In the continuous production process of injection molding, it is challenging to prevent the occurrence of defective products, leading to significant labor being invested in manually inspecting each item.
If it were possible to predict molding defects such as silver streaks, weld lines, and foreign material inclusion, defective products predicted to occur in a single shot could be automatically picked up in collaboration with robots, significantly reducing inspection labor.
MAZIN, Inc. is committed to developing AI systems that solve production challenges on a process-by-process basis.
In the injection molding process, we are working end-to-end, from developing algorithms that offer functionalities such as defect detection and automation of molding condition adjustments to providing systems that implement these algorithms to realize their capabilities.
This time, we introduce an initiative where we implemented our proprietary defect prediction algorithm into a system and verified its effectiveness on an actual injection molding machine.
In the initiative we're introducing, we verified whether it's possible to predict defective molding in actual injection molding machines that are continuously molding, and furthermore, whether it's feasible to output a digital signal for detected defects simultaneously, coordinating with a pickup robot to automatically eject the defective product.
In addition to continuing the development of our defect prediction algorithm, we began developing a system for data communication between the injection molding machine and the pickup robot. We implemented this defect prediction algorithm into the system and introduced it to an actual injection molding machine used in production settings to verify the accuracy of defect predictions.
Upon introducing the system into the operational injection molding machine, we started continuous molding under conditions set for producing quality goods. As long as the products were of good quality, the system showed no reaction. However, when we intentionally altered the conditions to differ from those for producing quality goods, the pressure waveforms inside the mold changed from those observed during the production of good parts.
At this moment, the defect prediction algorithm predicted that a defective product was being molded and output a digital signal to the pickup robot.
Once activated, the pickup robot picked up the molded product from the conveyor and removed it. Upon visual inspection of the picked-up product, it was confirmed to be defective.
This initiative has allowed us to verify the feasibility of a system operation that encompasses everything from defect prediction to the pickup of defective products.
Moving forward, we will embark on development efforts to enable data communication with injection molding machines, aiming to complete a system that automates the entire process from defect prediction to the ejection of defective products by the pickup robot.
For more detailed information about the technology or inquiries related to research and development, please contact us.
