In the continuous production environment of injection molding, the challenge of suppressing the occurrence of defective products is significant, with considerable labor invested in manually inspecting each item. If it were possible to predict defects such as silver streaks, sink marks, and foreign material inclusion, then defective products predicted to occur in a single shot could be automatically picked up in collaboration with robots, significantly reducing the labor required for inspection.
MAZIN, Inc. is dedicated to developing AI technologies aimed at solving production challenges at the process level. In the injection molding process, our focus is on addressing issues like skill succession and improving production efficiency. We aim to develop algorithms for defect detection and the automation of molding condition adjustments, among other applications, by conducting various experiments and analyses.
As part of these efforts, we have previously introduced experiments aimed at predicting molding defects such as silver streaks, sink marks, and foreign material inclusion.
Building on the data obtained from these initial experiments, further analysis has been conducted, revealing the potential for predicting the occurrence of defects. This update introduces the progress made in our efforts to predict and ultimately reduce the incidence of defective products through advanced analysis and AI-driven technologies.
In this initiative, we aim to determine whether specific molding defects such as silver streaks, sink marks, short shots, overpacking/flash, and foreign material inclusion can be identified by analyzing time-series waveforms of in-mold pressure, which are captured using pressure sensors attached to the mold.
Utilizing polycarbonate (PC), known for its impact resistance, durability, heat resistance, and moldability, and widely used in both consumer goods and industrial parts, we conducted injection molding experiments. By generating and analyzing original feature quantities that correlate with the state of the resin flowing inside the mold, we attempted to identify various molding defects.
This approach seeks to leverage the detailed data obtained from the pressure sensors to preemptively identify potential defects during the molding process, thereby enabling timely adjustments to the molding conditions or direct intervention, such as the automatic ejection of defective products. This method not only aims to reduce the labor-intensive process of manual inspection but also enhances the overall efficiency and quality of the production process.
We generated and analyzed original feature quantities that correlate with the state of the resin flowing inside the mold based on pressure waveforms.
The analysis revealed that there are distinguishable differences in the values on the vertical axis between the groups of good products and various types of defective products, as illustrated in the diagram below.
Notably, among the types of defects that showed significant differences from the values of good products were not only short shots and sink marks but also silver streaks, which have traditionally been considered difficult to detect. This outcome indicates a high possibility of distinguishing even appearance defects like silver streaks, suggesting an advancement in defect detection capabilities.
This initiative has successfully led to the development of an algorithm capable of predicting the occurrence of short shots, sink marks, and silver streaks.
Moving forward, we will continue to refine the algorithm to achieve higher accuracy and to accommodate a broader range of defect types.
For more detailed information about the technology or inquiries related to research and development, please contact us.
