Laser diode based vision system for viewing arc welding (May 2009)

Melton G, Schuhler C TWI Ltd

Houghton M University of Liverpool

Paper presented at Eurojoin 7, Venice Lido, Italy, 21-22 May 2009.

Summary

Vision-based systems are employed industrially for remote monitoring of the arc welding processes in mechanised applications. The main difficulty encountered by operators is the high intensity of light emitted by the arc, which prevents good visibility of the weld pool. The image quality that the operator obtains from vision-based monitoring systems is often not sufficient to allow judgement regarding the need for in-process parameter adjustments.

By filtering out most of the arc radiation with a band pass filter and illuminating the weld pool and surrounding area with light from laser diodes at the same wavelength, a good image of the weld pool is obtained.

This paper describes the application a vision system based on a CMOS camera and laser diode illumination at 905nm. This system is capable of viewing the weld pool, joint seam, wire position and electrode condition in TIG welding. In MAG welding the droplet transfer and weld bead position in a joint can be seen clearly.

This vision system can be used for real time momitoring and adaptive control of welding processes.

Introduction

There is an increasing need for automated welding systems to produce consistent levels of quality, to overcome a shortage of skilled welders and due to tighter health and safety regulations, requiring the operator to be removed from hazardous environments. These automated systems have to be able to tolerate part variations similar to manual welding using operators.

In mechanised applications, welding tends to be performed using preset welding parameters and so highly consistent and repeatable parts and alignment are required for high-quality results. Manufacturing tolerances of upstream operations, such as profile cutting and edge preparation, variation in the chemistry and accuracy of assembly, can all significantly affect the quality of the welding operation, yet can be challenging and expensive to control.

The weld pool characteristics provide important information about the weld quality. For example, there is a strong correlation between weld pool width and the degree of penetration in tungsten inert gas (TIG) welding. The weld pool width, volume and position relative to the joint are also important in determining the surface profile of the weld and the likelihood of imperfections such as lack of fusion and penetration. High quality images of the weld pool are essential in order to be able to detect weld pool deviations and to implement the necessary changes in the welding parameters or to track the joint in real-time, in order to maintain the weld quality.

Vision-based systems are employed industrially for remote monitoring of the arc welding processes in mechanised applications. The main difficulty encountered by operators is the high intensity of light emitted by the arc, which prevents good visibility of the weld pool. The image quality that the operator obtains from vision-based monitoring systems is often not sufficient to allow judgement regarding the need for in-process parameter adjustments. As a consequence, vision-based systems are generally limited in their industrial use to a visual aid for joint tracking and qualitative monitoring of the process. With improvements to vision systems, they could be used for quality monitoring and in-process control.

Fig. 1. TIG welding arc spectrum, 150A, argon shielding gas

The main difficulty in vision-based systems is to eliminate the high intensity arc light, which prevents good visibility of the weld pool. Figure 1 shows the broad spectral response of arc light emitted by TIG welding on stainless steel. The spectrum ranges from the ultraviolet at about 350nm through the visible region and into the infrared at about 850nm. In this vision system, a narrow band-pass filter has been placed in front of the camera to eliminate the arc light.^[1] This filter only allows light of limited wavelengths to pass through. In this case, the filter is chosen for a wavelength where the arc light intensity is low. However, the band-pass filter darkens the image, making it difficult to see the weld. By adding an illumination source, an overall view of the welding area can be obtained.^[2,3,4] The wavelength of the emitted light must be in the same range as the band-pass filter. This can be achieved using a laser light source. Figure 2 shows the effect of the band pass filter with the laser light source. The result of this combination is a much attenuated arc light with a dominant laser light which illuminates the weld pool and its surroundings.

Fig. 2. Principle of spectral filtering:

a) Unfiltered light;

b) Light captured with the bandpass filter.

The amount of light captured by the camera also depends on the shutter time of the camera. The longer the exposure time, the more arc light will be captured. This does not affect how much laser light is captured provided the pulse width of the laser pulse remains less than the exposure time. Laser light will only be attenuated if the laser pulse width is greater than the exposure time of the camera or if the laser is a continuous wave rather than pulsed. The contributions to the image brightness from the laser and the arc light are illustrated in Figure 3. This technique is known as temporal filtering. In order to optimise the image, the camera shutter opening time must be synchronised with the laser diode illumination time.

