Weld Microstructural Image Segmentation for Detection of Intermetallic Compounds Using Support Vector Machine Classification

Abstract

Weld microstructural images reveal information of presence of intermetallic compounds (IMC) and IMC layer width in the weld region. This is an important characteristic in evaluating the joint strength. With the evolution of machine learning approaches, automation of quality testing has drawn attention in the manufacturing lines. This effectively reduces the maintenance time. In this paper, an attempt has been made for pixelwise segmentation of microstructural images using support vector machine (SVM) classification. Segmentation could be used to detect the locations of intermetallic compounds and IMC layer width of the joint. The extracted pixel features such as color and texture of the weld microstructural images are used to train the SVM classifier model. The proposed SVM model is able to segment the intermetallic compounds from the base metal microstructures in the weld region with greater accuracy. Further, simulated SVM model results are in good coherence with the experimental results.