WEKO3
アイテム
製造データからの因果関係発見に向けて：分散不均一性と変数グループについて
http://hdl.handle.net/10441/0002000319
http://hdl.handle.net/10441/00020003197e18667c-dadb-40bc-b050-e0e2cda01268
| 名前 / ファイル | ライセンス | アクション |
|---|---|---|
博士論文全文甲50.pdf (4 MB)
|
|
|
|
博士論文審査要旨甲50.pdf (177 KB)
|
|
|
|
博士論文要旨甲50.pdf (187 KB)
|
|
| Item type | 学位論文 / Thesis or Dissertation(1) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 公開日 | 2024-06-21 | |||||||||||
| タイトル | ||||||||||||
| タイトル | Toward Discovering Causal Relations from Manufacturing Data: Heteroscedasticity and Variable Groups | |||||||||||
| 言語 | en | |||||||||||
| タイトル | ||||||||||||
| タイトル | 製造データからの因果関係発見に向けて：分散不均一性と変数グループについて | |||||||||||
| 言語 | ja | |||||||||||
| タイトル | ||||||||||||
| タイトル | セイゾウ データ カラ ノ インガ カンケイ ハッケン ニ ムケテ : ブンサン フキンイツセイ ト ヘンスウ グループ ニ ツイテ | |||||||||||
| 言語 | ja-Kana | |||||||||||
| 言語 | ||||||||||||
| 言語 | eng | |||||||||||
| 資源タイプ | ||||||||||||
| 資源タイプ識別子 | http://purl.org/coar/resource_type/c_db06 | |||||||||||
| 資源タイプ | doctoral thesis | |||||||||||
| アクセス権 | ||||||||||||
| アクセス権 | open access | |||||||||||
| アクセス権URI | http://purl.org/coar/access_right/c_abf2 | |||||||||||
| 著者 |
菊池, 元太
× 菊池, 元太
|
|||||||||||
| 抄録 | ||||||||||||
| 内容記述タイプ | Abstract | |||||||||||
| 内容記述 | Discovering causal relationships between quantities of interest is fundamental in many scientific disciplines. This thesis focuses on the field of manufacturing, where data-driven quality improvements are attracting increasing attention because of the more diverse data accumulated in the wake of Industry 4.0 and digital transformation. Understanding the causal relations among the various measurements, such as those of product qualities, machine parameters, and manufacturing environment, is crucial for data-driven quality improvement activities. Although controlled experiments are the recommended approach to infer cause?effect relations, such experiments can be unethical, technically challenging, or too expensive. For example, manufacturing a set of defective products during mass production is unrealistic, as it decreases overall equipment effectiveness and might affect subsequent products. Numerous methods have been developed to estimate causal relationships from observational data, termed causal discovery, to tackle this issue. Research that applies causal discovery methods to manufacturing data assumes that the data exhibit non-linearity, temporal dependencies, or both. However, they overlook a typical characteristic of manufacturing data, heteroscedasticity, which causes severe problems with many existing causal discovery methods. Another issue is handling groups of variables; when multiple measurements take similar values, selecting one of them or aggregating them by taking an average may impede the estimation performance. Several existing works on causal discovery address the aforementioned issues individually but not simultaneously. This thesis addresses the problem of performing causal discovery on non-linear timeseries data with heteroscedastic noise. We introduce an estimation method based on recently developed continuous optimization-based methods. Then, we extend the work to exploit the time structure and show that causal relationships can be uniquely recovered from data under specific assumptions. Furthermore, this thesis considers the problem of estimating causal relationships among multiple groups of variables where the functional relations are beyond linear. We propose a novel approach based on algebraic characterization of causal structure among multiple groups of variables that can be used as a constraint for the optimization problem on existing continuous optimization-based methods. |
|||||||||||
| 言語 | en | |||||||||||
| 書誌情報 |
発行日 2024-03-25 |
|||||||||||
| 著者版フラグ | ||||||||||||
| 出版タイプ | VoR | |||||||||||
| 出版タイプResource | http://purl.org/coar/version/c_970fb48d4fbd8a85 | |||||||||||
| 学位名 | ||||||||||||
| 言語 | ja | |||||||||||
| 学位名 | 博士（データサイエンス） | |||||||||||
| 学位授与機関 | ||||||||||||
| 学位授与機関識別子Scheme | kakenhi | |||||||||||
| 学位授与機関識別子 | 14201 | |||||||||||
| 言語 | ja | |||||||||||
| 学位授与機関名 | 滋賀大学 | |||||||||||
| 学位授与年月日 | ||||||||||||
| 学位授与年月日 | 2024-03-25 | |||||||||||
| 学位授与番号 | ||||||||||||
| 学位授与番号 | 甲第50号 | |||||||||||
