Quantitative detection for textured surface cracks based on cluster integration

Abstract

The zirconium plate is one of the essential components in nuclear reactor systems. During the manufacture of the zirconium plate, cracks occur on the surface inevitably suffering from the limitation of industrial technology, causing the failure of the zirconium plate and even leading to the leakage of nuclear fuels. Thus, the precise detection of cracks topography on zirconium plates is significant. However, when the frequency of the horizontal line structure is intensive, and the pattern of cracks is diverse, it is challenging to exactly detect and identify cracks. Therefore, the paper proposed a quantitative detection scheme, including a two-phase cluster integration strategy (TPCIS) and trunk reconstruction measurement evaluation to segment defect areas precisely and measure morphological characteristics of cracks quantitatively. In addition, this paper utilized artificial cracks as a dataset to analyze the performance of the algorithm. The adequate experiments demonstrate that the TPCIS achieves an average detection rate of 92.9%. For measuring the area, length, and width of cracks with the width of 50 µm, the proposed method can control the relative errors within 1.5%, which are of higher accuracy and stability compared with previous methods.