Biomedical Engineering Reference
In-Depth Information
In this section, we communicate the preliminary results of neural-network-based
vision system investigation. We have proved the system in off-line mode. For this
purpose, we have manufactured four groups of screws with different positions of
the machine tool cutter. The images of these screws were used to train and test the
developed system.
The task of shape recognition is well known [
10
]. The recognition of screw
images, in our case, is based on the recognition of the screw shape or profile. We
detected the screw contours and used this presentation as input for the recognition
system. The vision system was based on the neural network with the permutation
coding technique described in [
11
,
12
]. This type of neural network showed good
results in handwriting and face recognition tasks. Now we prove it for adaptive
algorithms in manufacturing [
13
].
9.2 The Problems of Adaptive Cutting Processes
To increase the precision of micromachine tools, it is possible to use adaptive
cutting processes [
2
]. Let us consider a lathe equipped with one TV camera, which
could be moved automatically from position 1 to position 2 (Fig.
9.1
). The images
from the TV camera in the first position could be used to measure the sizes of
partially-treated workpieces and to make the necessary corrections in the cutting
process. The images from the TV camera in the second position could be used to
correct the position of the cutting tool relative to the workpieces (Fig.
9.2
).
In both positions, the TV camera can give useful information about the passing
of the cutting process, for example, about chip formation, the contact of the cutter
with the workpiece, and so on. All such images must be treated with the image
recognition system. We propose to create such a recognition system on the basis of
a neural network with permutative coding.
Fig. 9.1 The lathe equipped with TV camera
