Details
| Original language | English |
|---|---|
| Pages (from-to) | 501-509 |
| Number of pages | 9 |
| Journal | Production Engineering |
| Volume | 13 |
| Issue number | 3-4 |
| Early online date | 21 Mar 2019 |
| Publication status | Published - 1 Jun 2019 |
Abstract
This paper presents a deep learning-based approach for the detection and classification of production defects that complements an existing thermographic online monitoring system for Automated-Fiber-Placement (AFP) processes. The detection and classification procedure is performed in two stages. In the first stage, the system monitors each tow individually and classifies its process status. Furthermore, it detects and classifies production defects that affect individual tows such as a tow-twist. In the second stage, the system monitors the total width of the faultless tows. In this stage, production defects effecting multiple tows, for example gaps or overlaps, are detected and classified. Twelve different deep convolution neural networks (CNN) with three various architectures are learned supervised relating to different data sets. The performance of both identification stages is explored separately before the entire system will be set up. Therefore, the thermal images of the data sets are superimposed by noise to test the performance of the selected CNN.
Keywords
- Automated-fiber-placement, Defects, Prepreg, Process monitoring, Thermography
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. 13, No. 3-4, 01.06.2019, p. 501-509.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - Deep learning-based classification of production defects in automated-fiber-placement processes
AU - Schmidt, Carsten
AU - Hocke, Tristan
AU - Denkena, Berend
N1 - Funding information: The authors would like to thank the Federal state of Lower Saxony and the Volkswagen Foundation for funding the research project “Multi-Matrix-Prepreg”. They also thank the Central Innovation Program for SMEs (ZIM) for funding the ongoing research. For further information, visit the website http://www.hpcfk.de . Berend Denkena was funded by Zentrales Innovationsprogramm Mittelstand (Grant no. KF2328125PO4) and Volkswagen Foundation (Grant no. ZN3063).
PY - 2019/6/1
Y1 - 2019/6/1
N2 - This paper presents a deep learning-based approach for the detection and classification of production defects that complements an existing thermographic online monitoring system for Automated-Fiber-Placement (AFP) processes. The detection and classification procedure is performed in two stages. In the first stage, the system monitors each tow individually and classifies its process status. Furthermore, it detects and classifies production defects that affect individual tows such as a tow-twist. In the second stage, the system monitors the total width of the faultless tows. In this stage, production defects effecting multiple tows, for example gaps or overlaps, are detected and classified. Twelve different deep convolution neural networks (CNN) with three various architectures are learned supervised relating to different data sets. The performance of both identification stages is explored separately before the entire system will be set up. Therefore, the thermal images of the data sets are superimposed by noise to test the performance of the selected CNN.
AB - This paper presents a deep learning-based approach for the detection and classification of production defects that complements an existing thermographic online monitoring system for Automated-Fiber-Placement (AFP) processes. The detection and classification procedure is performed in two stages. In the first stage, the system monitors each tow individually and classifies its process status. Furthermore, it detects and classifies production defects that affect individual tows such as a tow-twist. In the second stage, the system monitors the total width of the faultless tows. In this stage, production defects effecting multiple tows, for example gaps or overlaps, are detected and classified. Twelve different deep convolution neural networks (CNN) with three various architectures are learned supervised relating to different data sets. The performance of both identification stages is explored separately before the entire system will be set up. Therefore, the thermal images of the data sets are superimposed by noise to test the performance of the selected CNN.
KW - Automated-fiber-placement
KW - Defects
KW - Prepreg
KW - Process monitoring
KW - Thermography
UR - http://www.scopus.com/inward/record.url?scp=85071725443&partnerID=8YFLogxK
U2 - 10.1007/s11740-019-00893-4
DO - 10.1007/s11740-019-00893-4
M3 - Article
VL - 13
SP - 501
EP - 509
JO - Production Engineering
JF - Production Engineering
SN - 0944-6524
IS - 3-4
ER -