Industry 4.0 Foundry Data Management and Supervised Machine Learning in Low-Pressure Die Casting Quality Improvement

dc.contributorAalto Universityen
dc.contributor.authorUyan, Tekinen_US
dc.contributor.authorOtto, Kevinen_US
dc.contributor.authorSilva, Maria Santosen_US
dc.contributor.authorVilaça, Pedroen_US
dc.contributor.authorArmakan, Elvanen_US
dc.contributor.departmentDepartment of Mechanical Engineeringen
dc.contributor.groupauthorAdvanced Manufacturing and Materialsen
dc.contributor.organizationCevher Wheelsen_US
dc.contributor.organizationUniversity of Melbourneen_US
dc.descriptionFunding Information: This work was made possible with support from an Academy of Finland, Project Number 310252. The authors would also like to thank Cevher Wheels foundry team for providing the data and field support. Publisher Copyright: © 2022, The Author(s).
dc.description.abstractLow-pressure die cast (LPDC) is widely used in high performance, precision aluminum alloy automobile wheel castings, where defects such as porosity voids are not permitted. The quality of LPDC parts is highly influenced by the casting process conditions. A need exists to optimize the process variables to improve the part quality against difficult defects such as gas and shrinkage porosity. To do this, process variable measurements need to be studied against occurrence rates of defects. In this paper, industry 4.0 cloud-based systems are used to extract data. With these data, supervised machine learning classification models are proposed to identify conditions that predict defectives in a real foundry Aluminum LPDC process. The root cause analysis is difficult, because the rate of defectives in this process occurred in small percentages and against many potential process measurement variables. A model based on the XGBoost classification algorithm was used to map the complex relationship between process conditions and the creation of defective wheel rims. Data were collected from a particular LPDC machine and die mold over three shifts and six continuous days. Porosity defect occurrence rates could be predicted using 36 features from 13 process variables collected from a considerably small sample (1077 wheels) which was highly skewed (62 defectives) with 87% accuracy for good parts and 74% accuracy for parts with porosity defects. This work was helpful in assisting process parameter tuning on new product pre-series production to lower defectives.en
dc.description.versionPeer revieweden
dc.identifier.citationUyan, T, Otto, K, Silva, M S, Vilaça, P & Armakan, E 2023, ' Industry 4.0 Foundry Data Management and Supervised Machine Learning in Low-Pressure Die Casting Quality Improvement ', INTERNATIONAL JOURNAL OF METALCASTING, vol. 17, no. 1, pp. 414-429 .
dc.identifier.otherPURE UUID: 25138d01-9524-4f13-bfbc-78fa1bd0f164en_US
dc.identifier.otherPURE ITEMURL:
dc.identifier.otherPURE LINK:
dc.identifier.otherPURE FILEURL:
dc.publisherAmerican Foundry Society
dc.relation.ispartofseriesVolume 17, issue 1en
dc.subject.keywordalloy wheelsen_US
dc.subject.keywordindustry 4.0en_US
dc.subject.keywordlow-pressure die castingen_US
dc.subject.keywordmachine learningen_US
dc.subject.keywordsmart foundryen_US
dc.subject.keywordsustainable metals processingen_US
dc.titleIndustry 4.0 Foundry Data Management and Supervised Machine Learning in Low-Pressure Die Casting Quality Improvementen
dc.typeA1 Alkuperäisartikkeli tieteellisessä aikakauslehdessäfi