Embedded Files
Dataset Overview
Dataset Overview
IMS-Foxconn cutting wheel image dataset is provided by the NSF I/UCR Center for Intelligent Maintenance Systems (IMS) at University of Cincinnati, and Foxconn Technology Group. This dataset contains images of 6 cutting wheels which are collected in the real production lines. This dataset is publicly available, and it is highly welcome that researchers use this dataset to validate image-based prognostic methods. Publications using the IMS-Foxconn dataset are requested to cite the following paper.
IMS-Foxconn cutting wheel image dataset is provided by the NSF I/UCR Center for Intelligent Maintenance Systems (IMS) at University of Cincinnati, and Foxconn Technology Group. This dataset contains images of 6 cutting wheels which are collected in the real production lines. This dataset is publicly available, and it is highly welcome that researchers use this dataset to validate image-based prognostic methods. Publications using the IMS-Foxconn dataset are requested to cite the following paper.
[1] Xiang Li, Xiaodong Jia, Yinglu Wang, Shaojie Yang, Haodong Zhao, and Jay Lee, “Industrial Remaining Useful Life Prediction by Partial Observation Using Deep Learning with Supervised Attention”, IEEE/ASME Transactions on Mechatronics, 2020. DOI: 10.1109/TMECH.2020.2992331.
[1] Xiang Li, Xiaodong Jia, Yinglu Wang, Shaojie Yang, Haodong Zhao, and Jay Lee, “Industrial Remaining Useful Life Prediction by Partial Observation Using Deep Learning with Supervised Attention”, IEEE/ASME Transactions on Mechatronics, 2020. DOI: 10.1109/TMECH.2020.2992331.
Data Links
Data Links
Preparing. Coming soon.
Preparing. Coming soon.
Dataset Descriptions
Dataset Descriptions
The television liquid crystal display (LCD) panel cutting process is focused on, and the shunt production line is shown in Figure 1. The LCD panel is the key component in the LCD screen. The panel factory provides semi-manufactured LCD panel with the test pin, and the panel is cut by the wheels in this industrial scenario. The health conditions of the cutting wheels which can be largely affected by the wear, broken glasses, abrasion etc. during operation, have significant influence on the product quality, since the panel defects can not be repaired. Figure 1 also shows the cutting process and the panel defects during manufacturing. Furthermore, the wheels are also expensive tools in this process. Therefore, effective and accurate prognosis of the cutting wheels is of great importance, which improves yield rate, reduces defective products and increases economic profits.
The television liquid crystal display (LCD) panel cutting process is focused on, and the shunt production line is shown in Figure 1. The LCD panel is the key component in the LCD screen. The panel factory provides semi-manufactured LCD panel with the test pin, and the panel is cut by the wheels in this industrial scenario. The health conditions of the cutting wheels which can be largely affected by the wear, broken glasses, abrasion etc. during operation, have significant influence on the product quality, since the panel defects can not be repaired. Figure 1 also shows the cutting process and the panel defects during manufacturing. Furthermore, the wheels are also expensive tools in this process. Therefore, effective and accurate prognosis of the cutting wheels is of great importance, which improves yield rate, reduces defective products and increases economic profits.
Figure 1 Industrial LCD panel production line, cutting process and panel defects.
Figure 1 Industrial LCD panel production line, cutting process and panel defects.
Figure 2 Image collection environment in the cutting wheel dataset, and examples of normal and degraded wheels.
Figure 2 Image collection environment in the cutting wheel dataset, and examples of normal and degraded wheels.
Conventionally, it is difficult to monitor the health conditions of the small wheels, since passive cutting is implemented and most parts of the wheel are hidden inside the machine tool during operation. That makes the widely used signals less effective for prognostics, such as vibration data. In order to address this issue, a digital high-resolution camera is placed in front the wheel to capture the images, as shown in Figure 2. Some image examples are also presented in Figure 2 for illustration. It can be observed that the wheel health states can be well indicated by the images, and the abnormal conditions such as broken glass blockage, abrasion etc. can be reflected. That offers a promising approach for the cutting wheel condition monitoring and prognostics.
