Automatic recognition system of welding seam type based on SVM method
In this paper, an automatic recognition system of welding seam type based on support vector machine (SVM) method is presented. The hardware of the proposed system consists of an industry robot with six degrees of freedom, a vision sensor, and a computer. The system has two parts including input feat...
Gespeichert in:
Veröffentlicht in: | International journal of advanced manufacturing technology 2017-09, Vol.92 (1-4), p.989-999 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | In this paper, an automatic recognition system of welding seam type based on support vector machine (SVM) method is presented. The hardware of the proposed system consists of an industry robot with six degrees of freedom, a vision sensor, and a computer. The system has two parts including input feature vector computation and model building. In the input feature vector computation part, the depth values of a series of points of the welding joint are taken as feature vector, which are determined by four steps including main line extraction of the laser stripe, normalization of the laser stripe, selection of the left and right edge points of the welding joint, and normalization of feature vectors. In the model building part, SVM-based modeling method is used to achieve welding seam type recognition. At first, RBF kernel function is employed for classification of welding seam types. Then, the parameters of RBF are determined by a grid search method using cross-validation. After the optimal parameters of RBF being determined, the SVM model is built, and it could be used to predict welding seam type. Finally, a series of welding seam type recognition experiments are implemented. Experimental results show that the proposed system can achieve welding seam type recognition accurately and the computation cost can be reduced compared with previous methods. |
---|---|
ISSN: | 0268-3768 1433-3015 |
DOI: | 10.1007/s00170-017-0202-8 |