About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
A Processing-Aware Multimodal Data Pipeline for Porosity Quantification in High-Pressure Die Casting |
| Author(s) |
Junhyub Jeon, Ihnsik Weon, Munwon Lim, MiYoung Lee, JaeHwang Kim |
| On-Site Speaker (Planned) |
Junhyub Jeon |
| Abstract Scope |
High-pressure die casting (HPDC) of aluminum components generates heterogeneous data streams like process monitoring signals, dimensional inspection, and X-ray radiography that are rarely integrated for machine-learning-based defect analysis. We curate and align these sources for a structural die-cast automotive aluminum-alloy component. Process monitoring is reduced from 19 to 10 informative variables, with a three-stage outlier procedure yielding 12,505 shots after outlier removal. CAD geometry is parameterized from STEP models into nine descriptors linking shape to porosity susceptibility. A standardized 26-class defect taxonomy and region-wise porosity thresholds provide consistent labels. Because collected radiographs contained no rejects, physics-based synthetic defect injection sampled from region-specific casting-limit thresholds validates a quantification module computing porosity fraction, mean diameter, and cross-sectional area. The framework positions the curated process signals as the coupling target for processing-aware ML once real defect data are acquired.
This work was supported by the Technology Innovation Program(RS-2025-25455322) funded by the Ministry of Trade, Industry and Resources (MOTIR, Korea) and Korea Institute of Industrial Technology (KITECH) grant funded by the project (No. EO-26-0005) |
| Proceedings Inclusion? |
Planned: |