Printer Friendly

Wavelet based feature extraction method for quantitative characterization of porosity in gas tungsten arc welding by infrared thermography in AISI 316 stainless steel for on-line monitoring and control.


Gas Tungsten Arc Welding (GTAW) commonly referred as Tungsten Inert Gas (TIG) welding is best suited for precision welding in atomic energy, aircraft, chemical and instrument industries. It is an arc welding process wherein coalescence is produced by heating the job with an electric arc struck between a Tungsten electrode and the job. A shielding gas is used to avoid atmospheric contamination of the molten weld pool. TIG welding is one of the widely used methods for joining metals. In spite of the numerous advances in the science and technology of welding, failures do occur and weld is still considered to be the weakest portion. This is because the formation of the weld is affected by a number of process parameters, which make it difficult to ensure the quality of the weld. Conventionally, the quality of the weld is ascertained only after the welding has been completed through the use of Non Destructive Testing (NDT) such as ultrasonic or radiography. Since each of these techniques is applied only after the welding is completed, a lot of time, material and manpower are wasted before one comes to know about the soundness of the weld.

Inherent Limitations in conventional welding processes can be overcome if the weld is continuously monitored in real time for the assessment of defects and their automatic elimination by on-line control of the welding parameters. Moreover, the defective weld can be repaired immediately without continuing the process further. This strategy to monitor, control and maintain quality of welds is commonly known as adaptive welding or intelligent welding. Intelligent welding, as the name implies, combines welding equipment with intelligent sensing and control, knowledge of human experts, and Artificial Intelligence (AI) to improve joining efficiency and reduce the weld inhomogenities and defects.

Sensors are the key to success of intelligent welding. Non-Destructive Testing (NDT) sensors, which have been considered, for on-line monitoring include optical, radiography and Infrared (IR). Of the three, optical sensors provide information limited to the surface such as bead width; misalignment etc. Real time radiography using image intensifier based system has been used in process control of arc welding. However, the hazards involved in the use of radiation sources have limited the potential applications. Infrared, on the other hand, has the advantage that it can reveal surface and near surface perturbations.

After acquiring thermal images, features corresponding to weld defects are extracted. These feature vectors are related to the corresponding deviations in physical parameters responsible for the defect. With this mapping, the respective physical parameter is then controlled to produce defect free welds. This paper presents an algorithm for automatic identification and quantification of porosity.

The paper is organized as follows. Section II gives a brief review of Infrared Thermography in Non Destructive Testing. Section III provides experimental set up. Section VI and V review the feature extraction algorithm and wavelet Transforms. Section VI deals with the methodology. Section VII provides the Results and Section VIII concludes the work. The functions are implemented in Mat lab.

Related Work

Infrared thermography is not a new technique for on-line weld monitoring. Numerous groups worldwide have used Infrared investigation techniques in the inspection of subsurface defects and features, thermo physical properties, coating thickness and hidden structures. Thermographs are used to the control of welding process problems, such as arc misalignments [1]. Infrared sensors are best suited for weld quality detection as the perturbations that arise due to variations in arc positioning, heat input and the presence of contaminants distinctly manifests itself as differences in the spatial and temporal surface temperature distributions. Hence image analysis techniques can be developed to quantify the changes in the temperature distribution there by enabling adaptive welding techniques for automated weld control [2]. Infrared thermography was used for on-line control of torch path in Robotic Gas Tungsten Arc Welding. An infrared camera was used to record the temperature variations surrounding the welding torch. These images were transmitted to a central computer where an image processing algorithm was developed to determine the torch from the joint and also transmitted the corrective action to control the torch path. However the developed method was suitable for only single V-groove configurations [3]. Infrared Sensing and Computer image processing techniques can be used as a feasible method to improve the welding process through dynamic control of joint penetration parameters. An infrared camera was mounted on the front side of the weld pool and surface temperature distributions surrounding the weld pool were measured during the welding process. Welding parameters were varied to obtain different depths of penetration and the corresponding temperature distributions were noted. However relative temperature with respect to a specific chosen temperature was considered instead of the absolute temperature [4]. Infrared Sensing techniques were used to track curved contours of joints with a gap in fusion reactor welding. It was found that a gap in a weld joint produce a significant drop in the measured Infrared intensity or temperature and this temperature reduction can be used to determine the size and position of a joint gap [5]. Infrared thermography is highly suitable for sensing variations in bead width and depth of Penetration due to variations in plate thickness, shielding gas composition and minor element content in GTAW. Macroscopic temperature gradient determined from the peak temperature and the temperature at the solid-liquid metal surface was used to implement weld penetration control [6]. In recent years extensive research is carried out on the applications of Infrared thermography for Non Destructive Testing and Evaluation. Thermography offers non-contact, wide area detection of subsurface defects. Different passive and active thermographic techniques are used for defect detection. Active techniques include Pulse thermography, Lockin thermography, Pulsed phase thermography and vibrothermography [7-9]. Infrared thermography was used to determine the transient thermal field that accompanies fusion welding procedure in order to study the out-of-plane distortion. Thermographs reveal significant features of the thermal conditions that cannot be modeled theoretically in a practical sense [10]. Lack of Penetration and Tungsten Inclusion are detected and quantified from thermographs. Quantification of thermographs is achieved by image processing algorithm through histogram equalization, image segmentation and morphological image processing. These features extracted from the algorithm are then used for on-line weld monitoring to produce defect free welds [11]. An inversion based technique has been adopted to extract the features from thermographs. Phase and amplitude of the image are obtained by Fast Fourier Transform and it was found that phase profile for a defect exhibits a distinctive inflexion point at blind frequency and hence the defects are quantified [12]. In contrast to the conventional contrast based methods, Thermographic Signal Reconstruction method was used for the analysis of thermographic data to increase detectability and provide automated pass/fail processing in Pulsed Thermography [13].

