AI Machine Vision-based Vision Inspection for

Injection Molding Process Difficult Defect Detection Demonstration

This is a demonstration case of smart quality control that uses AI vision technology to accurately identify difficult defects in the injection molding process.



1. Pain Point 

Existing program aspects 

- Although a rule-based vision inspection program is installed for the injection molding process, one of the main processes, it is unable to make a judgment when the image is not in the correct position.

- The rule-based vision camera program is unable to properly distinguish between defects such as “gas,” “scratches,” “unformed,” and “gloss” that are caused by light reflection and actual defects among the various types of defects in the injection molding process.



Worker Aspect

- When defects occur in the injection molding process, on-site workers detect the defects and input the quantities by type into the MES system for collection and real-time monitoring, but they cannot accurately identify the causes of the defects.

- - Workers follow manuals and SOPs for production, but frequent changes in on-site workers due to generational shifts and frequent changes in items prevent production from being carried out under optimal conditions, resulting in defects.

2. Smart Factory Construction Details

Details

- Introduction of AI vision machine solutions instead of rule-based vision inspection equipment.

- Collect, analyze, and utilize structured and unstructured image data from the manufacturing processes of demand companies through AI and process expert consulting and solution validation, and build a data infrastructure and platform to support AI solution development.

- Analyze how key factors affect quality based on structured data generated in the target injection molding process to derive a process optimization model.

 

3. Effects of Smart Factory Construction

Before Introduction

- No quality prediction

- No vision inspection 

- Process defect rate of 8% 

After Introduction

- Achieved F-1 Score (quality prediction) of 0.96 

- Achieved F-1 Score (vision inspection) of 0.99 

- Process defect rate reduced to 4%