Vibration analysis for diagnosis of machines is a powerful instrument for
condition monitoring. Especially for diagnosis of antifriction bearings
various equipment and techniques have been developed and are available.
The envelope analysis is the most reliable method for bearing diagnosis.
Different bearing failures lead to different patterns of the envelope
spectrum. Checking the amplitudes of characteristic bearing frequencies
is mostly not sufficient for a reliable diagnosis. Therefore human experts
compare the pattern of the spectrum with typical patterns of bearing
defects. In this paper a diagnosis system is presented which imitates this
action of human experts. Using a neural net the system not only
distinguishes typical bearing defects but also detects unbalance, beating
of disengaged machine parts and other kinds of machine failures.