Motionwell has delivered machine vision inspection systems across medical device assembly (BD projects P22068, P23003), electronics panel assembly (ifm P23045), and pharmaceutical serialization lines in Singapore since 2009. This guide covers the technical fundamentals of machine vision inspection: how cameras, lighting, and algorithms work together to replace manual visual inspection on production lines.
Machine vision inspection is now the most common automation upgrade in Singapore manufacturing. According to the Singapore Economic Development Board, over 60% of precision engineering firms have adopted some form of automated inspection. The technology is mature, the hardware costs have dropped, and the integration patterns are well-established.
What Is Machine Vision Inspection?
Machine vision inspection uses cameras, controlled lighting, and image processing software to automatically evaluate parts on a production line. The system captures an image, processes it against defined criteria, and makes a pass/fail decision or records a measurement — typically in under 100 milliseconds.
A machine vision system consists of four components:
| Component | Function | What determines the choice |
|---|---|---|
| Camera | Captures the image (area scan or line scan) | Resolution requirement, field of view, frame rate |
| Lens | Controls magnification, focus, and depth of field | Working distance, feature size, depth of field requirement |
| Lighting | Makes defects or features visible and repeatable | Defect type, surface material, ambient light conditions |
| Software/processor | Runs algorithms, stores results, communicates with PLC | Inspection complexity, speed requirement, data storage needs |
The most common failure point in machine vision projects is not the camera or software. It is the lighting. A defect that is invisible under one lighting configuration becomes obvious under another. This is why Motionwell tests production parts under 4-6 different lighting setups before specifying any system.
Types of Machine Vision Inspection
Defect Detection (Pass/Fail Inspection)
The camera captures every part at line speed and compares the image against trained references. Defects — scratches, dents, contamination, missing features, misalignment — trigger a reject signal to a pneumatic diverter or robot sorting station.
Defect detection applications range from simple presence/absence checks (is the label there?) to complex surface inspection requiring AI-assisted classification (is this scratch cosmetic or functional?).
Key engineering decisions:
- Lighting geometry (diffuse, directional, dark-field, backlight) determines which defects are visible
- Camera resolution must match the smallest defect size with margin
- Algorithm selection: pattern matching for known defect types, anomaly detection for unknown types
- False reject rate target (typically under 0.1% for production use)
Dimensional Measurement (Inline SPC)
High-resolution cameras with calibrated telecentric lenses measure critical dimensions at production speed. Pin positions, gap widths, hole diameters, component heights, and alignment offsets are measured to sub-pixel accuracy and fed into SPC (Statistical Process Control) systems.
The challenge in inline measurement is repeatability across thermal drift, vibration, and part-to-part variation. Motionwell designs measurement stations with thermal compensation, anti-vibration mounting, and operator-accessible calibration routines.
Typical measurement capabilities:
| Parameter | Typical accuracy | Camera requirement |
|---|---|---|
| Linear dimension (2D) | ±0.01 mm | 5MP+ with telecentric lens |
| Position offset | ±0.005 mm | 5MP+ with sub-pixel edge detection |
| Angle measurement | ±0.05 degrees | Calibrated system with edge fitting |
| Height/flatness (3D) | ±0.01 mm | Laser profiler or structured light |
| Diameter/circularity | ±0.005 mm | Backlit configuration with calibrated optics |
Barcode and Data Matrix Grading (ISO/IEC 15416)
Barcode grading goes beyond simple reading. The system evaluates print quality parameters — symbol contrast, modulation, decodability, defects, quiet zone — and assigns a grade (A through F) per ISO/IEC 15416 for 1D barcodes or ISO/IEC 15415 for 2D Data Matrix codes.
Barcode grading is mandatory in pharmaceutical serialization, medical device UDI compliance, and automotive traceability. The grading result is stored with the production record as evidence of label quality.
Label Verification (OCR/OCV)
Optical Character Recognition (OCR) reads printed text. Optical Character Verification (OCV) confirms that printed text matches expected content. Label verification systems check lot numbers, expiry dates, serial numbers, product codes, and regulatory text on packaging lines.
In pharmaceutical applications, Motionwell integrates label verification with serialization databases to ensure every unit carries a unique, verified identifier. A red strobe lighting configuration is commonly used for reliable character contrast on reflective packaging materials.
