RESEARCH ON INTELLIGENT DETECTION OF DEFECTS IN HTR-PM GRAPHITE AND CARBON COMPONENTS BASED ON DEEP LEARNING
Jintao Fu,Renjie Liu,Yuewen Sun,Tianceng Zeng,Haoyu Liu,Shuo Xu,Peng Cong
DOI: https://doi.org/10.1299/jsmeicone.2023.30.1268
2023-01-01
The Proceedings of the International Conference on Nuclear Engineering (ICONE)
Abstract:Graphite and carbon components are important materials for the core and reflector of a high temperature gas-cooled reactor-pebble bed module (HTR-PM). Expansion to the critical point will cause the components to rupture and fall off, which will pose a huge threat to the structural safety of the HTR-PM and its safe and stable operation. Therefore, before the reactor is built, the defect detection of graphite and carbon components is particularly important, which is not only the basic guarantee for the long-term safe operation of the nuclear reactor but also an important support for predicting the fatigue state of irradiated materials based on the pre-service data of the reactor to evaluate the life of the reactor. Considering the large volume of reactor components, long production cycle, and small internal defect volume, the defect detection accuracy and efficiency are required to be high, and traditional non-destructive testing methods such as ultrasound, eddy current, and X-ray are no longer applicable. Multi-slice spiral CT technology has the advantages of high detection accuracy and fast detection speed and is suitable for 3D digital imaging of graphite and carbon components. However, the amount of reconstructed data from spiral CT images is huge, and the manual image review based on visual inspection has shortcomings such as false detection and missed detection. Therefore, it is particularly important to intelligently detect defects. In this paper, the detection images of graphite and carbon component samples by multi-slice spiral CT system are used as the dataset, artificial defect annotation is carried out, and the deep learning algorithm based on YOLOX is used to realize the intelligent detection of defects of large-size graphite and carbon components in HTR-PM. The results show that the YOLOX algorithm achieves good performance in defect detection of graphite and carbon components, and can achieve 96.4% accuracy and 64.85 fps value, with high detection accuracy and detection efficiency.