What Is a Visual Quality Inspection System
A visual quality inspection system is an automated quality control framework that uses an imaging system composed of industrial cameras, optical lenses, and illumination light sources, combined with image processing algorithms, to perform defect detection, assembly verification, and information identification on production lines. Its core value lies in converting the uncertainty of manual visual inspection into a quantifiable, traceable, and repeatable digital inspection process.
Defect Types and Detection Difficulty Comparison
Different defect types impose significantly varying requirements on imaging systems — this is the logical starting point for selection.
| Defect Type | Typical Characteristics | Detection Challenges | Recommended Imaging Method | Core Parameter Requirements |
|---|---|---|---|---|
| Surface Scratches | Depth 0.01–0.5mm, varying widths; easier to hide on reflective surfaces | Confused with normal textures; low contrast | Coaxial light + low-angle dark field | Resolution ≥ 5μm/pixel, contrast ≥ 30% |
| Missing Components | Components fully absent or position-deviated | Background interference; feature region identification | Backlight/ring light + front fill | Full FOV coverage, positioning accuracy ±0.5mm |
| Misassembly/Wrong Orientation | Component present but direction/position incorrect | Confused with correct assembly; subtle feature differences | Multi-angle light combination | Multi-camera coverage; feature extraction algorithm adaptation |
| Code/Character Recognition | 1D barcodes, 2D barcodes, OCR | Code damage, low contrast, curved surface interference | High-brightness uniform light + polarization | Resolution ≥ 0.1mm/pixel, depth of field covering product height variation |
The challenge of missing component detection lies in eliminating background interference. On hardware and plastic parts production lines common in Southeast Asian factories, workpiece surfaces often have oil residue, dust, or machining textures. These background features can easily be confused with the shadows appearing when components are missing. The typical solution is backlight imaging, where missing component areas form dark voids against a bright background, enabling precise identification through threshold segmentation algorithms.
Matching Method for Cameras, Lenses, and Light Sources — The Three Key Elements
Camera Selection: Applicable Scenarios for Area Scan vs. Line Scan
The core basis for industrial camera selection is motion state and detection precision requirements.
Area scan cameras are suitable for stationary or low-speed cycle scenarios (cycle time ≥ 1 second/piece). Their advantage lies in capturing the complete field of view in a single exposure, with relatively straightforward algorithm processing. Taking the CVM series area scan cameras from HuaRay Machine Vision as an objective example, mainstream resolutions range from 2MP to 29MP, with pixel sizes covering 3.45μm–5.86μm, capable of meeting various requirements from general inspection to high-precision measurement.
Line scan cameras are suitable for high-speed continuous production lines (cycle time < 1 second/piece) or large-area inspection scenarios. Their working principle is row-by-row scanning, requiring relative motion between the object under test and the camera. The advantage of line scan cameras lies in theoretically unlimited longitudinal resolution (dependent on line rate and motion speed), making them suitable for inspecting continuous materials such as coils, sheets, and bottle bodies. During selection, attention should be paid to the product of line rate (kHz) and resolution, which determines the imaging data volume per unit time.
Selection criterion can be simplified as: area scan for stationary products or relaxed cycle times; line scan for high-speed motion or ultra-long spans.
Lens Parameters: The Triangle Trade-off of Resolution, Depth of Field, and Working Distance
The essence of lens selection lies in finding balance among resolution, depth of field, and working distance.
Resolution is jointly determined by the lens MTF (Modulation Transfer Function) curve and camera pixel size. The empirical formula is: lens resolution (lp/mm) should be ≥ half the reciprocal of the camera pixel size (μm). For example, for a camera with 3.45μm pixel size, the effective lens resolution should be ≥ 145 lp/mm. Lenses below this standard will waste camera resolution.
Depth of field determines the tolerable movement range of objects along the optical axis. The calculation formula is: DOF = 2 × permissible circle of confusion diameter × (F-number) × reciprocal of magnification squared. For workpieces with significant surface variation (such as curved surfaces, stepped structural parts), sufficient depth of field margin needs to be reserved. It is generally recommended that actual depth of field be 2–3 times the product height variation.
Working distance affects equipment layout and protection design. Stamping workshops and welding workshops common in Southeast Asian factories have complex environments; overly close working distances may cause lens contamination or collision risks. It is recommended to prioritize lens configurations with working distance ≥ 100mm while meeting resolution and depth of field requirements.
