| 학회 | 한국재료학회 |
| 학술대회 | 2021년 가을 (11/24 ~ 11/26, 경주 라한호텔) |
| 권호 | 27권 2호 |
| 발표분야 | D. 구조 재료 분과 |
| 제목 | Image preprocessing and unsupervised learning classification for deterioration grade determination |
| 초록 | In order to evaluate the deterioration degree of power plant equipment, the damage type and deterioration grade are evaluated by comparing the metal surface with the reference image. Although it is intended to obtain the lifetime consumption rate using the deterioration grade, there are several problems in evaluating the deterioration grade. First, the images used to classify the degree of deterioration are very different from the actual driving environment. Second, human error may occur, and there may be a problem that the standards for grades are different. The last problem is that the existing images must be provided with various images according to the grain size and magnification for each grade in order to be classified accurately. In order to solve these problem, it was necessary to develop an algorithm that can prevent human error and make a quick diagnosis by applying an artificial intelligence technique to classify the image when analyzing it. In this study, samples of x20 steel grades 24h, 96h, 503h, 2015h, 9501h, and 10502h steam oxidation were used for image analysis. After acquiring a surface image and binary imaging, the area and perimeter of each particle were analyzed. The obtained image was able to analyze the area and circumference of the black particles with a binary technique that separates them into two colors, white and black. It can be seen that each particle is correctly calculated sequentially. Image classification using the merge clustering technique was performed based on the number, area, and perimeter characteristics of each analyzed image to rank the overall degradation degree. As the deterioration progresses, it can be seen that the spacing of the laths gradually increases and the average area and the perimeter tend to decrease. Through this classification technique, there is an effect that can almost zero the errors that occur from the results of the existing experts. Image classification is combined with quantitative operation history data to be used for accurate diagnosis of equipment. |
| 저자 | 오지은, 이한상, 김범신 |
| 소속 | 한전 전력(연) |
| 키워드 | <P>열화도; 열화 등급; 이미지 처리</P> |
