Abstract: In order to solve an existing problem in non-destructive testing of defects in graphite electrode, an electrode defect detection method based on the combination of X-ray detection and deep belief network is proposed in this paper. For this reason, an exposure curve of graphite electrode is made for the graphite electrode, and the exposure curve is used to select reasonable process parameters for prefabricated electrode blocks with different defects, and X-ray inspection is also carried out for the graphite electrode. The de-noising enhancement method is then utilized to improve the weak contrast and low signal-to-noise ratio of the electrode ray image, and the improved Canny operator edge detection method is also adopted to extract the edge of the electrode ray image, then the defect features can be selected and calculated easily. This is due to application of the deep belief network to identify intelligently electrode defects. All results from this research make us believing that it is feasible to apply the radiographic technology to detect graphite electrode defects. The improved Canny operator may be used to extract defect edges. The quality of the extraction has been improved greatly, and the accurate calculation of the characteristic parameters may be realized as a result. The deep belief network can be employed to identify electrode defects with an accuracy of 96.67%, which is a higher accuracy rate indeed.

Key words: X-ray detection, graphite electrode, feature extraction, deep belief network, defect identification