1. Overview

1.1 Importance of Preform Inspection

A preform is the intermediate product of a PET bottle, which is ultimately formed into various specifications of packaging bottles through a blow molding process. As a "semi-finished product" of PET bottles, the quality of the preform directly determines the performance of the final product. A defective preform can lead to:

• Failure in blow molding, reducing production efficiency

• Structural defects in the finished bottle, affecting usage safety

• Insufficient sealing performance, causing product leakage and deterioration

• Appearance defects, damaging brand image and market acceptance

In the quality control chain, preform inspection belongs to the "front-end prevention" link. Compared with finished bottle inspection, preform inspection offers higher economic benefits and efficiency advantages, allowing for the timely removal of unqualified products before defects are amplified, thereby avoiding resource waste in subsequent processing.

1.2 Technical Advantages of Vision Inspection

Traditional preform inspection mainly relies on manual visual inspection, which has inherent flaws such as low efficiency, poor consistency, strong subjectivity, and high labor intensity. With the development of machine vision technology, automated vision inspection systems have gradually become mainstream, with their core advantages including:

Inspection Precision: Capable of identifying subtle defects as small as 0.1mm

Inspection Speed: Up to 20-30 preforms per second

Consistency: Unaffected by human subjective factors

Comprehensive Information: Can detect multiple types of defects simultaneously

Data Traceability: Digital storage of inspection results, facilitating quality traceability and analysis

2. Components of a Preform Vision Inspection System

2.1 Hardware System Design

2.1.1 Imaging Unit

Camera Selection: Industrial-grade CCD or CMOS cameras are typically used, with resolutions ranging from 2 to 12 megapixels. Based on inspection needs, choices include:

• Area Scan Cameras: For overall appearance inspection

• Line Scan Cameras: For high-speed rotational inspection

• 3D Cameras: For dimensional and morphological inspection

Lens Configuration: Industrial lenses with appropriate focal lengths are selected based on field of view and working distance. Common configurations include:

• Telecentric Lenses: Eliminate perspective errors, improving dimensional measurement accuracy

• Macro Lenses: For high-resolution imaging of detailed areas

• Zoom Lenses: Adapt to preforms of different specifications

Lighting System: Lighting is a critical part of vision inspection. For the transparent characteristics of preforms, commonly used lighting solutions include:

• Coaxial Lighting: For detecting surface scratches and stains

• Backlighting: For contour and wall thickness uniformity inspection

• Dome Lighting: Eliminates reflections, used for thread and bottle mouth inspection

• Special Structured Light: For 3D morphological inspection

2.1.2 Mechanical Conveyance System

Feeding Mechanism: Vibratory bowls, conveyor belts, robotic arms, etc.

Positioning Mechanism: Servo motor-driven rotary tables, precision positioning fixtures

Sorting Mechanism: Pneumatic or mechanical rejection devices

Control System: PLC or industrial PC, coordinating the actions of various components

2.2 Software System Architecture

2.2.1 Image Processing Algorithms

Image Preprocessing:

• Filtering and Noise Reduction: Eliminate imaging noise

• Contrast Enhancement: Highlight target features

• Image Segmentation: Extract regions of interest

• Morphological Processing: Fill holes, remove burrs

Feature Extraction Algorithms:

• Edge Detection: Canny, Sobel operators

• Texture Analysis: Gray-level co-occurrence matrix

• Color Analysis: RGB/HSV space conversion

• Shape Analysis: Hough transform, contour matching

Defect Recognition Algorithms:

• Threshold Segmentation: Based on grayscale differences

• Template Matching: Compare with standard samples

• Machine Learning: Support Vector Machines, Random Forests

• Deep Learning: Convolutional Neural Networks

2.2.2 System Control Software

Inspection Process Management: Coordinates the entire process of image acquisition, processing, judgment, and sorting

Parameter Setting Interface: Provides a user-friendly interface for parameter configuration

Data Management System: Storage, query, statistics, and export of inspection results

Alarm and Prompt: Timely alerts for abnormal status, guiding maintenance

3. Main Inspection Items and Technical Requirements

3.1 Appearance Defect Inspection

3.1.1 Surface Defects

Scratch Detection: Linear defects with length > 0.5mm, depth > 0.05mm

Black Spot Impurities: Foreign matter contamination with diameter > 0.3mm

Bubbles: Bubbles with diameter > 0.5mm, density > 3/cm²

Silver Streaks: Micro-cracks caused by stress or humidity

Haze: Transparency not meeting requirements

Detection Challenges: Transparent materials are sensitive to light, with significant surface reflection interference. Solutions include using polarized light, multi-angle imaging, and diffuse reflection lighting.

