Session: CT-08-01 The L. Eugene Hulbert Memorial Session on New and Emerging Methods of Analysis and Applications

Paper Number: 106113

106113 - Defect Recognition and Classification From Ultrasonic Phased Array Total Focusing Method Imaging Based on Random Forest 

Pipeline transportation is one of the principal energy transportation methods. The safety performance of pipelines in construction and operation is closely related to the quality of welds. If the pipeline weld defects cannot be accurately identified promptly, late maintenance may cause leakage arising from the welding defects, which will give rise to considerably serious consequences. Ultrasonic phased array inspection is a widely used technology in non-destructive testing, which can significantly improve the sensitivity, coverage and accuracy of detection. In particular, the development of total focusing method imaging can obtain high-resolution images of cracks and other weld defects in pipeline. At the moment, imaging results necessitate manual judgment of defect type, which is highly dependent on the prior knowledge of the inspector. This leads to measurable amount of misjudgment, especially for the long-distance pipeline defect detection in which the amount of detection data is large and the imaging features are difficult to be recognized. Consequently, the automated recognition and classification of the defects is the emerging approach to resolve those. This study focuses on the linear defects and volumetric defects in welds(i.e. cracks and pores). Typical test blocks containing defects are fabricated and an ultrasonic phased array is employed to generate the images of the defects by using total focusing method (TFM) algorithm. To expand the sample size for small number of TFM results, the data augmentation approach is proposed and implemented. Then, the extracted features are input into the random forest algorithm, followed by debugging and optimization. The results suggest that the recognition and classification of defects from images are  realized, and the accuracy rate of defect recognition is more than 90%.

Presenting Author: Haibin Wang Hefei General Machinery Research Institute

Presenting Author Biography: Wanghaibin is currently working at Hefei General Machinery Research Institute as a engineer.His research interests mainly include pipeline monitoring,phased array imaging and machine learning.

Authors:

Haibin Wang Hefei General Machinery Research Institute
Zhichao Fan Hefei General Machinery Research Institute
Xuedong Chen Hefei General Machinery Research Institute
Jingwei Cheng Hefei General Machinery Research Institute
Wei Chen Hefei General Machinery Research Institute
Zhe Wang Hefei General Machinery Research Institute
Yangguang Bu Hefei General Machinery Research Institute