Proceedings of the National Seminar & Exhibitionon Non-Destructive Evaluation

NDE 2011, December 8-10, 2011

AUTOMATIC DEFECT RECOGNITION (ADR) SYSTEM FOR REAL TIMERADIOSCOPY (RTR) OF STRAIGHT TUBE BUTT (STB) WELDS

Deepesh.V1, R.J. Pardikar1, K.Karthik 2, A. Sricharan 2 S. Chakravarthy 2 and K. Balasubramaniam 2

1 BHEL, Tiruchirapalli-620014,India2 Indian Institute of Technology Madras, Chennai 600036,India

ABSTRACT

Non Destructive Evaluation (NDE) Methods, in particular Digital Radiography(DR), incorporated with AutomaticDefect Recognition (ADR) ,for industrial applications is a rapidly progressing area of research across the globe.Though ADR technology has been well established for Digital Radiographic inspection of cast and machinedcomponents, ADR is still considered a challenge in the case of many types of weld joints, mainly due to the non-uniformity in the radiographic images of weld joints. This paper introduces an indigenously developed AutomaticDefect Recognition System for Real Time Radioscopy (RTR) of Straight Tube Butt Weld(STBW) joints, which are thecritical joints of tubular components like Economiser,Super heater and Reheater of a Boiler.RTR system for inspectionof STB welds consists of a constant potential X-Ray equipment with swiveling arrangement, as the X-Ray source,Digital Flat Panel (DFP) as the Imaging device with it’s associated Image acquisition and Review Software, alongwith the ADR software for Defect recognition, classification and thereby evaluation of STB welds. ADR Algorithm,scans through the Digital X-Ray image of the STB Weld joint, and detects the defects present and takes the decisionof Acceptance / Rejection, based on the Acceptance Standards for STB welds. Artificial Neural Network (ANN)techniques enable the Algorithm, continuously learn and evolve as a more efficient program, with each joint it evaluates.This will enhance the reliability of defect detection and evaluation. The preprocessing uses concepts from digitalimage processing, image analysis, and pattern recognition. The development of this system involved validation witha wide range of weld samples with various types of discontinuities. The system has been implemented in one of theRTR stations in BHEL and the ANN training has so far resulted in over 95% accuracy level. This system replaces thehitherto used, humane evaluation and removes its inherent limitations like subjectivity, inconsistency, fatigue etc andaccomplishes a faster and more reliable evaluation.

Keywords: Automatic Defect Recognition, Real Time Radioscopy, Digital Radiography, Digital Image Processing,Pattern Recognition, Artificial Neural Network, Radial Basis Function.

INTRODUCTION

Radiography is very well established as an NDT technique,using both film and electronic X-Ray detection systems.Mainly used in the petroleum, petrochemical, nuclear andpower generation industries especially, for the inspection ofwelds, the radiography has played an important role in thequality assurance of the piece or component, in conformitywith the requirements of the standards, specifications andcodes of manufacturing. Most radiographic exposures and filminterpretations in RT are still carried out manually (1). Humaninterpretation of weld defects, however, is tedious, subjectiveand is dependent upon the experience and knowledge of theinspector (2).Human inspectors are not always consistent andeffective evaluators of products because inspection tasks aremonotonous and exhausting. It has been reported that humanvisual inspection is at best 80% effective. In addition,achieving human ‘100%- inspection’, where it is necessary to

check every product thoroughly, typically requires high levelof redundancy, thus increasing the cost and time forinspection(3).Here comes the importance of Automation ofevaluation , which reduces human involvement, thus makingthe inspection more reliable and faster.

REAL TIME RADIOSCOPY (RTR) OF STB WELDS

Tubular products form an important part of Steam Generatorsin thermal power plants. These include mainly Super heater,Reheater, Economizer coils, water wall panels etc. Thesecomponents consist of tube assembly of several meters lengthand the required length is achieved by Straight Tube Butt (STB)welding process. These tubes are made of carbon steel, alloysteel etc. It is a pulsed Metal Inert Gas welding process usingspray type of metal transfer both at average current levels andlow current levels which are very much suitable for out ofposition welding and thin gauge material welding. The major

356 Deepesh et.al : Proceedings of the National Seminar & Exhibition on Non-Destructive Evaluation

defects which occur in this weld are Porosity, Gas hole,incomplete penetration, Lack of fusion, Excess penetration,Burn through, etc.

