Development of a Methodology for Monitoring the Deposition Process in Gas Metal Arc Welding (GMAW)

Abstract

Gas metal arc welding (GMAW) processes need to guarantee the quality of the parts produced from this stability in the manufacture of a single bead. In addition to the visual inspection and subsequent characterization with the cutting of the samples, which consists of destructive analysis, it is possible to monitor the entire process and the quality of the part in a non-destructive way. Therefore, this work aims to develop a methodology for the analysis of non-destructive and online weld beads using high-speed cameras, thermal cameras, profilometer and algorithms in MatLab for data processing. The high-speed cameras allow the capture of images of the electric arc and the metallic transfer mode. The thermal cameras, on the other hand, allow the visualization of the melt pool and the temperatures reached during the deposition. Finally, the laser profilometer allows you to make a point cloud of the part and measure its geometry (height, width, height-to-width ratio, wetting angle) online and more accurately than the caliper. With this, it is possible to use the data, both for geometric quality, preliminary parameterization of additive manufacturing, and later for feeding simulations of welding. Promising the monitoring process during manufacturing.

1. Introduction

Welding processes with welds presenting defects in their geometry may also present changes in their metallurgical and mechanical characteristics, such as, for example, residual stresses and deformations and heat-affected zones with coarse grains and subsequent cracking. Understanding the variables involved (electrical, geometric, metallurgical) is of great importance to guarantee the final quality of the material.

With regard to the transfer mode, in GMAW, short circuit, globular, spray and pulsed modes can occur [1]. One way to study the metallic transfer mode of a material is through the construction of metallic transfer maps, which are schematic graphs that show regions or areas in which a certain type of transfer predominates. The maps serve as a support tool for analysis of the welding process and guidance for carrying out the operation and aim to establish transfer contours in terms of welding variables. Typically the ways constructed are in terms of voltage or arc length versus current or feed speed [2].

The identification of the transfer modes can be done at the instant the welding process takes place or after performing the weld, through the detection and analysis of different signals, generally provided by the arc region. Sound, light, current and voltage electrical signals, the appearance of the arc, spatter, among others, can enable the detection of the transfer mode. Sophisticated techniques, for example, based on analysis of images obtained through high-speed filming, allow obtaining a greater volume of information and determining, with greater security, the metal transfer mode, but they tend to be expensive and time-consuming. One of the most used techniques for the detection and analysis of some transfer modes is the analysis of current and voltage oscillograms, based on the arc current and voltage signals [3]. When it is not intended to see the arc, but the metal deposition, the shadowgraph technique can be used. This is a lighting technique used in conjunction with high-speed cameras in filming metal transfer in GMAW welding processes. The shadowgraph technique makes use of backlit directional illumination with a laser beam as the light source [4].

After shooting, pre-processing is important to improve its characteristics, which depend on the final objective intended for the image. Some processing typically involves techniques for contrast enhancement, noise removal, and region isolation [5]. Other processings used in digital images are erosion, dilation, thresholding, black and white, Gaussian and Fourier filters. It can also perform operations such as smoothing. Equalization and the convolution of matrices or images, for example, the multiplication of the original image with the images that have applied one or more filters. The dilation process consists of obtaining the reflection about its origin and then shifting this reflection from x. The dilation of A by B is then the set of all x displacements. Erosion is the remainder of displacements; dilation and erosion are dual operations with each other and with inverse behavior [6].

Burnoff’s criterion [1] or droplet detachment allows the appropriate wire speed and average current to be found for the material. The methodology seeks a balance between the melting rate of the wire and the wire feed speed. This criterion consists of building a functional relationship to represent all possible conditions of the pulse parameters (peak current, base current, peak time and base time) associated with an average current (Im). This relationship includes a workspace that encompasses all possible combinations of parameters and represents a region called the parametric zone.

For these reasons, this work intends to contribute to the monitoring of the deposition process. The first strategy consists of filming with high-speed cameras using the shadowgraph technique with MatLab programming for image processing in order to monitor the metallic transfer mode, number of drops per pulse, size and droplet detachment. The second monitoring stage consists of using a system with an infrared camera to observe the arc length and geometry of the weld pool and the bead. Finally, the third step consists of validating the measurements using the laser profilometer.

