Comparison of Logistic Regression and Support Vector Machine for the Classification of Microstructure and Interfacial Defects in Zircaloy-2

β-quenched Zircaloy-2 is a candidate material for Pressurized Heavy Water Reactors. Before the fabrication of PHWR components from Zircaloy-2 billets, it is very important to ensure that that the billets have undergone proper β-quenching treatment and they have the appropriate β-martensite structures. Improper β-quenching treatment given to the billets is strictly unacceptable. Previously, destructive techniques and semi-destructive techniques such as metallography and hardness measurements used to be employed to ensure proper quenching. Recently, ultrasonic velocity and attenuation measurement based techniques have been used with varying degree of success. In this work, Logistic regression and Support vector machine techniques have been attempted to ensure proper β-quenching treatment of Zircaloy-2 microstructures and for the classification of heat treatment given to the Zircaloy-2 specimens. The acquired A-scan signals (time domain signals) meant for ultrasonic measurements have been utilized in this study. Results indicate that both Logistic regression and SVM techniques can be effectively utilized for comparative quick classification of microstructures of Zircaloy-2 specimens, out of which SVM approach gives better confidence level than logistic regression to ensure proper classification of microstructures of β-quenched Zircaloy-2 specimens.