Pressure vessel weld defect identification method and device based on neural network

Abstract

The invention discloses a pressure vessel weld defect identification method and device based on a neural network. A computer reads in a radiographic inspection negative film image, and whether the quality of the negative film image is qualified or not is determined according to the read-in average darkness value of the negative film image; the negative film image is enhanced according to a fuzzy theory, and de-noising processing is conducted on the enhanced negative film image according to median filtering; a weld joint region is separated from the negative film image obtained after de-noising processing is conducted according to the morphological feature, and an original weld joint gray value is retained; defect filtering is conducted on the separated negative film image according to a mean filter, and a simulated ideal weld joint is formed to serve as a virtual background, the virtual background is subtracted from the original weld joint, and the weld joint defect is obtained after enhancement and de-noising are conducted. The invention further discloses the pressure vessel weld defect identification device, radiographic inspection weld joint images can be identified effectively, identification accuracy reaches 92%, the detection cost can be reduced greatly, and efficiency is improved.

Description

technical field [0001] The invention relates to the technical field of image processing and defect recognition of radiographic weld seam, in particular to a neural network-based method and device for identifying weld seam defects of pressure vessels. Background technique [0002] With the rapid development of radiographic inspection and image processing technology, the detection of weld defects has gradually transitioned from manual evaluation to computer intelligent detection and identification. Using a computer to analyze and identify the digitized weld seam image not only improves the detection efficiency and economic benefits, but more importantly, the method is convenient, practical and easy to operate. Therefore, how to improve the intelligence, automation and quantification of radiation detection is a hot issue in the current research of radiation detection technology. [0003] Due to the characteristics of radiographic inspection itself, the generated radiographic i...

AI Technical Summary

Problems solved by technology

[0003] Due to the characteristics of radiographic inspection itself, the generated radiographic inspection images have the characteristics of low contrast, blurred defect edges, high image noise, and large background fluctuations. Therefore, how to quickly and accurately extract defects in radiographic inspection weld images , and its characteristics, and its classification and identification become the main problem

[0004] When using the traditional identification system to process the information about the quality of the weld, the geometric features of the known defects will be collected in advance, and the corresponding standard template will be designed, and then the image to be identified will be normalized according to the size of the template, and finally used All templates are carefully compared with the images to be recognized, and the recognition results are inferred according to the similarity. However, due to the complex shape of the defect, it is difficult to create an ideal template, so it is impossible to perform accurate recognition of weld defects.

Images