Experimental Setup

The experiments were conducted using Precision TIG 375, automatic TIG welding machine. TIG welding or GTAW uses a non-consumable tungsten electrode protected by an inert gas. The electrode is either made of pure tungsten electrode, mixed with small amounts of oxides (Thoria, Zirconia) improving the stability of the arc and makes it easier to strike. Since, the process uses a non-consumable electrode, extra filler material is usually added. The experiment was conducted using 3mm thick American Iron and Steel Institute (AISI) type 316 steel plates measuring 125 x 50 mm in size. The edges and the surfaces of the plates were prepared using standard preparation techniques to facilitate butt-welding. The experiment was performed without filler material. Direct Current technique with electrode connected to negative polarity and workpiece connected to positive polarity. The Infrared camera mounted at an angle of 45 degrees to the weld plate captures the temperature distribution at the weld pool. The Infrared camera obtains thermal maps called thermographs from the distribution and a custom built interface transfers these images from the camera to the computer for further analysis. Later reprocessing is done to expand the thermal scale of the thermographs to standardize them for comparison. The camera detected the infrared radiation used to characterize the thermal distribution of the plates being welded. The infrared camera determines the temperature distribution by sampling a portion of the emitted energy within a wavelength band of 8 to 12 ?m. Each scan of the camera was transferred as an image of size. The frame rate of the camera is 40 ms.

Porosity is a group of small voids that occur mainly due to the entrapped gases. The parent metal melted under the arc tends to absorb gases like Hydrogen, Carbon monoxide, Nitrogen and Oxygen if they are present around the molten metal pool [14]. It is caused by improper electrode or longer arcs or faster arc travel speeds or too low and too high arc currents or incorrect welding technique or electrode with damp and damaged coating or unclean job surface or improper base metal composition. Porosity is introduced deliberately by including grease in the weld plate. In a thermograph, porosity appears as a relatively low temperature region in the hot spot which is the highest temperature region in the thermograph. The acquired thermograph file consists of 1747 frames. Welding arc is present up to 547 frames after which the welding torch is withdrawn. Porosity begins to appear from the 97th frame. As an example, 181, 182 and 183rd frames depicting porosity are shown in Fig.1, Fig.2 and Fig.3 respectively.




Feature Extraction Algorithms-A Review

Image processing algorithms are to be developed to isolate porosity and quantify the porosity. Conventional image processing algorithm involves color to gray level conversion, edge detection by Sobel or Canny filters, morphological image processing operators to isolate porosity, quantitative characterization of porosity. However the performance of this algorithm is dependent on the threshold value chosen for edge detection, size and shape of the structuring elements chosen for morphological image processing. Hence this algorithm involves writing different programs for each thermograph even of the same nature. Moreover Edge detection, dilation, region growing and erosion involve applying mask on every pixel which is a time consuming process. Time is also a very important parameter to be considered as this algorithm is to be used for on-line weld monitoring. Hence it is important to develop a parameter independent, less time consuming standardized algorithm for feature extraction to facilitate on-line weld monitoring. Such a parameter independent algorithm that consumes less time is based on region growing. However Region growing technique also involves choosing the seed pixel value and threshold. Euclidean distance based feature extraction techniques can also be used for extracting porosity directly from color images. But all the above techniques are not suitable if defects are present at different resolutions. Hence a standardized algorithm that is suitable for extracting defects at different resolutions is to be developed. Wavelet Transforms are suited fro MultiResolution Analysis.

