A knowledge-based computer system has been developed to
automatically scan and interpret graphical displays
of vibration data taken from strain gauges mounted
on the blades of gas turbines performing
acceleration tests. The various characteristic lines
which appear on these displays are recognised and
classified and the lines representing modes of
vibration are examined to determine which mode is
being excited and the severity of the excitation.
All lines are compared, by a “blackboard” technique,
against those predicted by the system working from
the engine specification, laboratory tests and
finite element analyses. Any unexpected features are
reported for manual inspection.
Several series of tests have now been made on the
system. In the first test on 130 diagrams, every one
of over 3000 features visible to the human eye was
detected and then classified.