3.1.2 Color Abnormalities

Color Difference Detection: ΔE value > 1.5 compared to standard color chart

Color Streaks: Uneven color distribution

Discoloration: Caused by raw material decomposition or contamination

Detection Method: High-precision color cameras capture images under standard light sources (D65, D50, etc.), with color space conversion and color difference formulas for calculation.

3.2 Dimensional Accuracy Inspection

3.2.1 Critical Dimensions

Total Height: Tolerance ±0.3mm

Bottle Mouth Outer Diameter: Tolerance ±0.1mm

Thread Dimensions: Pitch, tooth profile, completeness

Neck Dimensions: Support ring diameter, position

Weight: Weight deviation < 0.5g

3.2.2 Geometric Tolerances

Concentricity: Coaxiality of bottle mouth and body < 0.2mm

Verticality: Straightness of bottle body

Wall Thickness Uniformity: Wall thickness variation at different positions < 0.1mm

Detection Technology: 3D vision systems can accurately measure various parts of the preform, obtaining three-dimensional dimensional information through point cloud data processing.

3.3 Structural Integrity Inspection

3.3.1 Bottle Mouth Defects

Missing Threads: Missing or broken threads

Bottle Mouth Deformation: Ovality out of tolerance

Support Ring Defects: Incomplete, burrs

Sealing Surface Defects: Scratches, dents

3.3.2 Bottle Body Defects

Wall Thickness Abnormalities: Locally too thin or too thick

Stress Whitening: Uneven molecular orientation

Gate Defects: Residual sprue, shrinkage marks

Parting Line: Too obvious or misaligned

3.4 Special Defect Inspection

3.4.1 Raw Material-Related Defects

Foreign Material Inclusion: Impurities of different materials

Excessive Moisture Content: Bubbles generated during injection molding

Degradation Products: Raw material overheating decomposition

3.4.2 Process-Related Defects

Short Shot: Insufficient injection pressure causing material shortage

Flash: Excessive mold gap causing overflow

Sink Marks: Uneven cooling shrinkage

Weld Lines: Weakness at melt convergence points

4. Inspection Algorithms and Technological Innovation

4.1 Traditional Image Processing Techniques

4.1.1 Threshold-Based Segmentation

Separate defect areas from the background by setting appropriate grayscale thresholds. Suitable for defects with obvious contrast, such as black spots and bubbles.

Limitations: Sensitive to lighting changes, difficult to adapt to complex backgrounds.

4.1.2 Template Matching

Use a standard preform image as a template, perform correlation matching with the image to be inspected, and calculate the difference.

Improvement Methods: Multi-template matching, deformable templates, local feature matching.

4.1.3 Texture Analysis

Detect surface defects by analyzing image texture features, suitable for scratches, haze, etc.

Common Features: Contrast, correlation, energy, uniformity.

4.2 Machine Learning Approaches

4.2.1 Feature Engineering

Extract shape, texture, color, and other features from images to construct feature vectors for training classifiers.

Common Classifiers: Support Vector Machines, Random Forests, AdaBoost.

4.2.2 Transfer Learning

Use models pre-trained on large-scale datasets, fine-tune them for preform inspection tasks, reducing the need for training data.

4.3 Deep Learning Methods

4.3.1 Convolutional Neural Networks (CNN)

Detection Frameworks:

• Faster R-CNN: Two-stage detection, high accuracy

• YOLO Series: Single-stage detection, fast speed

• SSD: Balances speed and accuracy

Application Scenarios: Suitable for general detection of multiple defects, especially those difficult to define with traditional methods.