STB weld joints are subjected to online monitoring systemfor quality assessment. This system is called Real TimeRadioscopy system (RTR).These systems consist constantpotential duel focal (large focal size 3 mm x 3 mm and smallfocus 0.8x 0.8 mm) X-Ray machine with capacity 320kV,10mA as the X-Ray source and Digital Flat Panel Detector(FPD) as the Imaging device .X-Rays from the source penetratethrough the weld thickness and the differential absorption ofradiation gives a two dimensional X-Ray image of the weldjoints. Incoming X-rays first strike a Cesium Iodide scintillatorof the Digital Flat Panel, which converts the X-Rays intolight.The light then passes through a photodiode matrix ofamorphous silicon, and get converted to electrical signals,which are amplified and digitized (4).The light is directed ontothe silicon without lateral diffusion, which ensures imagesharpness (5).The digital data is then processed into imagesvia a corresponding gray value table, and is displayed, printedor sent to computer as required. The system offers theadditional advantages of image post-processing and archiving.Compared to other imaging devices FPD provides high qualitydigital images, better signal to noise ratio and dynamic rangeof 12 to 16 bit (6), which provides high sensitivity forradiographic application. The present RTR system in BHELuses an Amorphous Silicon Flat Panel (model:

DXR250RT).The Images obtained by Flat Panel ImageAcquisition and Review computers are in DICONDE format.

These images are evaluated by experienced, qualified NDEpersonnel, decision of acceptance /rejection taken as perstandards (7) and the feedback is given to the welder. Thethickness range usually is 4 mm to 12 mm .The Radiographictechnique used here is Double Wall Double Image (8).

AUTOMATIC DEFECT RECOGNITION (ADR) OF STBWELDS

The manual evaluation has certain limitations like subjectivity,and humane dependency, which affect the productivity andreliability. Here comes the importance of Automation inevaluation, which reduces human involvement, thus makingthe inspection more reliable and faster. The ADR system scansthrough the Digital X-Ray Image of the STB weld joint, andrecognizes the defects and takes the decision of Acceptance /Rejection, based on the Acceptance Standards.ADRtechnology is already available for Castings (9) especiallyAluminuium Wheels, Magnesium components (10) and weldjoints (11).There are also ADR systems available for general

Figs (1&2): RTR System and Control Station

Fig. 3 : RTR image of STB welds

Fig. 4 : pictorial representation of defect recognition stages.

NDE 2011, December 8-10, 2011 357

NDT methods (12) .However, there is no customized packageavailable as such, for ADR of Straight Tube Butt welds. Thistriggered the development of an ADR system for integrityassessment of STB welds. The new system is different fromother ADR systems in the aspect that, it uses different detectionapproaches for different class of defects where as many of theother ADR systems do defect classification only after defectdetection.

ADR ALGORITHM (13-19)

The first stage of operations performed by ADR algorithm ispreprocessing of the input image. This stage involvesenhancement of the image properties to such a level that, thepattern recognition can be executed without errors. In the caseof the STB weld image, the Region of Interest(ROI) includesthe elliptical weld region and adjacent raw material region ofjoining tubes. The next step is the extraction of the ROI, byfinding it’s boundaries by analyzing the vertical summationof the extracted tube for regions of heightened activity.

Table 1 : ADR trial results

Type No of test No of Correctsamples Classifications Percentage

Category I(Gas Hole &Pores) 560 546 97.5%

Category II(ICP & LF) 95 92 96.8%

Category III(Burn Through& Excess Penetration) 100 85 85%

Non defective joints 840 823 98%

The code was tested initially on a set of 1500 weld imagesamples .These welds include non defective as well asdefective weld joints covering a wide range of defects ofdifferent size,shape,location and orientations. The results ofthe trial are shown in table 1.0.Defect recognition algorithmbased on RBF can efficiently overcome the shortcomings oftraditional methods, which require large number ofsamples.

Fig. 6 : ROI extracted images of welds with BT,GH and ICP respectively,Binary image of ICP feature extraction

Fig. 5 : Input image before & after contrast enhancement ,ROI extracted from contrast enhanced image

358 Deepesh et.al : Proceedings of the National Seminar & Exhibition on Non-Destructive Evaluation

REFERENCES1. F Fucsok and M Scharmach, ‘Human factors: The NDE

reliability of routine radiographic film evaluation’,Proceedings of 15th World Conference on Non-Destructive Testing, Roma 2000.

