Looking into the pool...A low cost real time vision system for arc viewing.
TWI Bulletin, July/August 2007
Professor Bill Lucas is Technology Manager in the Arc, Laser and Sheet Processes Department at TWI. He received his doctorate from Queen's University, Belfast. He was employed by Leyland Motors for four years and joined TWI in 1970. After 16 years in process research, he became Head of the Arc Welding Department in 1986. In 1987, he was the first research engineer at TWI to be awarded Doctor of Science for his contribution to arc welding and computer technology. He has been a Visiting Professor at the University of Liverpool since 1997.
Badr Abdullah is a final year PhD student at the University of Liverpool working on vision systems for monitoring of arc welding processes. Badr received the MEng degree in Electrical Engineering and Electronics from the University of Liverpool, UK, in 2004, having specialised in solid state electronics. During his master degree, Badr won the Farnell InOne prize for best master project.
For traditional manual welding, the positioning of the welding torch and the selection of the welding parameters are controlled by the operator, so the quality of the resultant weld is determined by the practical skill of the welder and their knowledge of the welding process. For automation and control, a vision system is required to detect the various deviations and implement the necessary changes. In this feature by Bill Lucas and Badr Abdullar a novel, inexpensive vision system is based on the use of a low-cost CMOS camera and a laser source to illuminate the weld area in the presence of an intense arc.
The design features and the specification of the different components that make up the system are detailed. A combination of optical filtering, short pulse laser illumination and fast-shuttering imaging techniques along with precise synchronisation and very accurate triggering methods has enabled high quality images of the weld area irrespective of the intense arc brightness and the use of a low-cost CMOS camera.
The overall objective of the system is to provide reliable real-time measurements of the molten weld pool for use with a process controller. The weld pool contains important information about the welding process, which can be used by the process controller to adjust the welding parameters and regulate the weld pool width in order to allow for consistent welds. Several pulsed and continuous wave lasers have been tested with this system as a potential source of illumination. It was demonstrated that the arc light could be almost totally eliminated, enabling the weld pool and surrounding area to be clearly imaged.
Control of weld penetration is at all times a puzzling problem in automated and mechanised welding. The weld quality can also be determined by backside weld width and penetration. Although the backside bead width could be directly detected from the rear of the work-piece, sensing from the backside is inaccessible in most cases. Therefore, control of weld quality by sensing the topside has become a challenging problem in automated welding. Extensive experimentation has shown that the geometry, ie the size and shape of the weld pool, provides sufficient information on the weld penetration. It is also known that the weld penetration is a major determinant of the weld quality, so the geometrical features of the weld pool should be sensed and controlled.
The visual information that is accessible in real-time will also enable a quick reaction to various types of malfunction. This information can be used to control the system and modify the welding parameters to maintain the required weld quality and productivity. Figure 1 demonstrates the importance of having a vision system in order to obtain high quality images. Elimination of the arc light of the image makes it possible to extract accurate information about the weld pool and its geometry, weld bead geometry, the electrode tip profile and the surrounding area.
Fig.1. Characteristic appearance of the TIG process in argon shielding gas, arc light is totally eliminated
One of the major advantages of using a visual sensing technology to monitor welding operations is the fact that the visual sensor is not touched or interfered with during the welding process. The visual images of the weld pool therefore provide more accurate information about the welding dynamics. The main difficulty encountered in vision-based sensing of the weld pool geometry is extensive interference from the arc light across a wide spectrum.
Many attempts have been made in the past to find a method of sensing weld penetration so that it can be adequately controlled:
- Infrared thermography from the backside of the weld sheet to make isothermal maps.
- Brightness pyrometry coupled with an axial view torch to measure and control the width of the weld bead.
- Artificial intelligence based approach to sensors for process monitoring and control.
Most of these techniques are not suitable for production systems because they are often bulky, very expensive to produce and only suitable for specific applications.
Reduction in the cost of cameras and illumination systems over the last five years has allowed vision systems to be used increasingly as sensors to extract information about the weld pool. Recently, intensive research has been done to develop a system which can be used for direct weld pool viewing. However, the techniques to date are too expensive to apply in industry, so other alternatives were investigated. The total cost of the vision system introduced here is less than £3000, excluding the cost of the illumination equipment.