Practical steps: Step 1, calculate required pixel resolution based on defect detection precision requirements; Step 2, calculate magnification based on field of view range; Step 3, calculate required depth of field based on product height variation; Step 4, select the model with the most relaxed working distance among lenses meeting the first three criteria.
Light Source Types: Combinatorial Logic of Angle, Color Temperature, and Uniformity
The light source is the most easily overlooked yet most impactful component on imaging results in a visual system.
Coaxial light (light source coaxial with the lens incident beam) is characterized by eliminating reflections and highlighting surface defects, suitable for scratch detection on high-reflectivity materials such as metals, glass, and lenses. Its drawback is limited illumination coverage; multi-light source stitching is required for large field-of-view scenarios.
Ring light illuminates from all sides at a certain angle, producing brightness-darkness contrast effects, suitable for contour extraction and edge detection. The lower the angle (closer to horizontal), the more pronounced the shadow effect, but specular reflection is more likely to occur; the higher the angle (closer to vertical), the more uniform the illumination but the lower the contrast.
Backlight illuminates from behind the object under test, presenting transparent or semi-transparent defects (such as bubbles, cracks) as high-contrast images, and is also commonly used for silhouette imaging in missing component detection.
Color temperature selection affects defect detection rates. White light sources (5500K–6500K) are suitable for general inspection; blue light sources (470nm) can enhance contrast of metal scratches and cracks; red light sources (630nm) can suppress ambient light interference, suitable for outdoor or semi-outdoor scenarios (such as photovoltaic module inspection).
Light source uniformity is an easily overlooked indicator. Non-uniform light sources produce gradient backgrounds in images, interfering with algorithm defect identification. It is recommended to use an integrating sphere to calibrate light source uniformity; brightness variation within the field of view should be controlled within ±5%.
Trade-off Strategy Between Cycle Time and False Rejection
Cycle time and false rejection rate (False Reject Rate / False Pass Rate) are a core contradiction in visual quality inspection systems.
Cycle time constraints determine the upper limits of image acquisition time and algorithm processing time. Taking a 2-second cycle as an example, if 0.3 seconds is reserved for mechanism actions, the total time window for image acquisition and processing is 1.7 seconds. Under this constraint, high-resolution image (29MP) acquisition and transmission may occupy 0.5–1 second, leaving a very limited time window for algorithm processing.
False rejection types are divided into two categories: missed defects (False Pass — defective products are released) and false rejections (False Reject — normal products are rejected). Missed defects directly affect customer complaints and recall risks; false rejections increase rework costs and production capacity loss. Different industries have varying tolerances for both types: the automotive parts industry typically requires missed defect rate < 0.1% and false rejection rate < 5%; the consumer electronics industry has slightly higher tolerance for false rejection rates, but equally stringent requirements for missed defect rates.
Trade-off methodology: First, clarify the quality risk level of the detection scenario — safety-related parts (such as braking system components) prioritize controlling missed defect rates and can accept higher false rejection rates; appearance parts prioritize controlling false rejection rates and can accept slight missed defect rates. Second, adopt a multi-level detection strategy at the algorithm level: the first level uses lightweight algorithms to quickly filter obviously normal products; the second level uses high-precision algorithms for secondary review of suspicious products. A two-level strategy can compress overall processing time by 30%–50% while maintaining detection accuracy.
Practical recommendation: Establish a detection database for each production line, recording each product's inspection images, judgment results, and backend final inspection/customer complaint feedback. Regularly (such as weekly) use actual data to calibrate algorithm thresholds, avoiding systematic bias caused by fixed thresholds.
Special Considerations for Southeast Asian Factory Deployment
When deploying visual quality inspection systems in Southeast Asian factories (Thailand, Vietnam, Indonesia, Malaysia, etc.), the following environmental factors require additional attention:
Temperature and humidity: Southeast Asian climates are high-temperature and high-humidity; equipment should select industrial-grade cameras with operating temperature range ≥ 50°C, with anti-fog treatment for lenses (nitrogen-sealed or with self-heating modules). Light source heat dissipation also needs to be considered; continuously operating LED light sources may require active cooling in 40°C ambient temperatures.
Voltage fluctuation is relatively common in some Southeast Asian industrial parks. It is recommended to configure voltage stabilization power supplies or UPS for visual systems, avoiding camera restarts or light source flickering caused by voltage drops.
