engleski

Comparison of the Pull-out Strength of Steel Bars Glued in GluLam Elements Obtained Experimentally and Numerically

The paper presents the comparison of results obtained experimentally, numerically and analytically as well as the efficiency of trained Neural Network in prediction of the strength capacity of the threaded steel bars glued in GluLam elements. Results of FE linear analysis of 3D parametrically prepared numerical models had been used for generating of NN database. Joint specimens have been tested in the laboratory using destructive one-side pull-out test. Test method together with test results is very briefly described in further text. The experiment was the starting point of the whole process with purpose of determining the tensile strength of joint. As their final result, tests carried out the (Fracture) Force- Displacement diagrams. Experimental models have their basis in fracture mechanics approaches to study of modes of strength failure. FE parametric numerical models of an experiment using FEM package program COSMOS/M had also been prepared. Comparison of results of experimental test with results of analysis of 3D FE models was the next step of this researching. Numerical FE analysis is based on linear fracture mechanics analysis of the model of the adhesive layer's behaviour. Satisfactory correlation between experimentally obtained results and those that numerically produced was an ultimate issue we have intended to achieve. Confirming of numerically results by analytic model with rational theoretical bases was the final goal of comparison. The models have their bases in fracture mechanics and the key material property is the strength failure characteristics of the decisive fracture area and the mode of the fracture. EC5 standards, with a little exception of their informative annexes, do not regulate this type of joints in a legal way.

destructive one-side pull-out tests; glued-in threaded steel bars; GluLam Fir elements; 3D-parametric FEA modelling; analytic modelling; results comparison; Neural Networks; strength prediction

nije evidentirano