Review of Wavelet Transforms

Wavelet transform on the image produces four subband image coefficients. They are the approximation, horizontal detail coefficients, vertical detail coefficients and diagonal detail coefficients.

Let f(t) be any square integrable function. The continuous-time wavelet transform of f(t) with respect to a wavelet [PSI] (t) is defined as

W(a,b) [??] [[SIGMA].sub.-[infinity]].sup.+[infinity]] f(t) (1/[square root of (a)) [PSI]*((t-b)/a)dt (1)

where a and b are real and * denotes complex conjugation, a is referred to as scale or dilation variable and b represents time shift or translation [16].

CWT provides a redundant representation of the signal in the sense that the entire support of W(a,b) need not be used to recover the original signal f(t). A new nonredundant wavelet representation is of the form


This equation does not involve a continuum of dilations and translations; instead it uses discrete values of these parameters. The two dimensional sequence d (k,l) is called as the Discrete Wavelet Transform. The discretization is only in the a and b variables.

The Haar basis is obtained with a multiresolution of piecewise constant functions. The scaling faction is [phi] = [1.sub.[0,1]]. The filter h[n] is given by

H[n] = {[2.sup.-1/2] if n = 0,1 o otherwise (3)

The above equation has two non-zero coefficients equal to [2.sup.-1/2] at n=0 and n=1[17]. The Haar wavelet is

[PSI](t) = { -1 if 0 [less than or equal to] t < 1/2 1 if 1/2 [less than or equal to] t < 1 0 otherwise (4)


The captured thermographs are acquired as moving pictures. The first step involves converting movie file into frames. The initial frames do not contain porosity. Hence porosity quantification begins at nth frame where the porosity appears. On each frame the following operations are performed. The color image is converted into gray scale to avoid computational complexity. Discrete Wavelet Transform (DWT) with Haar wavelets is applied on the gray scale thermographs. Four different images namely approximation image, vertical, horizontal and diagonal detailed images are obtained from DWT. Thermographs are then reconstructed after zeroing the detailed coefficients. Detailed coefficients are zeroed because the abnormality appears as a low-resolution region. Then a threshold is chosen to retain the porosity region. The porosity region is thus isolated. The feature vectors used for describing the abnormality are major axis length, minor axis length, area, mean and variance.


DWT is applied on gray scale images. The final thermographs that depict porosity in frames 183, 195 and 203 are shown in Fig.4, Fig.5 and Fig.6 respectively. The quantitative characterization is as shown in Table 1. The shape of the porosity region is treated as an imaginary ellipse and the major axis length, minor axis length, area, mean and variance are used to describe the defect quantitatively. The shape of the defect is described quantitatively using mean and variance.


The developed wavelet based feature extraction technique clearly identifies and quantifies the porosity. In contrast to the conventional image processing algorithm, this algorithm is parameter independent and is also not image specific i.e. it can be standardized for porosity extraction of any shape and size. Moreover this technique is also suited for various defects that occur at different resolutions in the same thermograph. The time consumed by this algorithm is also much lesser than the algorithm for it does not involve convolution operations as in edge detection or morphological image processing algorithms. The feature vectors extracted from this algorithm can be used for training the neural network to facilitate in on-line weld monitoring. If porosity is caused by variation in physical parameters then the parameters can be varied correspondingly as indicated by the neural network. On the other hand, if porosity is caused by grease or unclean surfaces then welding may be stopped and is allowed to proceed with only cleaned surfaces.





[1] Khan. M.A, Madsen. N.H., Goodling. G.S. and Chin. B.A., 1986, "Infrared Thermography as a Control for the Welding Process", optical Engineering, Vol.25, pp.799-805.

[2] Nagarajan, S., Chen. W.H. and Chin. B.A., 1989, "Infrared Sensing for Adaptive Arc Welding", Welding Journal, Vol.68, pp.462s-466s.

