image 2
1 / 3
AI/ML Quality Inspection: Computer Vision NG Defect Detection — Gallery Image (1 of 3)

1. Context & Problem Statement

An automated visual inspection system powered by fine-tuned deep learning object detection models. The system inspects manufactured parts in real time, detecting missing domsheet pcb that previously bypassed manual operator checks.

2. System Architecture & Workflow

                         DEFECT AI INSPECTOR

┌─────────────────────────────────────────────────────┐
│                INSPECTION HARDWARE                  │
│                                                     │
│  [Part / PCB Jig] → [Sensor] → [Industrial Camera] │
└──────────────────────────┬──────────────────────────┘
                           │
                           │ Image Frame
                           ▼
┌─────────────────────────────────────────────────────┐
│                  AI VISION ENGINE                   │
│                                                     │
│  Image Acquisition                                  │
│        ↓                                            │
│  Image Preprocessing                                │
│        ↓                                            │
│  ROI Extraction                                     │
│        ↓                                            │
│  AI Defect Detection                                │
│        ↓                                            │
│  Bounding Box + Confidence                          │
│        ↓                                            │
│  OK / NG Decision                                   │
└──────────────────────────┬──────────────────────────┘
                           │
                   ┌───────┴───────┐
                   ▼               ▼
                  OK               NG
                   │               │
                   │               ├────► [Warning Lamp]
                   │               │
                   │               ├────► [Alarm / Buzzer]
                   │               │
                   │               └────► [PLC Interlock]
                   │                         │
                   │                         ▼
                   │                  Block Pick & Place
                   │
                   └──────────────► [PLC]
                                     │
                                     ▼
                              Allow Pick & Place

3. Technical Highlights & Automation

  • AI-Based Defect Detection: Detects and localizes visual abnormalities on PCB assemblies using bounding boxes and confidence scores.
  • PLC Production Interlock: Inspection results are integrated with the PLC to control whether the production sequence is allowed to continue.
  • Fail-Safe NG Handling: An NG result automatically blocks the pick-and-place process until the condition is acknowledged or reset.
  • Visual & Audible Alarm: Defect detection activates a warning lamp and alarm to immediately notify the operator.
  • Automatic Production Permission: An OK result sends a production-ready signal to the PLC, allowing the pick-and-place sequence to continue.
  • Real-Time Vision Processing: The Python-based AI pipeline processes inspection images locally for low-latency decision making.
  • Realtime Inspection Dashboard: Displays detection images, bounding boxes, OK / NG status, processing time, counters, and system logs.
  • Traceability: Inspection results and defect information can be stored for production analysis and quality tracking.

4. Measurable Results & Impact

  • Automated visual defect inspection and reduced dependency on continuous manual operator checking.
  • Prevented detected NG parts from proceeding to the pick-and-place process through PLC interlock integration.
  • Improved defect traceability by recording inspection results, confidence scores, and system events.
  • Reduced response time to production defects by automatically triggering warning lamps and alarms when NG conditions are detected.
  • Improved inspection consistency by applying the same AI-based validation criteria to every inspected part.
  • Enabled real-time monitoring of OK / NG counts, processing latency, and inspection status from a centralized dashboard.

Engineering Retrospective

Real-world dataset curation and domain-specific lighting setup represent 80% of successful computer vision deployment in industrial environments.

Related Case Studies