CCD vs. CMOS
The benefits of a CMOS based sensor system for monitoring the arc welding process include:
- Low light sensitivity; To be able to operate properly, the CMOS sensor requires a fairly bright environment such as a very bright arc light.
- Anti-blooming; the ability to drain localized over exposure without compromising the rest of the image in the sensor. CMOS generally have natural blooming immunity. CCDs on the other hand, require specific engineering to achieve this capability. Many CCDs that have been developed for consumer applications do, but those developed for scientific applications generally do not.
- Windowing; CCD image sensors only allow for a very restricted selection of an arbitrary window. It is only with the advent of CMOS sensors and cameras that freely defined ROIs (Regions Of Interest) are possible, without a loss of pixel transfer rate.
- Fast shuttering speed. This is because all camera functions can be placed on the camera sensor.
- Wide dynamic range.
Optical spectrum of TIG welding
Anritsu optical spectrum analyser with a wavelength range from 350nm to 1750nm was used to obtain the spectra of TIG welding. Fig.2a and Fig.2b show the emissions emitted by the argon and helium shield gas respectively. Fig.2c and 2d show the arc spectra when argon and helium are used respectively.
Fig.2c) TIG welding arc spectrum, 150 A, pure argon shield gas
Fig.2d) TIG welding arc spectrum, 150 A, pure helium shield gas
Basis of system operation
A typical spectral response range for a standard visible-light camera is 400-900 nm, falling sharply outside this range. On the other hand, laser illumination is at a very specific wavelength, with a typical wavelength much smaller than 1 nm. This is illustrated in Fig.3a, which shows how the light from a laser light-source might compare with the arc light. However, if a band pass filter is used, then only the laser light will pass with some arc light. The result is a much-attenuated arclight with little effect on the laser light as illustrated in Fig.3b.
The amount of arc light captured by the camera also depends on the shutter time of the camera. More arc light will be captured if longer exposure time is used. This does not affect how much laser light is captured as long as the pulse width remains shorter 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 a pulsed laser. The contributions to the image brightness from the laser and the arc light are illustrated in Fig.4a and Fig.4b.
Fig.4a) Before temporal filtering
Fig.4b) After temporal filtering
Vision system structure
The laser light-source can either be mounted on the same side as the camera ie backward reflection or opposite the camera ie forward reflection as shown in Fig.5a and Fig.5b. More laser light is reflected to the camera when the laser is on the opposite side. Hence, less laser energy is required. However, uneven reflection from the weld pool occurs. Varying the angle between the laser beam and the work-piece can improve this. On the other hand, mounting the laser on the same side as the camera gives natural lighting with little glare but requires more laser energy, since most of the laser light reflects away from the camera.
Fig.5a) Forward reflection
Fig.5b) Backward reflection
Laser-camera synchronisation
Figures 6a to 6d show four synchronisation methods that can be used to capture the laser pulse. In Fig. 6a the camera is driving the laser; a strobe output from the back of the camera is connected to the laser external input. In Fig.6b the laser is driving the camera; this will also introduce a delay to the camera shutter, which can be overcome by introducing a pulse delay. In the third method, which is shown in Fig.6c both the camera and the laser are driven by an external master-triggering unit. This is the preferred method for synchronising pulsed lasers. Since both the camera and the laser are triggered at exactly the same time nodelay is introduced. The frame rate, laser frequency and timing of the system can be controlled when connected to a pulse generator. The fourth method shown in Fig.6d is employed when using a continuous wave laser as an illumination source. Both the camera and the laser are free running and are independently controlled. Figure 7 shows that the illumination pulse is synchronised with the camera shutter.
Fig.6a) Camera driving laser
Fig.6b) Laser driving camera
Fig.6c) Camera and laser driven by an external trigger
Fig.6d) Camera and laser driven independently
Fig.7. Camera-laser synchronisation diagram
Illumination systems
Three different illumination sources were investigated to illuminate the weld pool; laser systems, LED diodes and laser diodes. Laser systems produce enough power to illuminate the weld pool. However, they are expensive compared tolaser diodes which will sufficiently illuminate the weld pool using a cluster of high power laser diodes. On the other hand, LED diodes do not provide enough power to illuminate the weld pool due to their characteristics which arelisted in Table 1. Several key characteristics of laser diodes determine their usefulness as an alternative illumination source for the welding vision system.