[3] Groom K.N., Nagarajan. S and Chin B.A., December 1990, "Automatic Single V-Groove Welding Utilizing Infrared Images for Error Detection and Correction", Welding research, Supplement to the Welding Journal, pp. 441-s-445-s

[4] Chen W. and Chin B.A., April 1990, "Monitoring Joint Penetration using Infrared Sensing Techniques", Welding Research Supplement, pp. 181-s-185-s

[5] Nagarajan S., Wikle III H.C. and Chin B.A., 1992, "On-line Weld position control for Fusion Reactor Welding", Journal of Nuclear Materials, 191-194, pp.1060-1064

[6] Banarjee P., Govardhan S., Wikle H.C., Liu J.Y, Chin B.A, August 1995, "Infrared Sensing for On-line Weld Geometry Monitoring and Control", Journal of Engineering for Industry, Vol. 117, pp. 323-330

[7] Ibarra-Castanedo. C., Galmiche. F., Darabi. A., Pilla. M., Klein. M., Ziadi. A., Vallerand. S., Pelletier. J.-F., Maldague. X., 2003, "Thermographic nondestructive evaluation: overview of recent progress", SPIE Proc. Thermosense XXV (SPIE: Society of Photo-Optical Instrumentation Engineers), 5073: 450-459, Orlando, X. Maldague, A. Rozlosnik eds., FL.,

[8] Maldague. X., 2002, "Introduction to NDT by Active Infrared Thermography," Materials Evaluation, 6[9]: 1060-1073.

[9] Maldague. X., 2000, "Applications of Infrared Thermography in NonDestructive Evaluation," Trends in Optical Nondestructive Testing (invited chapter), Pramod Rastogi ed., pp. 591-609.

[10] Duncan Camilleri et al, "Use of Thermography to Calibrate Fusion Welding Procedures in Virtual Fabrication Applications", Inframation 2004 Proceedings ITC 104 A 2004-07-27.

[11] Venkataraman B., Baldev Raj and Menaka. M., 2005, "Online Infrared Detection of Inclusions and Lack of Penetration during Welding", Materials Evaluation, pp. 933-937.

[12] Ibarra-Castanedo C. et al, 2004 "Automatic Algorithm for Quantitative Pulsed phase Thermography calculations", CD of Proceedings 16th WCNDT--World Conference on Nondestructive Testing, Montreal (Quebec).

[13] Shepard S. et al, "Quantitative Characterization of Thermographic Sequence Data",

[14] Khanna O.P., 1999, "A Text book on Welding technology", Dhanpat Rai Publications, pp. 628-630

[15] Rafael C.Gonzalez, Richard E.Woods, 2005,"Digital Image Processing", Prentice Hall of India Pvt. Ltd., pp. 612-613.

[16] Stephane G. Mallat, 1989, "Multi frequency channel decomposition of images and wavelet models", IEEE Trans on Acoustics, Speech and Signal Processing, Vol.37, No.12,

[17] Raghuveer M.Rao, Ajit S Bopardikar, 1998, "Wavelet Transforms: Introduction to Theory and Applications", Prentice Hall.

N.M. Nandhitha (1), B. Sheela Rani (1), N. Manoharan (1), B. Venkataraman (2), M. Vasudevan (2), Chandrasekar (2), P. Kalyana Sundaram (2) and Baldev Raj (2)

(1) Sathyabama University, Jeppiaar Nagar, Old Mamallapuram Road, Chennai 600 119, Tamil Nadu, India Email:

(2) Indira Gandhi Centre for Atomic Research (IGCAR), Kalpakkam 603 102, Tamil Nadu, India
Table 1: Quantitative Characterization of porosity.

Feature vector/frame         183           195           203
Major axis length (pixels)   8.9938        8.1074        6.6131
Minor axis length (pixels)   6.5550        6.0491        4.2583
Area (pixels)                44            36            20
Mean                         5.7292e-004   4.6875e-004   2.6042e-004
Variance                     1.6573e-005   1.1808e-005   3.7614e-006
COPYRIGHT 2009 Research India Publications
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2009 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Nandhitha, N.M.; Rani, B. Sheela; Manoharan, N.; Venkataraman, B.; Vasudevan, M.; Chandrasekar; Sund
Publication:International Journal of Applied Engineering Research
Article Type:Report
Geographic Code:1USA
Date:Apr 1, 2009
Previous Article:Fluidization and surface-to-bed heat transfer in a aound assisted bubbling fluidized bed of fine powders.
Next Article:Design a system for monitoring human health VIA power line based on basic embedded system.

Terms of use | Privacy policy | Copyright © 2021 Farlex, Inc. | Feedback | For webmasters |