Quantitative detection of wire rope damage based on local structural characteristics

Different number of broken wires produce different grooves on the surface of steel wire rope. Based on the local structural features of these grooves, a new broken wire identification method is proposed. By comparing the processing effects of various image enhancement methods, a processing method ca...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of intelligent & fuzzy systems 2023-08, Vol.45 (3), p.4337-4347
Hauptverfasser: Ye, Qiang, Zhang, Juwei, Chen, Quankun
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Different number of broken wires produce different grooves on the surface of steel wire rope. Based on the local structural features of these grooves, a new broken wire identification method is proposed. By comparing the processing effects of various image enhancement methods, a processing method called adaptive histogram equalization is selected to process the broken wire image. Aiming at a large amount of useless information in structural features extracted by HOG algorithm, a encoder-decoder neural network is designed to reduce the dimension of features. In addition, to effectively avoid information loss caused by the output layer of the BP neural network, a joint algorithm of the BP neural network and the support vector machine is proposed. The experimental results show that using image enhancement technology to process broken wire images can effectively improve the recognition rate of broken wires; The structural features extracted by HOG algorithm are more beneficial to the quantitative recognition of broken wires than the texture features extracted by LBP operator; Compared with various dimensionality reduction methods, neural network can retain more effective information; The joint algorithm can improve the recognition rate of broken wire by at least 0.25% on the basis of BP neural network.
ISSN:1064-1246
1875-8967
DOI:10.3233/JIFS-231259