Vision-Guided Assembly and Robot Alignment
The camera locates parts and communicates position offsets to a robot or servo stage for alignment. Vision-guided assembly eliminates the need for precise fixturing — the robot compensates for part placement variation in real time.
Motionwell has delivered vision-guided SCARA assembly systems achieving ±0.01 mm repeat accuracy for electronics panel assembly and medical device component insertion. The camera-robot calibration procedure is critical: we use multi-point calibration with verification runs to ensure accuracy is maintained across the working envelope.
Hardware Comparison: Leading Vision Platforms
Motionwell integrates multiple vision hardware platforms. The selection depends on application complexity, budget, and existing factory infrastructure.
| Platform | Strengths | Best fit applications | Typical price range (SGD) |
|---|---|---|---|
| Keyence IV3 | Simple setup, built-in lighting, fast deployment | Presence/absence, simple pass/fail, packaging inspection | 3,000-8,000 per sensor |
| Keyence CV-X | Multi-camera support, high-speed processing, large image library | Multi-point inspection, high-speed lines, complex pattern matching | 15,000-40,000 per system |
| Cognex In-Sight | AI-based defect classification (ViDi), spreadsheet programming | Complex defect detection, variable products, AI-assisted classification | 10,000-35,000 per system |
| Basler ace / ace 2 | High resolution, GigE/USB3, open software ecosystem | Custom measurement, high-resolution imaging, R&D and OEM integration | 2,000-5,000 per camera (+ software) |
| HIKrobot | Cost-effective, wide model range, growing industrial adoption | Budget-conscious projects, standard inspection, volume applications | 1,500-4,000 per camera |
Keyence IV3 and CV-X are Motionwell’s most frequently deployed platforms for production inspection. The IV3’s built-in lighting and one-click setup make it practical for simple inspection points. The CV-X handles complex multi-camera configurations with up to 8 cameras on a single controller.
Cognex In-Sight with ViDi AI tools is selected when defects are visually complex or variable — situations where traditional rule-based algorithms struggle with false rejects. The AI model trains on example images rather than explicit rules.
Basler and HIKrobot industrial cameras are used when the application requires custom optics, non-standard resolution, or integration with third-party image processing software (HALCON, OpenCV, NI Vision).
Applications by Industry
Medical Devices
Motionwell’s medical device vision systems include BD syringe assembly inspection (projects P22068, P23003) where cameras verify component presence, orientation, and dimensional conformance at each station of a 12-station rotary assembly machine. Inspection criteria include needle hub seating depth, cap alignment, and label position.
Medical vision systems require GR&R (Gauge Repeatability and Reproducibility) validation studies to demonstrate measurement system capability. Motionwell delivers GR&R reports as part of every medical device inspection project.
Electronics and Semiconductor
For ifm’s panel assembly line (P23045), Motionwell integrated ABB SCARA robots with Keyence vision for component alignment at ±0.01 mm accuracy. The vision system locates panel datums, communicates offsets to the SCARA controller, and verifies assembly completion after each operation.
Electronics inspection commonly includes solder joint inspection, connector pin alignment, PCB dimensional verification, and component presence detection. See our electronics and semiconductor solutions for project details.
Pharmaceutical Packaging
Pharmaceutical serialization requires every unit to carry a unique 2D Data Matrix code that is verified for print quality (ISO/IEC 15415 Grade C minimum) and data accuracy. Motionwell integrates vision stations with serialization software to verify codes, lot numbers, and expiry dates on blister packs, bottles, and cartons.
Red strobe lighting at 45-degree incidence provides consistent contrast on glossy pharmaceutical packaging. The vision system rejects any unit with an unreadable code or incorrect data, maintaining 100% traceability as required by pharmaceutical packaging regulations.
Food and Beverage
Vision inspection on food and beverage packaging lines verifies fill levels, cap seal integrity, label placement, date codes, and package completeness. Line speeds of 200-600 packages per minute demand high-speed cameras with minimal latency.
How to Select a Vision System
Step 1: Define the Inspection Task
Before evaluating cameras, define exactly what the system must detect or measure. Collect 50-100 sample parts including known good parts and parts with each defect type. Document the smallest defect or tightest tolerance the system must reliably catch.
