INVESTIGATION ON LOCAL BEARING BEHAVIOR IN HIGH STRENGTH STEEL SINGLE-BOLT CONNECTION WITH MACHINE LEARNING TECHNIQUES
Yi-Fan Lyu1*, Yan-Bo Wang1, and Guo-Qiang Li2
1 Department of Civil Engineering, Tongji University, Shanghai, China
2 State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai, China
This paper conducts an investigation on the local bearing behavior based on 15 groups of high strength steel single-bolt connections. The experimental load-displacement curves are used for a further numerical simulation, which shows a good agreement with the experimental results. With the validated numerical model, distribution of the equivalent plastic strain near bolt hole is extracted as the pattern for local bearing behavior. Two patterns of the varying distribution of equivalent plastic strain related to growing bearing resistance is found for connections with different end distance. The first pattern represents the state when distribution of equivalent plastic strain has an uniform or mountain shape. The second pattern represents a state where a saddle-shaped distribution of equivalent plastic strain occurs. Machine learning techniques for pattern identification is introduced to quantify the boundary between the two patterns from the view of Bayesian posterior probability. Based on the further comparison with current Eurocode3, it is found that obvious bolt hole elongation cannot be avoided in the design by Eurocode3 due to the saddle-shaped distribution of equivalent plastic strain. A modified resistance considering limitation of bolt hole elongation is suggested in the future research.
Connection; Bearing resistance; Equivalent plastic strain; Machine learning