Details
| Original language | English |
|---|---|
| Pages (from-to) | 735-742 |
| Number of pages | 8 |
| Journal | Production Engineering |
| Volume | 16 |
| Issue number | 6 |
| Early online date | 16 Apr 2022 |
| Publication status | Published - Dec 2022 |
Abstract
Errors in milling processes such as tool breakage or material inhomogeneities are a major risk to the quality of machined workpieces. Errors like a broken tool may also lead to damages to the machine tool. Process monitoring systems allow for autonomous detection of errors, therefore, promoting autonomous production. The parameterization of these systems is a trade-off between high robustness (low false alarm rate) and high sensitivity. Even though several monitoring systems have been introduced for single-item and series production, a universal parameterization technique that weighs off sensitivity and robustness does not exist. In this paper, a novel, model-independent and adjustable parameterization technique for monitoring systems is introduced. The basis for the parameterization is the material removal rate that indicates the temporal and quantitative impact of process errors (ground truth). The ground truth allows calculation of the established Fβ-score, which is used to evaluate the monitoring system. An adjustment of the β-parameter influences the weighting of sensitivity and robustness. Accordingly, the β-parameter allows to easily control the sensitivity-robustness trade-off so that the monitoring system is economic for the company’s specific situation. In this paper, a look-up table for hyper-parameters of the state-of-the-art tolerance range monitoring model is provided using the introduced parameterization approach. With this table companies and researchers can set the hyper-parameters of their monitoring models for 5-axis-milled single items user-specifically. To demonstrate, that introduced parameterization approach works for different kinds of monitoring models, a one-class support vector machine (SVM) is parameterized also.
Keywords
- Hyper-parameter, Machining, Monitoring, Parameterization, Process control, SVM
ASJC Scopus subject areas
- Engineering(all)
- Mechanical Engineering
- Engineering(all)
- Industrial and Manufacturing Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: Production Engineering, Vol. 16, No. 6, 12.2022, p. 735-742.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - User-Specific Parameterization of Process Monitoring Systems
AU - Denkena, B.
AU - Klemme, H.
AU - Becker, J.
AU - Blech, H.
N1 - Funding Information: This research was funded by the Federal Ministry of Economics (BMWi), project “IIP Ecosphere” (01MK20006A). The authors would also like to thank the “Sieglinde Vollmer Stiftung” for the financial support of this research work.
PY - 2022/12
Y1 - 2022/12
N2 - Errors in milling processes such as tool breakage or material inhomogeneities are a major risk to the quality of machined workpieces. Errors like a broken tool may also lead to damages to the machine tool. Process monitoring systems allow for autonomous detection of errors, therefore, promoting autonomous production. The parameterization of these systems is a trade-off between high robustness (low false alarm rate) and high sensitivity. Even though several monitoring systems have been introduced for single-item and series production, a universal parameterization technique that weighs off sensitivity and robustness does not exist. In this paper, a novel, model-independent and adjustable parameterization technique for monitoring systems is introduced. The basis for the parameterization is the material removal rate that indicates the temporal and quantitative impact of process errors (ground truth). The ground truth allows calculation of the established Fβ-score, which is used to evaluate the monitoring system. An adjustment of the β-parameter influences the weighting of sensitivity and robustness. Accordingly, the β-parameter allows to easily control the sensitivity-robustness trade-off so that the monitoring system is economic for the company’s specific situation. In this paper, a look-up table for hyper-parameters of the state-of-the-art tolerance range monitoring model is provided using the introduced parameterization approach. With this table companies and researchers can set the hyper-parameters of their monitoring models for 5-axis-milled single items user-specifically. To demonstrate, that introduced parameterization approach works for different kinds of monitoring models, a one-class support vector machine (SVM) is parameterized also.
AB - Errors in milling processes such as tool breakage or material inhomogeneities are a major risk to the quality of machined workpieces. Errors like a broken tool may also lead to damages to the machine tool. Process monitoring systems allow for autonomous detection of errors, therefore, promoting autonomous production. The parameterization of these systems is a trade-off between high robustness (low false alarm rate) and high sensitivity. Even though several monitoring systems have been introduced for single-item and series production, a universal parameterization technique that weighs off sensitivity and robustness does not exist. In this paper, a novel, model-independent and adjustable parameterization technique for monitoring systems is introduced. The basis for the parameterization is the material removal rate that indicates the temporal and quantitative impact of process errors (ground truth). The ground truth allows calculation of the established Fβ-score, which is used to evaluate the monitoring system. An adjustment of the β-parameter influences the weighting of sensitivity and robustness. Accordingly, the β-parameter allows to easily control the sensitivity-robustness trade-off so that the monitoring system is economic for the company’s specific situation. In this paper, a look-up table for hyper-parameters of the state-of-the-art tolerance range monitoring model is provided using the introduced parameterization approach. With this table companies and researchers can set the hyper-parameters of their monitoring models for 5-axis-milled single items user-specifically. To demonstrate, that introduced parameterization approach works for different kinds of monitoring models, a one-class support vector machine (SVM) is parameterized also.
KW - Hyper-parameter
KW - Machining
KW - Monitoring
KW - Parameterization
KW - Process control
KW - SVM
UR - http://www.scopus.com/inward/record.url?scp=85128169244&partnerID=8YFLogxK
U2 - 10.1007/s11740-022-01130-1
DO - 10.1007/s11740-022-01130-1
M3 - Article
AN - SCOPUS:85128169244
VL - 16
SP - 735
EP - 742
JO - Production Engineering
JF - Production Engineering
SN - 0944-6524
IS - 6
ER -