MACHINE LEARNING APPROACH FOR THE PREDICTION OF BUCKLING STRENGTH OF CORRODED STEEL PLATES

Abstract

The mechanical behavior of corroded steel members is of interest due to the increased age of steel structures around the world. Though considerable literature exists on the tensile behavior of corroded steel plates, relatively few studies have been reported on buckling behavior. For example, while simple evaluation formulas have been proposed by a few researchers, buckling test results were not used in the development of many of these equations. As a result, the accumulation of data on the buckling behavior of corroded steel plates is lacking. To address this issue, we conducted both buckling tests and finite element analysis to obtain extensive information about the effects of surface profiles on the buckling strength of steel plates. A finite element model of corroded steel plates was developed from a spatial autocorrelation model, which can represent several types of corroded surfaces, including pitting corrosion.
After obtaining the results, we trained an artificial neural network model, of which input is the characteristics of corroded steel plates including the surface profile and materials properties, and verified the accuracy of the model by leave-one-out cross validation. It was found that the model can evaluate the buckling strength of corroded steel plates with much higher accuracy than conventional equations.