Case Study: ‘Poka Yoke’ Check Facility

IMG_0607Our client instructed ESA to produce a new vision checking facility to provide error proofing on their existing line for the turbine ‘shaft and wheel’ assembly.

Essentially the system is looking for a Bearing-Spacer-Bearing combination on the shaft, combined with a Piston Ring. Through initial checks, it soon became apparent that this facility had to cope with many more variants. Differences in the shape profile, quantity of rings and the overall sizes.

IMG_0881These ‘shaft and wheel’ assemblies were placed into a series of groups by the Client and each program had to follow these groups.

The system consisted of a Cognex camera from the Insight 7000 series to take an image when the product is placed into a profiled receptacle (cup) and then transferred into the checking area on a linear slide

IMG_0894The method of approach taken to check for all the correctly placed parts involved casting a silhouette by backlighting the product. Checking for correctly placed parts could now be done across the range by looking for changes in the contour of shaft.

We now look for a Piston Ring to absent from the oil well groove, and to be present and correctly seated in the Piston groove.

This detection of the Piston Ring had to be adaptive since each ring and groove combination varied from a single ring, two rings or a double ring (two rings in a wider groove). In attempting to detect this ring, issues still needed to be overcome, the fact that the ring can move around the groove and be justified to one extreme or the other.

Within the camera’s vision tool library, we used an ‘Edge’ tool to detect both extreme edges of ring from the 2D image. A distance measure could then be taken between the edges to give a value that is checked against an acceptable tolerance range. But this proved not very reliable due to two reasons, part variation within each group of parts and the quality of the silhouette image.

We were using the neck of the ‘shaft and wheel’ groups but found this neck on the product varied too much and thus could not be used as the definitive reference/datum point detection.

In order to overcome this we introduced two methods of post image filtering, the first post image filter essentially produces a negative of the original acquire image.

The top left of the shaft neck can be clearly seen against the black background, in using an ‘intersection’ tool with a large search area the reference/datum point could be defined for all variants for each group.

This new datum method proved successful through repeated testing, as the reference point never changed across the shaft & wheel grouping.

The second post image filter relied upon an ‘edge contrast/graduation’ tool, which produces an outlined or contoured version of the original image. This contoured image assisted in the detection of, and increases the reliability of correctly identifying the required number of Piston Rings.

Typical screen shot showing the vision tool groups, etc.

Each program for each ‘shaft and wheel’ group could then take the same format.


Ultimately, what seemed a straight forward application became a lot more complicated due to product variation, With the clever programming ESA were able to produce a system that’s could handle 20 shaft & wheel groups for product variants totalling over 50 and comply with the client’s wishes.