Conventionally, it is difficult to monitor the health conditions of the small wheels, since passive cutting is implemented and most parts of the wheel are hidden inside the machine tool during operation. That makes the widely used signals less effective for prognostics, such as vibration data. In order to address this issue, a digital high-resolution camera is placed in front the wheel to capture the images, as shown in Figure 2. Some image examples are also presented in Figure 2 for illustration. It can be observed that the wheel health states can be well indicated by the images, and the abnormal conditions such as broken glass blockage, abrasion etc. can be reflected. That offers a promising approach for the cutting wheel condition monitoring and prognostics.
In practice, the wheel is pressed down to passively cut the moving panel for roughly one meter, and lifted up. When the wheel stops spinning, the image is taken by the camera. Afterwards, the wheel is pressed down again for cutting and this procedure repeatedly proceeds. Due to the mechanical structure of the machine tool, limited view of the cutting wheel can be captured by the camera, which only covers roughly 60 degrees of the edge. Furthermore, since passive cutting is implemented, the wheel position is of significant randomness at each time step. That indicates the specific region shown in each image is not known, and the availability of the complete information of the wheel is not guaranteed. Therefore, this industrial process and the collected data form a challenging partial observation problem.
In practice, the wheel is pressed down to passively cut the moving panel for roughly one meter, and lifted up. When the wheel stops spinning, the image is taken by the camera. Afterwards, the wheel is pressed down again for cutting and this procedure repeatedly proceeds. Due to the mechanical structure of the machine tool, limited view of the cutting wheel can be captured by the camera, which only covers roughly 60 degrees of the edge. Furthermore, since passive cutting is implemented, the wheel position is of significant randomness at each time step. That indicates the specific region shown in each image is not known, and the availability of the complete information of the wheel is not guaranteed. Therefore, this industrial process and the collected data form a challenging partial observation problem.
Based on the on-site experience in the factory, each wheel is replaced after cutting for around 5000 meters of the panels, which is subject to some fluctuations in the actual production scenarios. The images of four wheels in this situation are included in this dataset, which are used from the brand new healthy state and replaced after cutting for around 5000 meters. Since the wheels are basically not broken when replaced, the data are not strictly supervised for RUL predictions.
Based on the on-site experience in the factory, each wheel is replaced after cutting for around 5000 meters of the panels, which is subject to some fluctuations in the actual production scenarios. The images of four wheels in this situation are included in this dataset, which are used from the brand new healthy state and replaced after cutting for around 5000 meters. Since the wheels are basically not broken when replaced, the data are not strictly supervised for RUL predictions.
Furthermore, in order to collect run-to-failure data for RUL estimation, two additional wheels are put back to the production line, which have been replaced after the normal operation. The two wheels continue cutting until noticeable defects are observed from the manufactured panels by the experienced technicians, that indicates the wheels reach the end of their lives. The detailed information of the dataset is presented in Table 1. In the dataset folders, the image numbers follow the sequential collection patterns in each case.
Furthermore, in order to collect run-to-failure data for RUL estimation, two additional wheels are put back to the production line, which have been replaced after the normal operation. The two wheels continue cutting until noticeable defects are observed from the manufactured panels by the experienced technicians, that indicates the wheels reach the end of their lives. The detailed information of the dataset is presented in Table 1. In the dataset folders, the image numbers follow the sequential collection patterns in each case.
Related Information
Related Information
IMS Center: http://www.imscenter.net/
IMS Center: http://www.imscenter.net/
Industrial AI Institute: https://www.iaicenter.com/
Industrial AI Institute: https://www.iaicenter.com/
Contact
Contact
If you have any questions, please feel free to contact:
If you have any questions, please feel free to contact:
Prof. Jay Lee (jay.lee@uc.edu)
Prof. Jay Lee (jay.lee@uc.edu)
Dr. Xiang Li (li5xi@ucmail.uc.edu)
Dr. Xiang Li (li5xi@ucmail.uc.edu)
Google Sites
Report abuse