2019 CSCE Annual Conference - Laval (Greater Montreal)

2019 CSCE Annual Conference - Laval (Greater Montreal) Conference


Title
Prediction of Compressive Capacity of Gusset Plates Using Computational Intelligence

Return to Session

Author(s)
Dr. Meisam Safari Gorji, University of Alberta (Presenter)
Dr. J.J. Roger Cheng
Abstract

Gusset plate buckling is a major failure mode in bracing systems and truss bridges. Since the stress distribution in the connection area under compression is complex, it is difficult to accurately evaluate the compressive capacity of gusset plates. Current procedures for the compressive design of gusset plates involve highly simplified approaches, which typically result in poor and inconsistent prediction of compressive strength and design reliability. In this research, two computational intelligence techniques, namely (1) Genetic Programming (GP) and (2) Adaptive Neuro-Fuzzy Inference System (ANFIS), are employed to develop predictive models for compressive capacity of corner gusset plates. The GP-based model is an empirical expression and the neuro-fuzzy model is a black-box model, which can be used as a part of a computer software. The predictive models are developed based on a comprehensive database, consisting of test results and test-validated finite element models, collected from the literature. The database under consideration covers several key parameters influencing the buckling capacity of gusset plates and encompasses reasonably wide ranges for the mechanical and geometrical properties of typical gusset plates used in the industry. The predictive models correlate the ultimate compressive strength of gusset plates with their mechanical and geometrical properties. A comparative study is conducted to evaluate the prediction performance of these models, and compare their results with those of the well-known column analogy method (i.e., Thornton method). The results indicate that both of the developed models are capable of estimating the compressive capacity of gusset plates with high accuracy, and their prediction performances are significantly better than that of the current design procedures.