Research on automatic identification method for pipeline girth weld defects based on X-ray images and sparse representation

Objective Girth weld quality is considered a significant factor influencing the safe operation of pipelines. X-ray testing serves as a vital technology for identifying weld defects. Nevertheless, differentiating between the feature values of X-ray images for pipeline girth weld defects and noise poses a considerable challenge. Methods This paper proposes a high-accuracy automatic recognition method for X-ray images of pipeline girth weld defects. Based on Suspected Defect Region (SDR) and formulated gray densities, a clustering-based SDR segmentation algorithm was constructed, aimed at precise segmentation of defect SDRs in various shapes. To ensure a high success rate of image recognition after segmentation, judging whether an X-ray SDR image represents a defect was treated as a pattern recognition process. By considering SDR images to be evaluated as a linear combination of sample SDR images (i.e., dictionary matrix), the approach of deriving coefficient vectors was employed to determine whether the segmented SDR image signifies a defect. To facilitate judgments with sparse coefficient vectors, coefficient vectors were solved by 0-norm optimization. In addition, the inclusion of a smooth and differentiable function with 0-1 penalty terms made it possible to solve 0-norm optimization using a penalty function method. To maximize the library of image features within the dictionary matrix, an optimal model was established for X-ray SDR images of welds based on orthogonal optimization, along with a dictionary matrix solving algorithm featuring orthogonal optimization. Results The established model and algorithm led to the development of pipeline girth weld defect detection software integrating sparse representation and X-ray image detection technology. This software was used to recognize X-ray images of weld defects for a pipeline with a diameter of 762 mm and a wall thickness of 10.3 mm. The outcomes demonstrated an impressive defect detection rate of up to 98%. Conclusion The proposed defect identification method demonstrates its capability to greatly improve the detection quality and efficiency of potential safety hazards in pipeline girth welds, underscoring its promising prospects in industrial applications.