2. Materials and methods

The welding process was carried out on a Fronius Trans Plus Synergic 5000 power source. The power source operated in constant voltage mode and the power source was manipulated using an industrial computer via an interface for ROB 5000 robots. PS 130/6 – 5-P from ABB Robotics and programming for one or more beads with the ROB 5000 Interface, RCU 5000 controller and programming done with the Sport S3 Software. The base metal used was low carbon steel AISI 1020 with the aim of developing the methodology with a better cost-effectiveness ratio, with ESAB OK TUBROD 410 NiMo tubular wire and AWS INOX ER 316 wire, and shielding gas with 94% argon and 6% CO₂ and also pure argon. Tests were carried out combining two types of wire and two types of gases, but the methodology carried out can be adapted for any material and shielding gas that is in demand. The initial studies were carried out in pulsed GMAW mode due to the greater stability of the transfer mode. Decreasing the range of working values for parameters such as average current, peak current, base current, cycle or pulse time, wire feed speed, welding speed and inductance, to obtain a weld with good characteristics. The varied parameters are shown in Table 1.

Variable parameters	Fixed parameters
Travel speed: 6–12 mm/s	Contact tip distance piece: 15 mm
Wire feed speed: 2–8 m/min
Pulse frequency: 55–300 Hz	Torch angle: 0°
Peak current: 300–450 A
Base current: 15–80 A	Electrode diameter: 1.2 mm
Drop separator current: 50–110 A
Indutance: 0–50%	Stand off: 12 mm
Base time: 40–50% of period
Peak time: 40–50% of period	Gas flow: 15 L/min
Drop separator pulse time:0–0.8 ms

Table 1.

Parameters used in the depositions.

Other aspects of the research that should be highlighted are the metallic transfer technique used and the amount of energy and heat input for the metallic transfer. It is also relevant to point out that the result of these data or parameter values was evaluated in different parts of the process using the shadowgraph technique for the analysis of drops, type of transfer and number of drops per pulse, using, for this purpose, a high-resolution camera. CMOS speed captures 1000 frames per second, enough to capture between 5 and 7 images per pulse, depending on the frequency of the pulse and the frequency of the drop. The camera was placed 5° in relation to the bead and its displacement was always carried out in front of the wire in the same position to visualize it since it is necessary to see the drop, its characteristics, its detachment and the transfer mode. The oscillograms were evaluated using programming made in MatLab, an action that helped in the analysis of the stability of the weld, caloric input, energy per pulse during the process and average current. Then, the Burnoff analysis was performed and the empirical analysis based on data collection was used.

The shadowgraph assembly has as its principle the passage of a collimated laser beam through the arc region so that the resulting image represents the shadow of these elements (wire, metal in transfer, weld bead). An optical bandpass interference filter is placed between the arc and the camera, so that only the laser beam and the respective shadows will appear in the image, suppressing the light produced by the arc that is not in the region of the filter. This arrangement uses a Galileo-type laser beam expander and the expander uses a diverging lens as the beam input and a converging lens as the output, which ideally produces a flat wavefront at the output of the expander, thus having no distortion or magnification in the geometries. of the elements. The scheme of the technique is illustrated in Figure 1.

Figure 1.

Scheme used in the shadowgraphy technique [4].

For the storage of data and/or images, such as, for example, current and voltage data, in addition to the images obtained through the profiling technique, an industrial computer from the ADVANTECH brand, reference ICP-622, was used, which contains a PCI Eagle 703S data acquisition board in charge of acquiring the electrical signals of the process, in addition to being responsible for carrying out the communication between the computer and the ROB 5000 interface, which divides its tasks in the communication providing a time of reading and data response with the power source. The acquisition of signals is done through a program developed in the LabVIEW virtual instrumentation software, in the case of current and voltage measurements. The manipulation of the source for shadowgraph has an additional data acquisition board that records 10,000 data per second, in the case of shadowgraph, the program made through the board is also in charge of manipulating the linear displacement table.

The shadowgraph technique was complemented by the synchronization of the images obtained by filming with the current and voltage oscillograms. The method couples the variations of the welding parameters with the images and makes it possible to correlate the information from the two sources. The shadowgraph, together with the analysis of the oscillograms, allowed the monitoring of the transfer mode in order to elaborate the metallic transfer maps. After capturing the images, pre-processing was carried out to improve their characteristics with erosion, dilation, thresholding, black and white filters, Gaussian and fast Fourier transform techniques.

