Determination of notch factors for welded cruciform joints based on numerical analysis and metamodeling
The effective notch stress approach to estimate fatigue strength of welded components requires knowledge of stress concentration factors of an idealized weld geometry using notch radii. This paper covers the estimation of those stress concentration factors for welded cruciform joints with double-fil...
Gespeichert in:
Veröffentlicht in: | Welding in the world 2019-09, Vol.63 (5), p.1339-1354 |
---|---|
Hauptverfasser: | , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | The effective notch stress approach to estimate fatigue strength of welded components requires knowledge of stress concentration factors of an idealized weld geometry using notch radii. This paper covers the estimation of those stress concentration factors for welded cruciform joints with double-filled welds, and K- or double-Y-butt welds with partial or full penetration. Thin, medium, and thick-walled joints are covered in the resulting estimations as well as different notch radii and weld angles. In comparison with different existing estimations, new methods using metamodeling (a) by the response surface method based on polynomial regression using coupling terms and (b) based on artificial neural networks are presented. Both methods show similar and superior quality. Much lower errors compared to existing estimation methods are obtained. The methods were trained by a large data base of reference results obtained by finite element analysis for 5973 design alternatives (samples) in total. Besides higher quality of prognosis, the new metamodels enhance the range of allowable parameters of the cruciform joints compared to existing ones. The resulting methods provide sound means of obtaining stress concentration factors fast and of sufficient quality which could also be embedded in more complex applications as programmed solutions. |
---|---|
ISSN: | 0043-2288 1878-6669 |
DOI: | 10.1007/s40194-019-00751-y |