 |
 |
Quality Control
Fuzzy Logic and Neural Networks for
Acoustic Quality Control
Quality control problems often consist of signals that have to be
analyzed for the occurence of certain patterns. The analysis of
acoustic signals is frequently employed in quality control to
determine the quality of final products, or in machine diagnosis to
detect machine faults.
The objective of analyzing signals is so as to be able to
classify them into one or more categories. Fuzzy logic and neural
networks are able to classify even highly complex patterns. Moreover,
fuzzy classification yields gradual, more detailed quality judgements.
The example on this page shows the application of
DataEngine V.i to vibration analysis of antifriction bearings. This
on-line diagnosis system uses the fact that the frequency spectrum of
the sound emitted by the bearing yields a characteristic shape
depending on the kind of bearing fault. Possible failures are, for
example, unbalanced or damaged bearing races.
The pattern indicating the type of failure may or may not be
obvious in the signal. This is complicated by simultaneously occurring
failures whose effects may overlap. For these reasons a fuzzy c-means
classifier is used in the analysis. The diagnosis system combines the
data acquisition from the acceleration sensors; signal preprocessing;
classification and visualization. It can easily be adapted to new
situations (such as new bearing types or different fault classes) by
simply re-training it.
Similar applications are found wherever rotating devices have to
be checked (tape recorders, motor, turbine test benches etc.). Similar
procedures can be applied for inspecting manufactured products such as
the quality control of ceramic products or materials.