The use of intelligent algorithms in the prediction of bonding strength in steel-concrete interfaces

Resumo

The current study proposes the use of intelligent systems and a statistical technique to predict the
strength of steel-concrete bond using a database of academic literature. The work used as the reference deals with
the pull-out test that evaluated the steel-concrete bond behavior in thin bars. The experimental program employed
concretes of class C25, C35 and C40, and CA-50 ribbed bars (with diameters of 6.3, 8.0, and 10 mm) and CA-60
notched bars (with diameters of 5.0, 6.0, 8.0, and 9.5 mm). The database was subjected to data mining strategies
for statistical treatment. Conventionally, bond strength is obtained by pull-out and beam tests, as proposed by BS
EN:10080, involving expensive and lengthy experimental tests. Alternative ways are the application of machine
learning-based methods and the use of statistical techniques. Using these methods it is possible to assess their
efficiency in predicting the maximum pull-out force. In the present research two particular methods are used to
solve the problem. One technique is Multiple Linear Regression which is a generalization of Least Squares, where
the minimization of the sums of the n-th powers of the residuals is considered. Multiple regression analysis is also
very useful in experimental situations, where the experimenter can control the predictor variables. The other model
is based on statistical learning theory and is called Support Vector Machines (SVM). It is a machine learning and
computational intelligence technique, where it is possible to obtain a classification of data from the same domain
in which the learning is performed. The method uses a principle called induction, where it is possible to draw
generic conclusions from a training set. The main objective of this work is to propose an alternative way to predict
the bond strength of thin bars using computational methods and a statistical technique. The methods used is
compared via performance metrics to verify which one proves to be more reliable to predict steel-concrete interface
bond, considering the safety coefficients used in engineering.