The research team led by Dr. Seong-hoon Kang in the Department of Materials AI & Big-Data at Korea Institute of Materials

Science utilized artificial intelligence to develop a technology to accurately detect microscopic

defects on the surface of an object which may not be detected with a bare eye. 

Visual inspection requires a lot of time and cost since microscopic defects appear in various forms. In particular, as the surface 

of a part with a complex hierarchical structure is three-dimensional, it was difficult to accurately inspect with a microscope or

naked eye.

To solve the issue, the researchers developed an artificial intelligence (AI) technology that automatically transfers the three

dimensional shape into a two-dimensional image of a hierarchical structure and detects microscopic defects that deviate from

the standard image.

The artificial neural network performs learning tasks to build the knowledge of the characteristics of defective factors and then

conducts defect detection and classification. Since microscopic defects, about half the thickness of a hair (50㎛), cannot be

accurately detected with the naked eye, the technicians have to take dozens of photos per specimen and inspect with a

microscope one by one. But the technology now allows accurately detecting a defect in only about 1 second.

Aside from defect detection, Class Activation Mapping (CAM) was introduced to increase the reliability of microscopic defect

detection as the user can find out what kind of surface characteristics was selected by the artificial intelligence (AI).

Dr. Seong-hoon Kang, the principal researcher, said, “The process automation and smart factory are emerging issues.” “If the 

technology is applied to various automated processes, microscopic defects which are difficult to detect with the naked eye can

be detected swiftly.”

The research was funded by a project to establish a virtual engineering platform for lightweight materials for vehicles of the 

Ministry of Trade, Industry and Energy as well as supported by the Fundamental Research Program of KIMS.