Visibility in vision systems

Machine vision systems often lack proper means of producing high quality images for vision processing. One either fights with the lack of light intensity, the direction of lighting or instability of ambient lighting conditions, or, as in welding applications, with an excess of light. These conditions normally require delicate planning in order to achieve suitable filtering for the visioning process.

Inspection for Curved Metal Surfaces

Lighting has a significant role in the inspection of curved metal surfaces. The surfaces under inspection cannot reflect light straight to the camera. On the other hand the lighting must elicit adequate contrast between the flawed surface and its surroundings. The curve and texture of the inspected surface may cause unpredicted changes in the reflection of light. For example, the weakness of lighting methods of dark-field type is the narrowness of the lighting pattern. Instead, a diffuse source of light produces a lighting pattern that is quite even and extensive on shiny and curved surfaces. Diffuse illumination is also less sensitive to the texture of the surface.

Welding monitoring

Surveillance from a distance carried out by video cameras can be exploited when monitoring mechanised welding. Typically the parameters concerning the quality of the weld, such as the form and the penetration of the weld pool, are of the most interest. The visual information that is accessible in close to real time enables a quick reaction to various malfunctions. The distance surveillance system based on video monitoring is at its simplest composed of a video camera set at the welding location and a video monitor. The radiation of a wide spectrum and strong intensity caused by the arc is likely to lead to overexposure or even to the destruction of the camera’s CCD element. The radiation can, however, be significantly reduced by using optical filtering and by adjusting the aperture of the optics and the cell exposure time.

The object of the study was to find methods for developing video monitoring on a mechanised narrow gap TIG welding station. Various types of steel and shielding gases were used in measuring the radiation spectrum emitted by the arc. The results of the spectrum calibration were later used when choosing a filtering method. Test imaging was performed with CCD and CMOS cell equipped cameras which were connected to a PC via frame grabber board. In order to attenuate the intensity of the light reaching the cell an optical filter was mounted to the camera lens. The quality of the image was adjusted to an adequate level by regulating the aperture of the optics and the exposure time. Substantial information was gathered from the video image, such as the form of the melt in the gap, the location of the welding electrode in relation to the sides of the gap and the melting of the filler metal wire. The emphasised features enable the implementation of machine vision based monitoring of the welding station. For instance, the position of the electrode can be adjusted by means of image based measuring.

Optical Seam Tracking

The groove surveillance improves welding quality and productivity. There are various form and measure deviations in welded objects. The wire nozzle’s movement must be adjustable in accordance to the deviations of the welded object. Nearly all of the seam surveillance systems are based on optical measurement without contact.

The optical measurement can be performed by directly imaging the seam or by a triangle measurement principle, in which a laser line is projected on the object to be measured and then imaged. Direct imaging cannot be applied to the measuring of MIG welding, because the strong spark interference requires efficient optical filtering. The band width of the optical filter has the same wavelength as laser. Thus it is the laser line only that penetrates the filter unabsorbed.

In addition to optical filtering efficient methods for filtering and segmenting images are required. The spark interference in MIG welding is strong enough to be seen even after optical filtering. The interference must be eliminated from the image by nonlinear filtering. Laser lines are quite uneven. The profile can be read using the intensity centre point. The reading of the outline results in an uneven profile.

The seam is differentiated from the profile according to its derivate. The methods defined above have given good results when monitoring seams in MIG welding. Spark and smoke interference has been completely eliminated. The measuring accuracy of the method is +/- 0.15mm and the follow-up is reliable. The cost of the system’s equipment is only a part of the cost of a complete system. Investment in good design and efficient signal processing enable building reliable and cost-effective systems out of basic components.

Contact Information: Ville Saarinen, Jyrki Tuominen Satakunta Polytechnic Faculty of Technology and Maritime Management Tekniikantie 2, 28600 Pori Tel. +358 2 620 3166, fax +358 2 620 3160 ville.saarinen@samk.fi, jyrki.tuominen@samk.fi http://www.samk.fi