4.3.2 Generative Adversarial Networks (GAN)

Defect Synthesis: Generate various defect samples to expand the training dataset

Anomaly Detection: Determine abnormalities through reconstruction errors

4.3.3 Attention Mechanism

Enable the network to focus on key areas, improving detection efficiency and accuracy, especially for small target defect detection.

4.4 Multi-Sensor Fusion

4.4.1 Vision + 3D

2D vision detects appearance defects, 3D vision detects dimensional and morphological defects, complementing each other.

4.4.2 Vision + Spectroscopy

Combine visible light with near-infrared spectroscopy to detect internal characteristics such as raw material purity and moisture content.

5. System Implementation and Optimization

5.1 Implementation Steps

5.1.1 Requirements Analysis

• Define inspection objectives: Defect types, inspection standards

• Determine performance indicators: Detection rate, false positive rate, inspection speed

• Assess on-site conditions: Space, lighting, power supply, air source

5.1.2 System Design

• Hardware Selection: Cameras, lenses, lighting, mechanical structure

• Software Architecture: Image processing algorithms, control logic, human-machine interface

• Integration Plan: Method of interfacing with the production line

5.1.3 Installation and Commissioning

• Mechanical Installation: Ensure positioning accuracy

• Optical Commissioning: Optimize lighting and imaging quality

• Parameter Calibration: Establish correspondence between pixels and actual dimensions

• Algorithm Optimization: Adjust detection parameters for optimal performance

5.1.4 Verification and Acceptance

• Performance Testing: Test detection rate using standard samples

• Stability Testing: Continuously run the system to test reliability

• User Training: Operation and maintenance training

5.2 Key Parameter Optimization

5.2.1 Lighting Optimization

• Angle: Incident angle, reflection angle

• Intensity: Avoid overexposure or underexposure

• Uniformity: Eliminate uneven lighting

• Spectrum: Match camera and material characteristics

5.2.2 Camera Parameters

• Resolution: Balance between field of view and accuracy

• Exposure Time: Avoid motion blur

• Gain: Balance signal-to-noise ratio

• Frame Rate: Meet inspection cycle requirements

5.2.3 Algorithm Parameters

• Thresholds: Segmentation thresholds

• Filtering Parameters: Noise reduction intensity

• Feature Parameters: Weights of shape, texture features

• Classification Thresholds: Discrimination thresholds

5.3 System Maintenance

5.3.1 Daily Maintenance

• Cleaning: Lenses, lighting, protective windows

• Calibration: Regular dimensional calibration

• Backup: Parameter and program backup

• Logging: Operation logs, defect statistics

5.3.2 Periodic Calibration

• Performance Verification: Verify detection rate using standard samples

• Parameter Optimization: Adjust parameters based on production conditions

• Software Upgrades: Update algorithms and functions

• Hardware Inspection: Check the status of components

6. Industry Applications and Challenges

6.1 Beverage Packaging Industry

Application Characteristics: High speed, high precision, multiple varieties

Special Requirements: Food-grade materials, compliance with FDA, EFSA certifications

Development Trends: Lightweight preform inspection, compatibility with recyclable materials

6.2 Cosmetic Packaging

Quality Requirements: Perfect appearance, accurate color

Inspection Challenges: Multiple colors, shapes, transparencies

Development Trends: Personalized preform inspection, adaptation to small batch, multiple varieties

6.3 Pharmaceutical Packaging

Regulatory Requirements: Compliance with GMP, pharmacopoeia standards

Special Inspections: Seal integrity, cleanliness, material consistency

Development Trends: Sterile packaging inspection, integration with traceability systems

