Improved weighted extreme learning machine with adaptive cost-sensitive strategy for imbalanced fault diagnosis of rotating machinery

Since the fixed weight strategy and the ignored significance of class may reduce the diagnostic performance of weighted extreme learning machine (WELM) in real scenarios, this paper proposed an improved weighted extreme learning machine (IWELM) method with adaptive cost-sensitive strategy for imbala...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	Mechanical systems and signal processing Jg. 217; S. 111526
Hauptverfasser:	Zhao, Yinghao, Yang, Xu, Huang, Jian, Gao, Jingjing, Cui, Jiarui
Format:	Journal Article
Sprache:	Englisch
Veröffentlicht:	Elsevier Ltd 01.08.2024
Schlagworte:	Cost-sensitive matrix Imbalanced fault diagnosis Multi-objective multi-verse optimizer algorithm Weighted extreme learning machine Weighted extreme learning machine Cost-sensitive matrix Multi-objective multi-verse optimizer algorithm Imbalanced fault diagnosis
ISSN:	0888-3270, 1096-1216
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	Since the fixed weight strategy and the ignored significance of class may reduce the diagnostic performance of weighted extreme learning machine (WELM) in real scenarios, this paper proposed an improved weighted extreme learning machine (IWELM) method with adaptive cost-sensitive strategy for imbalanced fault diagnosis. At first, framing and signal processing technology are applied to increase the amount of data in the data layer and weaken the influence of one-sidedness of single-domain features. On this basis, the learned multi-domain deep features are sent to IWELM for training, which takes into account the differences in various data volumes and the cost of class misclassification in equipment faults to improve the fault diagnosis performance of minority classes. Meanwhile, considering the balance of various items in the cost strategy of classifier and the generalization ability of the model, multi-objective multi-verse optimizer algorithm (MOMVO) is designed to adaptively search for hyperparameter combinations and fine-tune the cost-sensitive matrix. Finally, rolling bearing test rig and industrial reciprocating pump are used to verify the effectiveness of the method. The results show that the proposed method has competitive classification results for data sets with different imbalanced rates (IR).
ISSN:	0888-3270 1096-1216
DOI:	10.1016/j.ymssp.2024.111526