Step 2: Determine Resolution Requirements
The minimum resolution depends on the smallest feature you need to detect. A general rule: you need at least 4-5 pixels across the smallest defect to reliably detect it.
| Field of view | Smallest defect | Minimum resolution needed |
|---|---|---|
| 100 x 75 mm | 0.1 mm | 5 MP (2448 x 2048) |
| 200 x 150 mm | 0.2 mm | 5 MP |
| 200 x 150 mm | 0.1 mm | 12 MP (4096 x 3000) |
| 400 x 300 mm | 0.5 mm | 5 MP |
| 400 x 300 mm | 0.1 mm | 20 MP+ |
Step 3: Design the Lighting
Lighting design is the most critical and most underestimated step. The same defect can be invisible or obvious depending on the light angle, type, and color.
| Lighting technique | Best for | How it works |
|---|---|---|
| Diffuse (dome/cloudy day) | Uniform illumination, reducing glare | Light from all angles; minimizes surface reflections |
| Directional (bar/ring at angle) | Surface defects, scratches, texture | Light at controlled angle creates shadows at defect edges |
| Dark-field (low angle) | Surface scratches, raised features | Near-horizontal light; only defects scatter light to camera |
| Backlight | Silhouette measurement, edge detection | Light behind part; camera sees outline only |
| Structured light | 3D profiling, height measurement | Projected pattern deforms on surface; triangulation computes height |
| Coaxial | Flat reflective surfaces, wafer inspection | Light along camera axis; specular surfaces appear bright |
Step 4: Select Processing Speed
Vision processing time must be shorter than the available inspection window. If parts move on a conveyor at 30 parts/minute, the system has 2 seconds per part. Subtract trigger latency, image acquisition time, and communication time to determine the available processing budget.
Step 5: Plan PLC and Data Integration
The vision system communicates inspection results to the line PLC for reject control, and optionally to MES/SCADA for production data logging. Standard communication protocols include EtherNet/IP, PROFINET, Modbus TCP, and discrete I/O.
Motionwell designs the PLC integration so that every inspection result is linked to a part identifier (serial number, lot code, or sequence counter) for traceability. This is essential in regulated industries.
Integration with PLC and Robots
Machine vision systems do not operate in isolation. They are part of a production line controlled by PLCs and robots. The vision system must trigger at the right moment, process within the cycle time budget, and communicate results to downstream equipment.
| Integration point | Protocol | What gets communicated |
|---|---|---|
| PLC trigger and result | EtherNet/IP or discrete I/O | Trigger signal, pass/fail result, measurement values |
| Robot offset data | TCP/IP socket or dedicated protocol | X, Y, theta offsets for vision-guided alignment |
| MES/SCADA data logging | OPC UA or database write | Inspection images, measurement data, statistical summaries |
| Reject mechanism | Discrete I/O from PLC | Diverter, pusher, or robot sorting command |
Allen-Bradley CompactLogix and Siemens S7-1500 are the PLC platforms Motionwell most frequently integrates with vision systems. The PLC handles line sequencing, safety interlocks, and reject control while the vision system focuses on image processing.
Motionwell’s Machine Vision Expertise
Motionwell is a machine vision system integrator based in Singapore with delivered projects across medical devices, electronics, pharmaceutical packaging, and precision manufacturing. We handle the full scope: feasibility testing with your actual parts, hardware selection, lighting design, algorithm development, PLC integration, and validation.
Our approach starts with a lighting feasibility study. We test your parts under multiple lighting configurations in our lab to confirm that the target defects or features are reliably detectable before any system is specified. This avoids the most common failure mode in vision projects — buying a system that works in the demo but fails with real production variation.
For detailed information on our vision integration approach and project references, contact the Motionwell engineering team.
Getting Started
If you are evaluating machine vision inspection for your production line, the most productive first step is a feasibility study with your actual production parts. Contact Motionwell with:
- Sample parts (5-10 good parts, 5-10 parts with each defect type)
- Current inspection criteria and tolerance specifications
- Line speed and cycle time requirements
- Existing PLC platform and communication infrastructure
We will test your parts, recommend a system configuration, and provide a budget estimate before you commit to a project.