Overview. Optimization of Industrial Radiography Technique for Defect Detection of Oil and Gas Pipelines in Weld Regions by Image Processing

Transcription

1 Optimization of Industrial Radiography Technique for Defect Detection of Oil and Gas Pipelines in Weld Regions by Image Processing Alireza karimian (University of Isfahan- Isfahan - Iran) Monir Torabian (Azad University- Najaf abad- Iran) Mohammadreza Yazdchi (University of Isfahan- Isfahan - Iran) Overview What s s NDT Radiography testing method Weld defects The way for detection and automatic classification of defects Results Conclusion What s s NDT NDT is the use of physical methods which will test material, components and assemblies for flows in their structure without damaging their future usefulness 1

2 Radiographic Testing Method Definition Radiography is a method of NDT which uses penetrating radiation (x-ray, gamma ray, neutron) to produce a graphic record on film which indicates comparative soundness of the material under test. Basic Radiographic Testing. There are three basic essentials for the radiographic testing process, namely; Radiation source. X-ray machine, Ir-192 and Co-60. Object or specimen. Welded plate and pipe, casting, etc. Film. Industrial grade x-ray x film. Examples - Fuji, Agfa Geavert,, Kodak, etc. Radiographic Arrangement (source, specimen and film) Radiation source Specimen Radiographic film. Weld defects Transverse cracks Incomplete or Lack of Penetration Porosity Burn Through 2

3 Transverse Crack Incomplete or Lack of Penetration Porosity 3

4 Burn Through In this research work a method for detection and automatic classification of four kinds of weld region defects (i.e. transverse crack lack of penetration porosity and burn through ) has been suggested. The research was done in the following steps: 4

5 A: Image digitization The images inside the film* were converted to digital images (JPEG format) by using HR scanner Microtek 9800 XL 2000 DPI * Two sets of RT images from NIGC( National Iranian Gas Company) and IIW (International Institute of Welding) were used. B: ROI selection Region Of Interest (ROI) could be the total welded region or the defect (semi-defect) region ROI 5

6 C: Image preprocessing Histogram equalization C: Image preprocessing Median Filter P1 P2 P3 3 χ 3 mask P4 P5 P6 P7 P8 P9 Median value p1, p2, p3, p4, p5, p6, p7, p8, p9 6

7 C: Image preprocessing Histogram specification (Gray levels slicing) 256 gray level will classified for example in 5 levels 7

8 D: Image segmentation In this work, Global threshold is used to recognize the defect from other region. Global method computes a single threshold value for the entire image and then pixels having a gray level value less than the threshold value are marked belong to one class, otherwise the other class g ( x, y)= 1 0 if f( x, y) T if f( x, y) T New Value of pixel New Gray level of pixel E: Features extraction extraction of geometric features of the weld bead was done by region growing method To do this a pixel (seed) inside the (ROI) is selected by user The pixels with the same or semi gray scale join to each other and produce the ROI 8

9 Seed Growing Creating the ROI and defect region inside the smallest rectangular Length Defect Geometric Features Roundness (measuring its degree of the circumference ) P=perimeter & A=Area of defect L W Defect area/smallest perimeter rectangular area R=P 2 /4ΠA I=ΠD 2 /4A Width A/AC θ Defect angle Lengthening Index (measuring its degree of the Elongation) D=Diameter of defect * 9

10 L W * A=L*W 10

11 * 0 درجه * D P is estimated by the number of pixels around the defect * P 11

12 F: Defect Classification multi-layer preceptron:mlp Back-propagation(BP) 10 Neuron L W I R θ A A C 5 Neuron 7Neuron Neural network input vector To train the neural network we used from 200 images including one or more defect (50 images for each defect) and 50 non-defect images 12

13 Defect name Some RT films from NIGC and IIW including defects used to check our method and algorithm which their results are as the followings: Transverse crack 13

14 Image digitization ROI selection Histogram equalization Median filter used Histogram specification Threshold Creating the smallest perimeter around the defect Defect classification Assigning and calculating the perimeter of defect Porosity 14

15 Image digitization ROI selection Histogram equalization Median filter used Histogram specification Threshold Creating the smallest perimeter around the defect Defect classification Assigning and calculating the perimeter of defect Burn through 15

16 Image digitization ROI selection Histogram equalization Median filter used Histogram specification Threshold Creating the smallest perimeter around the defect Defect classification Incomplete or Lack of Penetration 16

17 Image digitization ROI selection Histogram equalization Median filter used Histogram specification Threshold Creating the smallest perimeter around the defect Defect classification Assigning and calculating the perimeter of defect Conclusion The suggested method in this research work can detect and assign the defects for more than 90% of RT films Most of the RT machines in Iran are analogue (not digital) and therefore need experts for film description Our method can help the experts and moreover the users in the lack of expert The suggested method is cheap in comparable with digital RT 17

18 Acknowledgment The authors wish to thank from the National Iranian Gas Company (NIGC) for supporting this research Thank you 18