Alphanumeric character recognition on gas turbine blades

Many applications exist in which there is a desire to analyze object surface features within relatively unstructured operational environments.

One such example involves the machine reading of punched alphanumeric characters, previously applied to the surface of gas turbine blades. Here the base or root section of each blade is presented to a vision system for identification. Unfortunately, any slight variation in object pose can make the application of conventional structured lighting techniques, such as dark-ground illumination difficult to reliably implement. The problem is illustrated by the figures below.

Character 1
(a) Characters Clearly Visible

Character 2
(b) Characters Almost Imperceptible

At the correct attitude (Figure (a) above) it may be observed that the punched characters are readily discernible, however, when the blade surface is subject to only a relatively small alteration in attitude (Figure (b) above), of a few degrees, the characters may become almost illegible. The characters themselves are applied as a rectangular matrix (5 by 7) of circular dot-punched depressions, which locally alter the topography of the blade surface. The impressions made by the punch markings are discernible in terms of the local disruption caused in the light reflected from the object surface. As a result, the complex topography of the punch depressions cause incident light to be scattered. Under suitable conditions this diffuse behaviour tends to contrast well with the more specular behaviour of the surrounding metal surface, rendering the individual characters clearly visible. Unfortunately, in order to use a conventional structured lighting technique it becomes necessary to precisely fix the position and orientation of the blades relative to the incident illumination.

An alternative, less environmentally restrictive approach, is offered by considering the character markings as a surface description known as a 'bump map', applied to the underlying planar geometry of the turbine blade. In this manner the characters are apparent as perturbations in the nominal surface normals and hence, by using an object-centred approach, may be isolated at almost any position or attitude of the blade. Figure (a), below, shows an example acquired image. Figure (b) shows a rendered synthetic virtual image of the derived surface bump map. Using this approach, differing lighting configurations (in terms of geometry or even colour) can be positioned within the virtual viewing environment, in order to highlight details of interest. It may be noticed that, as expected, no surface data is captured within the blade holes, and hence only noise appears in these regions. Notice also that the details of the punched impressions appear more clearly in the synthetic image, when compared to the camera-acquired view of Figure (a). By suitable configuration of the virtual viewing environment, it is possible to give the surface a more three-dimensional appearance. Figure (c) shows the isolated surface albedo, or reflectance. Any variation in surface colouring should be visible in this view, but not in Figure (b) or Figure (d), which show only surface topography. (In theory, the stamped lettering should not be visible in the albedo image. The fact that it is would seem to indicate that the surface reflectance is altered within the punched marks. This may be attributed to a slight change in the microstructure of the surface caused by plastic deformation.) Figure (d) shows a bump map thresholded binary view of the surface bump map. Here all surface normals that are perturbed by more than 3 degrees, from the nominal surface normal, are shown in black. This demonstrates the successful isolation of the lettering. The white regions within each character are due to the relatively flat bottom of the punch markings.

Character 3
(a) Camera-Acquired Image

Character 4
(b) Synthetic View of Bump Map

Character 5
(c) Surface Albedo

Character 6
(d) Thresholded Bump Map

Hence by employing a photometric method, and given that the acquired surface data is essentially object-centred, it is not necessary to precisely constrain the position or orientation of the component part on presentation, (within limits imposed by occlusion). In addition, the lighting and camera configuration are able to remain relatively fixed across differing component geometric designs.

Back to top