A Study on the Development of Deep Learning Algorithm for Determining External Quality of Welded Parts Using Transfer Learning

Recently, deep learning has been applied to various welding techniques, such as laser welding, gas metal arc (GMAW), and resistance spot research on automation quality prediction is being conducted. Even for GMAW, many researchers have attempted predict through X-ray, current, voltage measurements. If judgment in real time not necessary, it most effective judge the of a welded part using an exterior image. Therefore, this study, appearance model was analyzed image learning. Welding defects were classified into pores, overlaps, craters, melting base material, cracks, undercuts, divided 7 categories including normal ones. In constructing deep-learning model, transfer performed existing networks, ResNet AlexNet. To improve accuracy learning, tests while optimization technique, maximum number epochs, minibatch size changed. It confirmed that weld defect improved increased, stochastic gradient descent had highest accuracy. Increasing data should make technique images GMAW welds proposed more widely applicable.