The second part of the monitoring is based on the development of infrared image capture equipment for detecting output variables such as arc length, width and reinforcement, along with image processing and adjustments in data synchronization, size and organization. To carry out measurements from the ultraviolet to the near-infrared spectrum, the SM-240 spectrometer was used. This equipment has a CCD sensor internally composed of four parts: (1) input mechanism with built-in slot, a fiber coupling adapter and a spectrum classification filter; (2) Czerny-Turner cross-array spectrograph, using high-quality focus; (3) linear charge-coupled device (CCD) sensor array and driving circuits; (4) Computer interface card. For the measurements, the spectrometer was placed 25 cm from the arc using a 0.2 light attenuator and a tube to direct the light from the weld and reduce the effects of external light. For the adjustment of the camera, a high-power infrared LED with the following characteristics was used:

• LED code: K2005;

• Emission Color: Infrared (IR) 4 chips;

• Power [W]: 5 W;

• Wavelength (nm): 940 nm;

• Luminous Flux or Luminous Intensity: 250 mW;

• Current:700 mA.

The equipment developed to capture the images comprises a 2.0-megapixel Web camera at 30 fps to 50 fps - adapted for IR; lenses, 1000 nm high-pass infrared filter; two 1.0 radiation attenuators; tele objective -zoom focal length (18–108)/Aperture 2.5; and (5); polarizer. To adapt the infrared camera to the camera used for shadowgraph, the location of the infrared camera was perpendicular to the bead and its displacement at 90° and always in front of the wire in the same position to avoid changes in the images in reinforcement and width. The experimental bench with the spectrometer and the two cameras is shown in Figure 2.

Figure 2.

Experimental bench for making the shadowgraph.

The laser profilometer was used to validate the geometric measurements performed by the infrared image monitoring system. Calibrations were performed as described in the studies by Idrobo-Pizo et al. [7].

3. Results and discussion

3.1 High-speed camera

As already explained at the beginning of this research, a welding process, to be used successfully, must have the following characteristics: deposit beads free of welding defects; be able to apply the process with a high deposition rate and in all welding positions; absence of spatter and splashes in the formation of deposits; easy opening of the electric arc; produce weld layers with excellent surface finish, among other features.

For the validation of the data collection methods of the developed sensors, it was necessary to legitimize the processes and steps through the shadowgraph technique. For this, it was necessary to carry out the tests synchronized with the capture of images by shadowgraph. Initially, the first method of capturing images by profiling and image pre-processing was used. Tests were performed to capture the droplet mode and characteristics in droplet detachment and arc length performing image processing. Figure 3 shows one of the tests with a voltage of 26.5 V, wire feed speed of 6.0 m/min and using Argon with 4% CO₂.

Figure 3.

Shadowgraph image capture for droplet mode.

The original image has excessive noise making it necessary to treat it. Thus, the second image from left to right received treatment with a Gaussian filter, erosion, dilation and binarization. In the third image, a treatment with erosion, dilation, mean filter and black and white filter was performed. Finally, the fourth image is a superimposition of the treatments carried out in the second and third. Additionally, an algorithm was developed to calculate the relationship between wire diameter, droplet size and arc length. Another analysis is shown below in Figure 4.

Figure 4.

Edge surface imaging with object recognition and pixel calculation.

In the first image, a treatment was carried out to visualize details of the wire and the surface of the pool with a blue background. In the second and third images on the right, different segmentation filters are applied with edge recognition such as Prewitt and Sobel. Finally, the fourth image is created based on the previous two edge recognition images. In this, an algorithm is applied that segments it and recognizes objects by dividing them and placing a different color for each object. The algorithm calculates the number of pixels per area within each object or color and subsequently determines the approximate volume of the drop through a relationship with the calculated area, this area is calculated through a relationship with the pixels within the drop area and the pixels of the wire diameter which is a known value in millimeters. By calculating the maxima and minima of the pixels at the edges, measurements of the arc length can be obtained. By synchronizing these measurements with measurements from other sensors and other variables, it allows for validating measurements from other devices or equipment.

3.1.1 Burnoff criterion

Burnoff’s criterion allowed the construction of parametric curves and the study of the regions where the transfer was more stable. Through tests carried out with pulsed current and with constant current, a graph of average current versus wire speed was constructed (Figure 5). The intersection point between the two generated lines allowed inferring the range of the transition current between the globular and spray/aerosol transfer modes at 167.8 A with a wire feed speed of 5.04 m/min. Such values can also be decreased or increased in the GMAW-P, using an Im lower than the calculated one, without loss of weld quality. However, if the value of Im is reduced beyond the acceptable level, instability in the process and the formation of weld spatter occur. On the other hand, if the value of Im is increased, the heat generated, the pool penetration and the material flow increase, in addition to converting the transfer mode from spray to projected spray, bringing instability in the process. In this sense, this study analyzed groups located below and above the intersection point in order to confirm the elucidated aspects. Using wire feed speeds of 5, 6 and 7 m/min with average currents of 175, 188 and 200A. The intersection between the two lines, based on Burnoff’s criteria, determines the average current value to obtain a stable spray/aerosol transfer with a droplet diameter size close to the wire diameter and with minimum feed speed for this current. The drop volume is considered equal to that produced by a sphere, as shown in Eq. (1), where r = wire radius (mm) = 0.6 and VD = drop volume (mm³) = 0.90478.