6.4 Technical Challenges

6.4.1 Imaging of Transparent Materials

• Reflection and refraction interference

• Difficulty imaging internal structures

• Solutions: Polarized light, multi-angle imaging, special coatings

6.4.2 High-Speed Inspection

• High frame rate imaging

• Real-time processing

• Solutions: Hardware acceleration, parallel computing, algorithm optimization

6.4.3 Complex Defects

• Micro-defects

• Defects with blurred boundaries

• Solutions: High-resolution imaging, deep learning, multi-scale analysis

6.4.4 Adaptability

• Quick switching between multiple varieties

• Adaptive parameter adjustment

• Solutions: Recipe management, automatic calibration, online learning

7. Future Development Trends

7.1 Intelligent Upgrades

Adaptive Inspection: System can automatically adjust parameters based on production conditions

Predictive Maintenance: Analyze equipment status based on inspection data, predict failures

Online Learning: System can learn from new samples, continuously optimizing models

7.2 Integrated Development

Deep Integration with Production Lines: Inspection data directly feedbacks to control injection molding parameters

Quality Big Data: Massive inspection data used for process optimization and quality analysis

Digital Twin: Establish virtual inspection systems for prediction and optimization

7.3 Integration of New Technologies

Hyperspectral Imaging: Simultaneously acquire spatial and spectral information

Terahertz Technology: Detect internal defects and delamination

Quantum Imaging: Breakthrough traditional optical limits

Edge Computing: Processing completed at the device end, reducing latency

7.4 Standardization and Modularization

Interface Standardization: Facilitates system integration and upgrades

Modular Design: Flexible configuration based on needs

Cloud Services: Algorithm updates, remote diagnostics, data analysis

8. Economic Benefit Analysis

8.1 Direct Benefits

Quality Improvement: Defect rate reduced by 30%-70%

Cost Savings: Labor reduced by over 80%, lower rework and scrap losses

Efficiency Increase: Inspection speed increased by 3-5 times, capable of 7×24 operation

8.2 Indirect Benefits

Data Value: Quality data used for process optimization

Brand Protection: Avoid brand damage caused by quality issues

Compliance Assurance: Meet increasingly stringent quality regulations

Technology Reserves: Accumulate core inspection technology

8.3 Return on Investment

Investment Cost: System prices range from hundreds of thousands to millions

Payback Period: Typically 6-18 months, depending on production scale

Long-term Value: Technology upgrades, data accumulation, brand premium

9. Implementation Recommendations

9.1 Phased Implementation Strategy

1. Pilot Phase: Pilot at key stations to verify effectiveness

2. Promotion Phase: Summarize experience, gradually extend to other production lines

3. Integration Phase: Integrate with MES, ERP systems for data sharing

4. Optimization Phase: Continuously optimize algorithms and parameters to improve performance

9.2 Key Success Factors

Management Support: Strategic emphasis, resource guarantee

Cross-Department Collaboration: Coordination among production, quality, equipment, and IT departments

Supplier Selection: Technical strength, industry experience, service capability

Personnel Training: Cultivate operation, maintenance, and data analysis capabilities

Continuous Improvement: Establish optimization mechanisms to adapt to changing needs

9.3 Risk Control

Technical Risks: Immature technology, poor adaptability

Implementation Risks: Delayed timelines, ineffective results

Operational Risks: Difficult maintenance, frequent failures

Countermeasures: Thorough verification, phased implementation, backup plans, professional support

10. Conclusion

Preform vision inspection technology is evolving from traditional single-function inspection to intelligent, integrated, and networked systems. With the integration and application of new technologies such as artificial intelligence, the Internet of Things, and big data, vision inspection systems can not only achieve more precise defect recognition but also provide data support for production process optimization, becoming an important component of intelligent manufacturing.

For PET packaging manufacturers, investing in advanced preform vision inspection systems can not only improve product quality and reduce production costs but also serve as a crucial step in digital transformation and enhancing core competitiveness. In the future, as inspection accuracy continues to improve, inspection speed increases, and system adaptability strengthens, vision inspection technology will undoubtedly play an increasingly important role in preform quality control, driving the entire industry to a higher level of development.

When selecting and implementing preform vision inspection systems, companies should consider their actual needs, comprehensively evaluate from aspects such as technological advancement, system stability, return on investment, and service support, adopt scientific and reasonable implementation strategies, and ensure that the system can truly add value and create sustainable competitiveness for the enterprise.