.png){kind=icon}
.png){kind=icon}
Return
Kernel principal component analysis as a non-linear unsupervised learning algorithm for assessing drivers of bolted angle steel connection tension failure
.png){kind=icon}
OA
AI
M
R
K
C
D
DOI：10.1016/j.rineng.2026.112263.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Bolted steel angle connection subjected to tensile loads entail several factors that drive their failure behavior, i.e., failure load and failure mode. This study examined the nature of influence of these factors on the failure load behavior. • The nature of the interaction between features, i.e., input-input, and input-output, were examined and found to be non-linear in nature. • The non-linearity between input-features prompted the use of Kernel Principal Component Analysis for factor analysis, a non-linear version of Principal Component Analysis (PCA). PCA is a data-driven unsupervised learning technique, i.e., was utilized to investigate the nature of this influence between factors, and in other cases to deal with dimensionality issues to do with factors. The analysis was conducted within Matlab. • A dataset with 472 instances, that entailed 13 input features, and 2 output features, was constructed from prior laboratory experimental and Finite Element Analysis (FEA) numeric modeling on tension failure behavior and utilized in the analysis in this study. • Findings from the KPCA revealed that global connection geometric properties with select profile section and material properties dominate in driving connection failure behavior, i.e., the number of angle steel leg(s) that are bolted, the number of bolt rows, the number of bolts per row, steel yield strength (fy), and thickness of the profile section (t). The ultimate strength (fu) of bolts manifested as having intermediate influence on failure behavior. The features that had the least influence included angle steel profile leg lengths (h and b), edge distance of outer most bolt (e1) and pitch (p1), and bolt diameter.
Keywords:
Bolted angle steel connections
Geometric properties
Material mechanical properties
Tensile failure behavior
PCA
KPCA
Unsupervised learning
Journal
IF:
7.9
Papers:
1.1W
Citations:
1.7W
Organization
Cited Papers