2. T Y Lim,MM Ratnam and M A Khalid, “Automaticclassification of weld defect using simulated data and anMLP neural network”,Insight Vol 49 No.3 ,march 2007,p154-159

3. Domingo merry, Workshop on Digital Radiography, GEGlobal Research Centre, Bangalore, June 27, 2005

4. Dr.P.R.Vaidya-Flat Panel Detectors for IndustrialRadiography-International Workshop on Imaging NDE-2007, April 25-28, 2007,Kalpakkam,Chennai,India.

5. Giakos,G.C.; Suryanarayanan,S.; Guntupalli,R.Odogba,J.; Shah,N.;Vedantham,S.;Chowdhury,S.;Mehta,K.; Sumrain,S.; Patnekar,N.; Moholkar,A.;Kumar,V.; Endorf,R.E.,‘Detective quantum efficiency ofCZT semiconductor detectors for digital radiography’,Instrumentation and Measurement,IEEE TransactionsVolume 53, Issue 6, Dec. 2004, P 1479 – 1484.

6. V.R.Ravindran, ‘Digital Radiography Using Flat PanelDetector for t e Non-Destructive Evaluation of SpaceVehicle Components’, Journal of Non-DestructiveTesting & Evaluation , Vol.4, Issue 2, September 2005.

CONCLUSIONS AND SUMMARY

Usually the approach for common ADR algorithm is therecognition of the defect, followed by classification based onthe features. However this Algorithm applies differenttechniques for detection of different classes of defects andhence defect detection and classification are parallel processes.The trials carried out gives good results for class I(Gas holeand porosity) and class II(ICP and LF).However theperformance needs to be improved in the case of the thirdcategory. ANN training with sufficient quantity of images isbeing carried out to enhance the detection level further.Application of other networks like support vector machine(SVM) is also being studied so that probability of detectionof class III can be enhanced. In the case of training samples,the size of the sample with one class of defect should alsotake in to account the probability of occurrence of that classduring actual field trial. It is also important to note that RTRoffers real time image and by rotation of the tube andswiveling, different images of the same weld joint can becaptured. Some defects which are not prominent in one imagebecome prominent in a different orientation of the weld withrespect to the source and detector. Hence a defect detectionapproach based on multi image or direct frame analysis of avideo image can enhance the detection and classification levelsignificantly.

Fig. 7 : ADR GUI during normal operation

NDE 2011, December 8-10, 2011 359

15. Daum W, Rose P, Heidt H, Builtjes JH. Automaticrecognition of weld defects in X-Ray inspection. Br JNDT 1987;29(2):79–82.

16. Liao TW, Li DM, Li YM. Extraction of welds fromradiographic images using fuzzy classifiers. Inform Sci2000;126:21–42.

17. Kato Y, Okumura T, Matsui S, Itoga K, Harada T,Sugimoto K, Michiba K, Iuchi S, Kawano S.Development of an automatic weld defect identificationsystem for radiographic testing. Weld World 1992;30(7/8):182–8.

18. Gayer, A, Saya, A, Shiloh, A. (1990). Automaticrecognition of welding defects in real-time radiography.NDT International, 23(4):131–136.

19. Lawson, S W, Parker, G A, (1994). Intelligentsegmentation of industrial radiographic images usingneural networks. In Machine Vision Applications andSystems Integration III, Proc. of SPIE, volume 2347,pages 245–255, November 1994.

7. ASME Section I.8. ASME Section V.9. Frank Herold, Rolf-Rainer Grigat, “A New Analysis and

Classification Method for Automatic Defect Recognitionin X-Ray Images of Castings”, Paper presented at the8th ECNDT, Barcelona, June 2002

10. Veronique Rebuffel, Subash Sood, “Defect DetectionMethod in Digital Radiography for Porosity inMagnesium Castings”.

11. G. Bonser and S. W. Lawson, “Defect detection inpartially complete SAW and TIG welds using on-lineradioscopy and image processing”, Miguel Carrasco andDomingo Mery, “Segmentation of welding defects usinga robust algorithm”.

12. US Patent No: 4896278,dt: 23rd January1990.13. Inoue, K. and Sakai, M., “Automation of inspection for

weld”, Trans. Of Japanese Welding ResearchInstitute,Osaka University, vol. 14(1), pp. 35-44, 1985.

14. Aoki L, Suga Y. Application of artificial neural networkto discrimination of defect type in automatic radiographictesting of welds. ISIJ Int 1999;39(10):1081–7.