Fig. 3. Principle of temporal filtering:

a) Normal exposure time (pulsed laser shown for clarity);

b) Reduced exposure time, synchronised with laser pulse

Vision system

Previous work has used a laser to provide the illumination, but lasers are expensive and not ideally suited for production applications. TWI Ltd. has collaborated with the University of Liverpool, on the development of a vision system for monitoring arc welding processes based on low-cost camera and laser diode illumination system. ^[5,6,7]

The illumination source is 16 laser diodes (OSRAM SPL PL90_3) operating at a wavelength of 905nm. Each laser diode has a beam divergence of 11° x 25° giving an elliptical illumination pattern. The pulsed laser diodes are designed to operate at 75W for 100ns for applications such as range finding. For welding, it is necessary to maximise the energy in the pulse while maintaining a small laser on to off ratio or duty cycle for arc light reduction. Video is typically 25 (fps) giving a duty cycle of just 0.025 to 0.030% for a 10µs shutter. A laser pulse width of 10µs was chosen because this is the minimum exposure time of the camera. A cluster of 16 diodes can produce laser pulse energy of 2.85mJ (17.8W of total power) at desired frame rates.

The diode cluster is mounted on a printed circuit board (PCB) to fit within a 40mm diameter tube. At the end of the tube there is a single lens of focal length 80mm (or alternatively 125mm). The distance between the lens and diode cluster is adjustable to allow a clearly focused illumination area to be obtained on the work piece. To protect the laser diode array and the cameral from heat and spatter from the process, a clear polycarbonate sheet was placed in front of both and cooling air was directed across the front.

The camera is a 1024 x 1024 pixels complimentary metal oxide semiconductor (CMOS) Photon Focus Hurricane _40_CL (MV-D1024E-40) camera fitted with a 50mm lens with adjustable aperture and manual focus and a 950nm band pass filter. The camera is positioned to view the weld and surrounding area and the laser diode cluster is positioned on the opposite side, Figure 4.

Fig. 4. Vision system showing the laser diode cluster and CMOS camera

A PC is used to control the system. A frame grabber card mounted in one of the PCI slots captures images from the camera. This card is controlled by software which allows the full specification of the camera system to be set up and includes frame rate and laser pulse width.

The system captures a series of images at a frame rate controlled by a timing circuit mounted in a control box. Once captured, each image is sent to the computer in real time. The images are analysed by software (Active Silicon Phonic) which also controls the camera. The number of pixels is set to the maximum which is 1024 x 1024 pixels. The frame rate is set by the control box to 25 frames per second (fps). The exposure time is fixed at 10µs and the lens aperture adjusted to control image brightness.

Applications

The vision is capable of producing good images of TIG and MAG welding, successfully eliminating most of the arc light from the image. Optimum results were achieved with the light source about 100-250mm from the torch at an angle of between 10 and 30° to the horizontal. The camera was positioned at about 250-500mm from the torch at an angle of 20 to 50° from the horizontal. Should it be necessary to mount the light source further away from the torch, higher power diodes would be required. Different camera positions could be achieved with alternative lenses. 

Fig. 5. TIG welding

a) autogenous mild steel,

b) stainless steel with filler wire

c) AC aluminium

The images from both DC TIG (Figure 5 a) and b)) and AC TIG (Figure 5 c)) welding, with and without a filler wire were very clear. The improved clarity of the weld pool images as a result of eliminating the arc is expected to be beneficial for real-time image processing applications in TIG welding. The weld pool edges are clearly defined, and it was also possible to see the joint line for tracking purposes. A good image of the weld pool width could be used in a feed back control system to automatically adjust the welding parameters to maintain weld pool geometry, for example when a component heats up or heat transfer is variable. With a filler wire the image could allow the correct adjustment of the wire feed rate and position of the wire in the weld pool. Also, in Figure 5c) the balled end of the electrode can be see clearly, demonstrating that tungsten electrode condition can also be monitored.

MAG welding arcs in general operate at higher current levels and the light intensity is greater. Consequently, achieving the desired level of illumination in MAG welding is more difficult than for TIG welding. With the current system not all the arc light could be eliminated for MAG welding. However, It was found that illumination and camera positions could be achieved which allowed viewing of both the weld pool and metal transfer modes, but for the latter focusing needed to be very precise.

