Limits...
Vision-based sensor for early detection of periodical defects in web materials.

Bulnes FG, Usamentiaga R, García DF, Molleda J - Sensors (Basel) (2012)

Bottom Line: For this reason, it is necessary to have a system that can detect these situations as soon as possible.A total of 36 strips produced in ArcelorMittal Avilés factory were used for this purpose, 18 to determine the optimal configuration of the proposed sensor using a full-factorial experimental design and the other 18 to verify the validity of the results.Next, they were compared with those provided by a commercial system used worldwide, showing a clear improvement.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, University of Oviedo, Campus de Viesques, Gijón 33204, Spain. bulnes@uniovi.es

ABSTRACT
During the production of web materials such as plastic, textiles or metal, where there are rolls involved in the production process, periodically generated defects may occur. If one of these rolls has some kind of flaw, it can generate a defect on the material surface each time it completes a full turn. This can cause the generation of a large number of surface defects, greatly degrading the product quality. For this reason, it is necessary to have a system that can detect these situations as soon as possible. This paper presents a vision-based sensor for the early detection of this kind of defects. It can be adapted to be used in the inspection of any web material, even when the input data are very noisy. To assess its performance, the sensor system was used to detect periodical defects in hot steel strips. A total of 36 strips produced in ArcelorMittal Avilés factory were used for this purpose, 18 to determine the optimal configuration of the proposed sensor using a full-factorial experimental design and the other 18 to verify the validity of the results. Next, they were compared with those provided by a commercial system used worldwide, showing a clear improvement.

No MeSH data available.


Differences in detection depending on the initial solution.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3472857&req=5

f8-sensors-12-10788: Differences in detection depending on the initial solution.

Mentions: A very important aspect that must be taken into account is that the individual defect selected to form the initial solution greatly affects the search, since the values of their characteristics iid.tr, iid.ar, iid.wi y iid.le are assigned to features pd.t, pd.a, pd.w y pd.l of its periodical defect respectively at the beginning of the search. Thus, the clustering algorithm includes in the periodical defect only those individual defects whose characteristics are similar to those of the individual defect included in the initial solution. Therefore, if this individual defect was detected by the vision system incorrectly (e.g., assigning a value to its dimensions and area much larger than they should be) the clustering algorithm will not work properly, since it will include only individual defects similar to it (whose dimensions and area are also incorrect), as shown in Figure 8. In this figure, the individual defect chosen to form the initial solution is surrounded by a black square, and individual defects included in the periodical defect by the clustering algorithm are surrounded by red squares. The clustering is better in the cases where the individual defect included in the initial solution is more like the majority of individual defects that form the periodical defect. Since the individual defect most appropriate to form the initial solution can not be determined before running the algorithm, it must receive as many initial solutions as individual defects are located at the transversal position determined by the histogram (see Figure 6). For each of them, both forward and backward searches are applied. The largest periodical defect (i.e., the one whose feature n is greater) is the one chosen to be part of the output, but only if its feature n is greater than the size threshold.


Vision-based sensor for early detection of periodical defects in web materials.

Bulnes FG, Usamentiaga R, García DF, Molleda J - Sensors (Basel) (2012)

Differences in detection depending on the initial solution.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3472857&req=5

f8-sensors-12-10788: Differences in detection depending on the initial solution.
Mentions: A very important aspect that must be taken into account is that the individual defect selected to form the initial solution greatly affects the search, since the values of their characteristics iid.tr, iid.ar, iid.wi y iid.le are assigned to features pd.t, pd.a, pd.w y pd.l of its periodical defect respectively at the beginning of the search. Thus, the clustering algorithm includes in the periodical defect only those individual defects whose characteristics are similar to those of the individual defect included in the initial solution. Therefore, if this individual defect was detected by the vision system incorrectly (e.g., assigning a value to its dimensions and area much larger than they should be) the clustering algorithm will not work properly, since it will include only individual defects similar to it (whose dimensions and area are also incorrect), as shown in Figure 8. In this figure, the individual defect chosen to form the initial solution is surrounded by a black square, and individual defects included in the periodical defect by the clustering algorithm are surrounded by red squares. The clustering is better in the cases where the individual defect included in the initial solution is more like the majority of individual defects that form the periodical defect. Since the individual defect most appropriate to form the initial solution can not be determined before running the algorithm, it must receive as many initial solutions as individual defects are located at the transversal position determined by the histogram (see Figure 6). For each of them, both forward and backward searches are applied. The largest periodical defect (i.e., the one whose feature n is greater) is the one chosen to be part of the output, but only if its feature n is greater than the size threshold.

Bottom Line: For this reason, it is necessary to have a system that can detect these situations as soon as possible.A total of 36 strips produced in ArcelorMittal Avilés factory were used for this purpose, 18 to determine the optimal configuration of the proposed sensor using a full-factorial experimental design and the other 18 to verify the validity of the results.Next, they were compared with those provided by a commercial system used worldwide, showing a clear improvement.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science, University of Oviedo, Campus de Viesques, Gijón 33204, Spain. bulnes@uniovi.es

ABSTRACT
During the production of web materials such as plastic, textiles or metal, where there are rolls involved in the production process, periodically generated defects may occur. If one of these rolls has some kind of flaw, it can generate a defect on the material surface each time it completes a full turn. This can cause the generation of a large number of surface defects, greatly degrading the product quality. For this reason, it is necessary to have a system that can detect these situations as soon as possible. This paper presents a vision-based sensor for the early detection of this kind of defects. It can be adapted to be used in the inspection of any web material, even when the input data are very noisy. To assess its performance, the sensor system was used to detect periodical defects in hot steel strips. A total of 36 strips produced in ArcelorMittal Avilés factory were used for this purpose, 18 to determine the optimal configuration of the proposed sensor using a full-factorial experimental design and the other 18 to verify the validity of the results. Next, they were compared with those provided by a commercial system used worldwide, showing a clear improvement.

No MeSH data available.