Success Story

Quality control of plastic components using industrial vision and AI

20230831 082516 Copy

The traditional approach to quality control using industrial vision and AI technology allows the system to detect both predictable and unpredictable anomalies across every single piece on the production line. Quality assessment is based on predefined criteria, standardizing defect classification and the required quality levels.

 

The fully automated industrial vision solution adheres to the footprint and cycle time constraints of the existing production line. It ensures consistent high-quality results and frees operators to focus on other, higher added value tasks.

Inea Cs Downloadpdf Inea Download PDF

Company profile

Industry:
Plastika Skaza is a Slovenian company with 45 years of experience in the field of plastics processing and injection moulding.

Product portfolio:
The company specializes in the production of plastic industrial components and housings, as well as innovative sustainable materials.

Location:
A Slovenian company, mostly exporting to Europe.

Impact

50 mil verified parameters

  • 100% quality control of all products
  • 9 scans per inspected product
  • 0% affected production cycle
Inea Inclass Raster Circle 100px

Blaž Ojsteršek

Head of Production Optimization, Packaging, and Automation

“INEA’s technology allows us to perform fully automated quality checks. In just a few seconds, the AI system detects the most complex defects and systematically categorizes them, relieving our operators to focus on other important tasks.”

Challenge

Plastika Skaza decided to integrate industrial vision quality assessment into their existing production line. Their objective was to establish a fully automated production process to enhance efficiency and production speed, reduce the number of defective products, minimize waste and labour costs, and ultimately boost the company’s competitiveness. By implementing industrial vision, they aimed to conduct comprehensive quality control across the entire production line, particularly focusing on plastic housing components for gas flow meters, which are exported to a top-tier client in the gas industry.

The production cycle had spatial and time constraints, meaning the new solution could not extend the production cycle and had to be strategically integrated into the existing production cell. Classifying defects on moulded housings presented another challenge due to an extensive and continually growing defect list. Additionally, many of these deformities were considered analytically non-deterministic, further complicating the classification process.

Image3

Solution

Consistency in image capture

High-performance cameras capture detailed images of each part from multiple angles, with precise handling by a robotic gripper. The part is uniformly illuminated by industrial-grade LED lights during the imaging process, ensuring consistent lighting and optimal conditions for accurate and repeatable image capture.
Synchronized
operation

The system analyses image contrast simultaneously across the entire image and in specific sections according to preset parameters. Based on these classification results, the synchronized robotic system accurately sorts the part, directing it either to the appropriate line for acceptable items or to a designated line for reclassification.
Deep learning
predictability

In addition to traditional image analysis, the complexity of the part requires the use of advanced deep learning techniques. These methods predict both known deviations and previously unidentified anomalies by recognizing error patterns within defined tolerance thresholds.

Our industrial vision solution has automated the quality control process for plastic housing components, enabling the system to inspect the quality of each piece during the production cycle. The quality control solution has been seamlessly integrated into the existing robotic cell of the production line. Despite adding an extra quality assessment step, the production cycle time remains unchanged, and productivity is unaffected.

Image3

User interface screen for tracking settings and quality history. Users can view defects and their specific locations on the last twenty products inspected. The interface displays the status of individual process phases and allows for manual activation or deactivation of specific functions.

User interface screen for reviewing detected defects. Products are inspected from both sides, providing the user with a dual view from the left and the right side. All defects are highlighted and properly marked according to predefined quality criteria.

Image4

This cell design employs classic quality control methods, enhanced with industrial vision and AI technology. The goal is to achieve results that closely replicate human assessments. This approach identifies defects that are either unclassifiable by existing standards or entirely new and previously undocumented. The system makes decisions that mirror the subjective judgments of experienced operators but doing so on basis of an incomparably vaster learning library.

Cs Kf Inspect Display Assy Robotvision

Industrial vision system

The visual control cell for the gas flow meter required two separate stations to accommodate the complex design of its plastic housing, which incorporates multiple components. This cell is equipped with tools to ensure reliable, industrial-grade scanning, including five high-resolution cameras and ten robust LED lights and backlights.

At the first station, cameras scan the product’s surfaces, which are evenly illuminated by LED lighting from all directions to ensure uniform conditions throughout the image scan. Comprehensive quality control is achieved via a robotic arm that rotates the product, enabling scans from multiple angles.

Backlighting illuminates the product from behind, accentuating any anomalies in the transparent window screen of the meter and aiding in the detection of defects or deformities.

Once scanning is complete, the system processes the images based on predetermined parameters. Each image is analysed both in segments and as a whole; this concurrent segmentation accelerates the calculation and analysis of parameters.

Image analysis

Due to the product’s complexity, the parameters must be carefully adjusted. The system identifies potential defects across the product surfaces by analysing nearby pixel groups. The industrial vision system detects changes in contrast between individual pixels and compares these with the contrasts of adjacent pixels. Based on the analysis of a segment of the full scan, the system can determine whether a change in contrast indicates a defect or a standard product feature such as a button, indentation, or transparent window.

The analysis must be time-efficient and not extend the cycle time; therefore, it is conducted on smaller image segments while the robot arm transfers the product from one station to another. This analysis includes assessing surface quality, verifying proper material application and coating, and identifying potential defects in the transparent display window. The system checks whether all injected components are dimensionally accurate and correctly positioned.

Synchronized with the robot, the system delivers quality assessment results to the robot before it picks up the product, ensuring seamless operation. The robot then sorts the product into either the OK or NOK line for further classification and possible repair.

Inspect Disp Assy Key Technologies 4 Vb

Inspect Deep Learning Vb Circle

Deep learning technology

Designing a defect detection solution for a product with considerable structural complexity, we chose to utilize AI technology, specifically deep learning techniques. We employed the YOLO (You Only Look Once) model, which facilitates fast and accurate object detection. This method allows for enhanced control over data output but requires a comprehensive data library input.

Each layer of the neural network contains more than 3 million nodes, totalling over 50 million nodes. It analyses nine images per product, each with a resolution of 20 million pixels.

We developed an extensive collection of product defect templates, based on operator assessments in real-life at Plastika Skaza. These templates were used to train the system to recognize patterns of acceptable and unacceptable deformities. We utilized test examples to train the system to detect changes in contrast between individual pixels, adhering to predefined tolerances and thresholds. Based on this data, the system determines whether a contrast change constitutes a defect.

Results

100%

quality control of all products

each component is subjected to a comprehensive inspection

0%

affected production cycle

the production cycle remains unchanged despite complex analysis

50 mil

verified parameters
per product

the deep learning technique identifies even new, unclassified defects
No stories found.

Contact us

Have questions or want more information? Fill out the form, and we'll connect with you promptly.


    Your personal data will be processed by INEA d.o.o., Stegne 11, Ljubljana, based on legitimate interest (communication with customers and the public, Article 6(1)(f) of GDPR) for the purpose of responding to your question or inquiry. We will only keep your personal data until the end of our communication. If you have given us consent (Article 6(1)(a) of GDPR) to receive personalized e-news and offers, we will process your personal data for the purpose of sending newsletters and offers until revoked. Predictions about your interests will be based on your inquiries regarding our solutions. You can unsubscribe from receiving emails at any time by clicking the unsubscribe link in the received message. For more details, please read our Privacy Policy.