Figure 5.

Relationship between Ia and Va for welding with constant current and pulsed current.

It is also known that the amount of material deposited on the base metal is the product of the droplet volume and the pulse frequency (F), or the product of the cross-sectional area of the electrode (s) and the velocity (v) Eqs. (2) and (3). With a wire feed speed of 6 m/min, the minimum frequency is F = 125 Hz and the drop formation time, called cycle time, is Tc = 8 ms. Thus, if the drops are smaller and the heat supplied is greater, the frequency must be higher. In this way, drops with smaller dimensions and higher energy have more stability in the metal transfer process. Therefore, the optimal values of time and frequency used in this study had as a criterion the smallest droplet size and the relationships between Im and Va established from the Burnoff criteria.

$$\mathrm{VD}=\left(4/3\right).\mathrm{\pi }.\text{r3}$$E1

$$\mathrm{VD}.\mathrm{F}=\mathrm{s}.\mathrm{v}$$E2

$$\mathrm{F}=3\mathrm{v}/\left(\text{2d}\right)$$E3

In this representation, the point of intersection between the two lines was used to calculate the cycle time (Tc = 8 ms) for the studied feed rates, but the values used for this research were Tc = 5 ms and Tc = 6 ms with frequencies of Fr = 166.7 Hz and Fr = 200 Hz, respectively and Im = 180 A in the construction of the Parametric Zones, with the aim of saving energy and increasing stability in the transfer process. Figure 6 shows the typical time graph of Ip versus Ib for establishing the parametric zone for a feed rate of 6 m/min. Such a representation exhibits a linear relationship between Ip and Ib, the minimum possible value being the common point Im = Ip = Ib. This zone presents the multiple combinations between Ip, Ib and Tp for a given Im, which guarantees a pulse condition that fulfills the Burnoff criterion.

Figure 6.

Parametric zones according to the Burnoff criteria for V = 6 m/min, Im = 178.404A and (a) Tc = 5 ms and (b) Tc = 6 ms.

From shadowgraph studies and Burnoff’s criteria, correlating with the current and voltage data from the oscillograms, it was possible to draw the current and voltage map showing the regions in which the transfer modes occur, as shown in Figure 7. With this, an optimal region of work is chosen depending on the transfer mode and the geometric quality of the bead. As will be further explored below.

Figure 7.

Voltage scan image to determine transfer modes and process behavior using wire 410 NiMo and 96%Ar + 4%CO₂ gas.

3.2 Infrared camera

The results obtained by the spectrometer are shown in Table 2. The spectra were captured using a sensor with manual adjustment and calibration of the equipment to obtain the wavelength in meters. For this, LEDs of different wavelengths with known values were used, in addition to considering that the spectrometer has a capture limit range between 100 nm and 900 nm. In addition, the results were crossed with studies by Mota [8] who investigated the spectra between 890 nm and 930 nm, obtaining results similar to those of the present study.

Wavelength (nm)	LED color	Intensity (lm/cm^2) – approx.
320-360-390	Ultraviolet	280–580
460	Blue	150
560–575	Green	200
595	Yellow	200
645	Red	170
815	Infrared	<100
850	High power infrared	<100

Table 2.

Spectrometer results.

With the data captured by the spectrometer and its subsequent analysis, results were obtained that allowed finding the tracks where it is possible to visualize the bead, opening the possibility of continuing with the project of the CCD sensor of the commercial camera. This sensor can capture a cloud of points of the infrared spectrum forming an image of the bead, in addition, the sensor manages to partially eliminate the spectra generated by the plasma or arc due to its maximum capture limit and the use of filters that restrict the values of capture in the proper range. The images obtained together with image processing and computational algorithms allowed obtaining the necessary information for the research, these being the measurements of width, reinforcement and length of the arc in real time with the 1000 nm filter as shown in Figure 8.

Figure 8.

