TWO-STAGE SUPPORT VECTOR MACHINE METHOD FOR FAILURE MODE CLASSIFICATION OF REINFORCEDCONCRETE COLUMNS

A two-stage support vector machine (SVM) method for the failure mode classification of reinforced concrete (RC) columns is proposed. A two-stage SVM model is established to classify the seismic failure modes of RC columns according to three failure modes, namely, flexure failure, flexure-shear failure and shear failure. The optimal values of model parameters (i.e., penalty parameters and kernel function parameter) of the two-stage SVM model are determined by using ten-fold cross-validation and grid-search based on 270 experimental data. Subsequently, the characteristic parameters including the axial load ratio, shear span ratio, hoop spacing to depth ratio (s/h₀), longitudinal reinforcement index and transverse reinforcement index on the seismic failure modes of RC columns are analyzed by the support vector machine-recursive feature elimination (SVM-RFE). The classification accuracy of the proposed classification method is validated by comparing with two classical machine learning methods and five traditional classification methods. The results indicate that the accuracy of the proposed method is generally higher than 90% for three failure modes, which is 10% higher than the classical machine learning methods and 20% higher than the traditional classification methods. The shear-span ratio and longitudinal reinforcement index have significant influences on whether the RC column fails in flexure. They are followed by the transverse reinforcement index and s/h₀, while the axial load ratio has negligible influence. The longitudinal reinforcement index has significant influence on whether the RC column fails in flexure-shear. It is followed by the shear-span ratio and s/h₀, while the transverse reinforcement index and axial load ratio have negligible influence.