Table 1. Characteristics of LED diodes and Laser Diodes
| Characteristic | LED diodes | Laser diodes |
| Current | 50 to 100 mA | 5 to 40 A |
| Power | Low | Much higher |
| Speed | Slow | Fast |
| Emission pattern | Wide | Narrow |
| Wavelengths available | 0.66 to 1.65 µm | 0.78 to 1.65 µm |
| Spectral width | Wide (40-190 nm FWHM) | Narrow (7 nm to 10 nm FWHM) |
| Ease of use | Easy | Hard |
| Lifetime | Longer | Long |
| Cost | Cheaper | Cheap |
| Pulse width | Long (min 100 µs) | Very short (min 30 ns) |
Results
Diode laser
The results obtained using a diode laser as an illumination light source are shown in Fig.8. Its wavelength of emission is 808 nm and it peak power is 200W. Although the peak power is relatively low, its pulse width is relatively long compared to other lasers, which can be manually set from 1µs to80µs. Pulse energy varies from 0.15 mJ to 15 mJ depending on the pulse width; the longer the pulse width the higher the pulse energy. For maximum power, a pulse width of 80µs was used, hence the exposure time was also set to 80µs.
Fig.8a) TIG welding on stainless steel, 50A, 80µs exposure time
Fig.8b) TIG welding on stainless steel, 100A, 80µs exposure time
Nd-YAG laser
This is a FlashLamp pumped Q-switch Nd-YAG laser with a frequency doubling option, pulse energy of 20 mJ and pulse width of 5ns. Figures 9a and 9b show the results obtained when the laser is operated at 532nm.
Fig.9a) TIG welding on stainless steel, 100A, 10µs exposure time, backward reflection
Fig.9b) TIG welding on stainless steel, 150A, 10µs exposure time, backward reflection
Copper vapour laser
This is considered a pulsed laser although repetition rate is 10kHz, which means that synchronising the laser pulse with the camera shutter is not possible due to the camera's limited frequency. The exposure time was set to100µs. This was the longest exposure time possible without capturing too much arc light. The pulse width is 5ns, average power is 5W, peak power is 70 kW and pulse energy is 2.75 mJ. Figure 10 shows the images obtained when the copper laser was used as an illumination source. They appear to be out of focus due to the narrow depth of field, which is governed by the aperture setting. The smaller the aperture, the greater the depth of field. In this case the aperture was set to 2.0, resulting in a small depth of field.
Fig.10a) TIG welding on stainless steel, 150A, 100µs exposure time, forward reflection
Fig.10b) TIG welding on stainless steel, 200A, 100µs exposure time, forward reflection
Continuous wave laser
Figure 11 shows the images obtained using a high power Nd-YAG continuous wave laser as an illumination source. The emission wavelength is 1064 nm, ie operating at the end of the camera's spectral response. The arc light is totally eliminated due to spectral and temporal filtering and operating at the end of the spectral response where emissions are mainly from the weld pool.
Fig.11a) TIG welding on stainless steel, 100A, 20ms exposure time, backward reflection
Fig.11b) TIG welding on stainless steel, 160A, 20ms exposure time, backward reflection
Laser diodes
Figure 12 shows the images obtained using a low-cost laser diode system as an illumination source with an emission wavelength of 905nm. The arc light is totally eliminated due to spectral and temporal filtering.
Fig.12a) TIG welding on stainless steel, 80A, 10µs exposure time, forward reflection, LD type 1
Fig.12b) TIG welding on stainless steel, 50A, 10µs exposure time, forward reflection, LD type 1
Fig.12c) TIG welding on stainless steel, 100A, 10µs exposure time, forward reflection, LD type 2
Fig.12d) TIG welding on stainless steel, 80A, 10µs exposure time, forward reflection, LD type 2
Discussion and conclusions
The vision system described in this paper is a low-cost system which has been shown to remove the arc light almost totally to produce reliable and high quality real-time welding images. The system incorporates a CMOS camera with a lens and narrow band pass filter along with a frame grabber and an illumination source. Several laser illumination light sources were used, including a pulsed laser, continuous wave laser and laser diodes. It has been demonstrated that novel and low-cost laser diodes can be used as an alternative illumination system to the bulky and expensive laser systems.