Measurements tests. (a) Reinforcement and width measurement test in the same camera, using 1000 nm filter, attenuators and polarizer. (b) Test to measure reinforcement only, using 1000 nm filter, attenuators and polarizer.

One of the objectives of the development of this camera is to obtain the width and the reinforcement using the same camera or the same sensor at the same time. Therefore, if it is necessary to obtain two images in the same camera at the same time, one of the images travels a greater distance as it is an indirect image that passes through two mirrors, which implies obtaining a small blurring of one of the images, which may be the image that arrives directly or the one that reaches the camera indirectly. In the image of Figure 8, a better focus can be observed when a single measurement is performed, this is because due to the differences in ways to measure both the width and the length with the same camera, it is necessary to focus the camera on a midpoint between the two paths, which generates a small distortion in the images, a fact that does not occur when only one parameter is being evaluated. To solve this problem, two cameras could be used, but this would increase the cost of the proposed methodology. Another solution comes from the fact that the images can also be improved, using a camera with higher resolution and changing the aperture of the diaphragm to reduce blurring, for this reason, a zoom with variations of the diaphragm and focus was used. Thus, for these tests, a low-resolution 2.0-megapixel web camera was used, which is economical and can capture 33fps or can be adapted to capture 50fps.

One of the tests adopted to carry out the different types of adjustments in the camera, such as, for example, the focus of the two images obtained in the same camera (direct and indirect), the position of the mirrors to capture the indirect image at the bottom of the camera and mainly the test of lighting levels to capture the infrared spectra of wavelengths greater than 950 nm, was the test shown in Figure 9, in which three infrared LEDs with different wavelengths (730 nm, 850 nm and 940 nm) were used, as well as high power LEDs each of 5 W. In these tests, it was only possible to capture images of the LED with a wavelength of 940 nm because its transmittance level is 950 ± 25 nm shown in Figure 8, which coincides with the limits of the infrared spectra of the filter used in the camera. The image in Figure 9 shows the test with the 940 nm LED, in the upper part of the image there is a direct image of the profile of the high-power LED, and in the lower part, its front face is seen, captured from a higher position, but with the use of two mirrors taking the image indirectly to the same camera at the bottom of the sensor.

Figure 9.

Infrared camera adjusts with high-power LED 950 nm.

The reason why it is possible to see the 940 nm infrared LED image using the 1000 nm high pass filter is that it can have an emission range with high transmittance up to 975 nm and very low transmittance levels up to 988 nm. Furthermore, the 1000 nm high-pass filter has a tolerance for capturing wavelengths lower than the limit value at −25 nm, so this filter begins to weakly capture the spectra that emit from 975 nm onwards. For this reason, even when testing with a high-power LED, the light intensity captured is weak. As for the other tested LEDs, with shorter wavelengths, it is not possible to capture images.

Figure 10 shows how the algorithm performs the measurements of reinforcement, width and length of the arc with images captured through the infrared camera, the equipment developed and previous treatment of images using media filter, black and white filter and Prewitt edge filter and Sobel.

Figure 10.

Measurements with MatLab algorithm of (a) height, (b) width and (c) arch lengh.

3.3 Laser profilometer

To validate the measurements made online with the infrared camera, an offline measurement of the reinforcement and width was carried out, with a scanner for each of the cords obtained, which are presented by a cloud of points in a 3D graph, exemplified in Figure 11. The scanner program generates a table with the maximum value data on the Y-axis and X-axis, collected from the point cloud obtained by the laser transversely around the bead. Thus, with this maximum value, determine the reinforcement and width at each measurement instant.

Figure 11.

Point cloud of a bead using a laser profilometer.

4. Conclusions

It is concluded that high-speed cameras are an effective way to control the transfer mode and parameterize the welding processes together with current and voltage sensors and algorithms for image post-processing, essential for better visualization of the transfer, increasing image quality and reducing noise. All of these techniques, together with the Burnoff criteria, guarantee the choice of the optimal working region and the stability of the welding process and the final quality of the bead. In addition, the use of results obtained via spectrometer and infrared camera allows online obtaining of width, reinforcement and arc length, allowing monitoring of the entire process and bead geometry. Profilometer validation is extremely useful to prove the quality of the monitoring system developed via cameras, making it a viable technique applicable to parameterization processes. Post-processing is complementary to increasing image quality, but online measurements can be performed without using it.

Acknowledgments

To CAPES for financial support and to the University of Brasilia for the necessary infrastructure to carry out the tests.

Conflict of interest

The authors declare no conflict of interest.