Zhuo Yang, Jaehyuk Kim, Yan Lu, H. Yeung, B. Lane, Albert T. Jones, Yande Ndiaye
{"title":"金属粉末床熔融增材制造过程监控的多模态数据驱动决策融合方法","authors":"Zhuo Yang, Jaehyuk Kim, Yan Lu, H. Yeung, B. Lane, Albert T. Jones, Yande Ndiaye","doi":"10.1115/iam2022-96740","DOIUrl":null,"url":null,"abstract":"\n Data fusion techniques aim to improve inference results or decision making by ‘combining’ multiple data sources. Additive manufacturing (AM) in-situ monitoring systems measure various physical phenomena and generate multiple types of data. Data types that occur at different scales and sampling rates during a build process. Data types that can be used to monitor the state of that process. Monitoring typically requires software tools to analyze multiple data sources. There are two reasons. First, data only from an individual data source may not be accurate enough or large enough to monitor the process stat. Second, a single source will be limited by the relevancy of the observations, signal-to-noise ratio, or other measurement uncertainties.\n This work proposes a decision-level, multimodal, data fusion method that combines multiple, in-situ, AM monitoring data sources to improve overall, process-monitoring performance. The work is based on a recent, laser powder bed fusion (LPBF) experiment that was conducted to create overhang surfaces throughout a 3D part. The data from that experiment is used to illustrate and validate the proposed method. The overhang features were designed with different shapes. angles, and build locations. The features are formed using constant laser power and scan speed. A high-frequency, coaxial, melt-pool, imaging system and a low-frequency layerwise staring camera are the two, in-situ, monitoring, data sources used in that experiment. The Naïve Bayes and the k-nearest-neighbors algorithms are first applied to each data set for overhang feature detection. Then both hard voting and soft voting are adopted in fusing the classification outcomes. The results show that while none of the individual classifiers are perfect in detecting overhang features, the fused decision of the 324 test samples achieved 100% detection accuracy.","PeriodicalId":184278,"journal":{"name":"2022 International Additive Manufacturing Conference","volume":"1 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-10-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A Multi-Modal Data-Driven Decision Fusion Method for Process Monitoring in Metal Powder Bed Fusion Additive Manufacturing\",\"authors\":\"Zhuo Yang, Jaehyuk Kim, Yan Lu, H. Yeung, B. Lane, Albert T. Jones, Yande Ndiaye\",\"doi\":\"10.1115/iam2022-96740\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"\\n Data fusion techniques aim to improve inference results or decision making by ‘combining’ multiple data sources. Additive manufacturing (AM) in-situ monitoring systems measure various physical phenomena and generate multiple types of data. Data types that occur at different scales and sampling rates during a build process. Data types that can be used to monitor the state of that process. Monitoring typically requires software tools to analyze multiple data sources. There are two reasons. First, data only from an individual data source may not be accurate enough or large enough to monitor the process stat. Second, a single source will be limited by the relevancy of the observations, signal-to-noise ratio, or other measurement uncertainties.\\n This work proposes a decision-level, multimodal, data fusion method that combines multiple, in-situ, AM monitoring data sources to improve overall, process-monitoring performance. The work is based on a recent, laser powder bed fusion (LPBF) experiment that was conducted to create overhang surfaces throughout a 3D part. The data from that experiment is used to illustrate and validate the proposed method. The overhang features were designed with different shapes. angles, and build locations. The features are formed using constant laser power and scan speed. A high-frequency, coaxial, melt-pool, imaging system and a low-frequency layerwise staring camera are the two, in-situ, monitoring, data sources used in that experiment. The Naïve Bayes and the k-nearest-neighbors algorithms are first applied to each data set for overhang feature detection. Then both hard voting and soft voting are adopted in fusing the classification outcomes. The results show that while none of the individual classifiers are perfect in detecting overhang features, the fused decision of the 324 test samples achieved 100% detection accuracy.\",\"PeriodicalId\":184278,\"journal\":{\"name\":\"2022 International Additive Manufacturing Conference\",\"volume\":\"1 1\",\"pages\":\"0\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-10-19\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"2022 International Additive Manufacturing Conference\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1115/iam2022-96740\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 International Additive Manufacturing Conference","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1115/iam2022-96740","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
A Multi-Modal Data-Driven Decision Fusion Method for Process Monitoring in Metal Powder Bed Fusion Additive Manufacturing
Data fusion techniques aim to improve inference results or decision making by ‘combining’ multiple data sources. Additive manufacturing (AM) in-situ monitoring systems measure various physical phenomena and generate multiple types of data. Data types that occur at different scales and sampling rates during a build process. Data types that can be used to monitor the state of that process. Monitoring typically requires software tools to analyze multiple data sources. There are two reasons. First, data only from an individual data source may not be accurate enough or large enough to monitor the process stat. Second, a single source will be limited by the relevancy of the observations, signal-to-noise ratio, or other measurement uncertainties.
This work proposes a decision-level, multimodal, data fusion method that combines multiple, in-situ, AM monitoring data sources to improve overall, process-monitoring performance. The work is based on a recent, laser powder bed fusion (LPBF) experiment that was conducted to create overhang surfaces throughout a 3D part. The data from that experiment is used to illustrate and validate the proposed method. The overhang features were designed with different shapes. angles, and build locations. The features are formed using constant laser power and scan speed. A high-frequency, coaxial, melt-pool, imaging system and a low-frequency layerwise staring camera are the two, in-situ, monitoring, data sources used in that experiment. The Naïve Bayes and the k-nearest-neighbors algorithms are first applied to each data set for overhang feature detection. Then both hard voting and soft voting are adopted in fusing the classification outcomes. The results show that while none of the individual classifiers are perfect in detecting overhang features, the fused decision of the 324 test samples achieved 100% detection accuracy.