The algorithm was tried on real sensed data, from the GRF-2 scanner, along a toolpath from a manufactured cover plate pocket. The scanner was not very accurate, so first we recognized pieces of lines and arcs out of the noisy points from the scanner and defined those as our nominal curve. This is not a bad approximation as the NC milling machine tool actually moves only in straight line and curve segments. For each points from the scanner we find the closest point to this nominal curve and - eventually - increase the radius of the sphere around the nominal point to include the point from the scanner. Finally, we smooth the values from the radius 40 times and define the surface with circles orthogonal to the path. Our algorithm compares it to the tolerance spline model, a few runs produced a good idea of the minimum specifications. Notice that both nominal curves from the model and from the scanner are quite different at some spatial instances, certainly because of a scale factor or a deformation from the scanner. Accurate data from a CMM along a toolpath would produce a much more precise input for the algorithm.
The figures 8 and 9 represent the inner profile, and figure 10 is the outer profile of the cover plate pocket. For the first one, we have found that a radius around the nominal curve of the model should be more than 0.12 cm. For the outer profile, we have found that it should be more than 0.065 cm. It should be obvious that more precise results can be obtained with more runs. As one can see on the cross section of the outer pocket (figure 10), a few bad points can badly influence the result, specially if there is already an error between the two nominal curves.