Dust and vibration are particularly prominent in metal processing workshops. Lenses need dust covers or airflow purging devices; camera installation positions should be kept away from large vibration sources (such as stamping equipment), or vibration-damping mounts should be used.
Local service response is an easily overlooked aspect for overseas enterprises. The visual system debugging phase often requires on-site parameter optimization. It is recommended to select suppliers with local technical teams in the region, or sign service agreements including on-site debugging. Subject to actual project, it is recommended to specify response time (such as 48-hour on-site arrival) and debugging man-hour limits in contracts.
Practical Recommendations: Five Steps to Complete Visual Quality Inspection System Selection
Step 1: Defect Type and Precision Definition
Align with process and quality teams on defect types to be inspected, missed defect tolerance, and false rejection tolerance. Quantify detection precision requirements as "minimum detectable defect size." For example: "Scratches with width ≥ 0.1mm and length ≥ 2mm must be 100% detected."
Step 2: Field of View and Cycle Time Calculation
Determine the field of view (FOV) required for single inspection based on product dimensions and installation space. Calculate the time budget for image acquisition and processing based on production line cycle time requirements (such as 3 seconds/piece), to constrain camera resolution and algorithm complexity.
Step 3: Camera and Lens Parameter Matching
Calculate magnification based on field of view and precision requirements (Magnification = camera field of view / sensor size). Calculate required pixel count based on magnification and precision requirements (Pixel Number = defect size / 0.3). Select combinations where both camera resolution and lens parameters meet requirements.
Step 4: Light Source Solution Testing
In a laboratory environment, use standard defect samples to perform imaging comparison tests on different light source types, angles, and color temperatures. Select the configuration with the highest defect contrast and best uniformity. It is recommended to retain 2–3 alternative solutions for on-site verification.
Step 5: Algorithm and Threshold Calibration
During the production line trial run phase, collect 500–1000 samples of actual products (including defective and normal items) to perform offline training and threshold optimization on algorithms. Establish a detection database to accumulate data foundations for subsequent continuous improvement.
FAQ
Q: What are the advantages of visual inspection systems compared to manual visual inspection?
A: The core advantages of visual inspection systems are stability and traceability. Manual visual inspection is susceptible to fatigue, emotions, and experience differences, with considerable fluctuation in missed defect rates; visual systems apply consistent inspection standards to every product, and inspection images can be archived for review, suitable for customers or regulatory scenarios with traceability requirements for quality data.
Q: How long does production line downtime need to be for adding visual systems to existing production lines?
A: Subject to actual project, the modification period depends on production line complexity and number of inspection stations. Typical modification流程包括：现场勘测（1天）、方案确认（2–3天）、设备安装与布线（1–2天）、调试与试运行（3–5天）。整体改造周期通常在1–2周内，建议选择生产淡季实施。 The typical modification process includes: on-site survey (1 day), solution confirmation (2–3 days), equipment installation and wiring (1–2 days), debugging and trial operation (3–5 days). The overall modification period is typically within 1–2 weeks. It is recommended to implement during the low-production season.
Q: How to determine if a visual quality inspection solution is suitable for my product?
A: Preliminary judgment can be made through two dimensions: "minimum detectable defect size" and "cycle time requirements." If defect size ≥ 0.1mm and cycle time ≥ 1 second/piece, visual system selection difficulty is relatively low; if defect size < 0.05mm or cycle time < 0.5 second/piece, higher-specification hardware configuration and more refined algorithm optimization are required. It is recommended to engage in in-depth technical exchange with visual suppliers.
Q: What is the lifespan of light sources, and do they need regular replacement?
A: The typical lifespan of industrial-grade LED light sources is 30,000–50,000 hours (under standard operating temperature). Actual lifespan is affected by ambient temperature and switching frequency; high-temperature environments significantly shorten lifespan. It is recommended to establish a light source usage time ledger and proactively replace at 80% of expected lifespan, avoiding detection failures caused by light source attenuation during production.
Q: How to handle disputes regarding visual system inspection results?
A: It is recommended to establish a "dispute handling process": for products judged as defective by the visual system, retain inspection images and judgment basis for manual review by quality engineers. If the review result is consistent with the system judgment, process as defective products; if the review result is normal products, feedback to the visual team to analyze the cause of false rejection and adjust algorithm thresholds or optimize light source configuration.