미세조직 결정립 크기 측정에 대한 기계학습

Abstract

Machine learning for grain size measurement from microstructure images Jun ho Jung Advisor: Prof. Heesoo Kim, Ph. D. Dept. of Advanced Materials Engineering Graduate School of Chosun University Observing microstructure is the most basic way to determine the properties of materials.The mechanical and electrical properties of metallic materials vary depending on their microstructure. The microstructure of a material is not only deeply related to physical properties and mechanical behavior, but can also help optimize the manufacturing process or prevent damage to material, so observing the microstructure of material can be said to be a fundamental step in materials engineering. The method of observing microstructure generally involves grinding a target specimen, etching it with a solution, and observing it with an optical microscope. This work requires a lot of time and effort, and there is a possibility that subjective judgment may be involved. However, the average grain size can be accurately obtained by dividing the entire microstructure image by the number of grains. Unlike the mechanical properties or composition of the material, the grain size of the microstructure is information that can be obtained directly from the image, so it is suitable for using convolutional neural networks. A convolutional neural network is an artificial neural network that performs an image operation called convolution. It is also used in materials engineering. Studies were conducted with the expectation that it would be possible to measure the grain size of the microstructure. The goal of this study was to evaluate the performance of convolutional neural networks applied to materials engineering and to understand how convolutional neural networks recognize the characteristics of microstructure.