With the camera mounted to observe the weld pool it is possible to see the tip of the wire and the droplet transfer mode. In Figure 6a) for spray transfer and Figure 6b) for pulse transfer the droplet size can be clearly seen. Also, the software enables the images to be viewed as a video in real time, or slowed down so that individual droplets can be seen.

Multiple lighting directions with higher laser power would be preferred to give a higher intensity of illumination and a more natural lighting. This would be especially convenient to observe metal transfer where a shallow camera angle is needed and the laser light reflection on the work piece is quite difficult to obtain. Moreover, this solution would reduce the laser light reflection in the weld pool which for some camera positions prevents the pool movements from being seen. 

Fig. 6. MAG welding

a) spray transfer

b) pulse transfer

c) welding in a groove

The shielding gas composition was observed to have an influence in MAG welding. The greater the carbon dioxide content of the shielding gas, the greater the emissions at 905nm and hence the lower reduction in arc light achieved by the filtering. Most of the arc light could be eliminated by closing the camera aperture, but the image becomes less bright, and the surrounding area cannot be seen.

The vision system can also be used to view bead placement in a joint. Figure 6c) shows a MAG weld in a groove. The position and volume of the deposited weld can be seen clearly.

In multi-pass welds the image quality was found to degrade after several passes. In joint preparation the sidewall initially acts as a reflector providing general illumination of the area, but, in time, the sidewalls become coated in welding fume and then the level of illumination is reduced.

For practical application the safety of the laser diodes needs to be considered. This illumination source is a Class 3 device, which means that the beam should not be viewed directly. However, scattered light from a reflective surface should not be a hazard. In a production environment, similar precautions as those required for laser seam tracking systems are required.

Conclusions

1. The laser diode based vision system is capable of producing high quality, real-time images of the TIG and MAG processes.
2. For TIG welding the arc light is eliminated. The extent and position of the weld pool, solidification lines, surface slag, and the onset of full penetration and filler wire melting are clearly visible during TIG welding. Thecondition of the tungsten electrode can also be seen.
3. For general viewing of MAG welding some arc light can be seen, but can be eliminated sufficiently for metal transfer viewing. Droplet size and metal transfer mode can be observed and the weld pool in a joint can be seenclearly.
4. For optimum viewing the laser diodes and camera need to be positioned about 100 - 250mm and 200 - 500mm respectively from the arc. For greater distances a higher laser diode power and different camera lenses would berequired.
5. The illumination source is a Class 3 laser so in a production environment similar precautions to those required for laser seam trackers are required.
6. Laser diode based vision systems offer a practical cost effective solution to real time monitoring of welding operations in a production environment.

References

1. Wezenbeek H C F M, 1992: 'A system for measurement and control of weld pool geometry in automatic arc welding'. (Thesis) Technische Univeersiteit Eindhoven, Netherlands.

    

2. Burke M A, Hackworth D T, Miller R A and Kasner W H, 1985: 'Weld pool viewing system'. United State Patent 5,475,198.

    

3. Jinqiang G, Chuansong W, Min Z and Yanhua, 2007: 'Vision-based detection of MAG weld pool'. China Welding. Vol. 16, No. 1, pp32-35.

    

4. Mulligan S J, 2007: 'Evaluation of a low-cost camera for monitoring arc welding processes'. TWI Members Report 873.

    

5. Lucas B and Abdullah B, 2007: 'Looking into the pool - a low-cost real-time vision system for arc welding'. TWI Bulletin.

    

6. Lucas W, 2007: 'Novel, low-cost and unique... ...a revolutionary arc viewing system takes centre stage'. TWI Connect, no.147. Mar.-Apr.2007. pp.1-2.

    

7. Houghton M A, Lucas J J and Lucas W, 2007 'Vision systems for monitoring and control of arc welding operations.' Soldagem & Inspecao, vol.12, no.4. pp.293-299.

Geoff Melton is a Consultant at TWI Ltd. He has a degree in Physics from the University of St Andrews and an MBA from Loughborough University in the UK. He has been involved in welding research for over twenty years with a particular interested in arc welding processes, including equipment, consumables, process developments, monitoring and control. He is also the Chairman of TC26A the European standards committee for arc welding equipment.