Prognostic Techniques for Mechanical Failure
Prediction
Abstract:
BAI developed two new diagnostic and prognostic
algorithms for processing machinery vibration data: (1) an
AM/FM demodulation technique, and (2) a bispectral-based statistical
change detection algorithm. Both methodologies demonstrated
to be ready for transition into practical use.
Problem:
The essential capability missing presently in state-of-the-art
mechanical diagnostic systems is the ability to predict remaining
useful life (RUL). Prognostic information is critical as it
provides a quantitative measure on which to base subsequent
actions. It is the essential improvement needed in future
health monitoring systems. The central technical problem lies
in estimating the RUL of individual components (e.g., gears
and bearings), rather than that of a population of components.
Population statistics do not have much to offer, since the
pilot or equipment operator needs to know if the present system,
operating under existing or modified loadings, can complete
its mission before failure.
Solution:
The AM/FM demodulation approach demonstrated the ability to
extract from machinery vibrations the amplitude and phase
modulation signals, which are indicators of the severity of
a particular type of localized component defect. Amplitude
and phase modulation information was extracted using a neural
network computational methodology that relies on nothing more
than knowledge of the bearing geometry and the frequency tones
at which a given type of defect will manifest itself. The
approach can also exploit very high frequency vibrations,
such as measurements collected using laser interferometry
systems.
With the
SCD approach, successful fault detection and isolation (diagnostics)
and estimation of RUL (prognostics) were demonstrated using
two incipient fault datasets. In particular, the SCD algorithm
correctly predicted the failure time of an overloaded helicopter
transmission (due to a gear tooth root fracture) 10 minutes
in advance. The naturally-occurring incipient fault in this
case progressed from initiation to complete failure over a
time course of only 23 minutes. In a second dataset, taken
from a gear and pinion laboratory testbed, an incipient fault
progressed from initial gear tooth cracking to complete failure
over a time course of 5 days. The same SCD algorithm (no change
in its parameters) accurately predicted actual failure time
more than 2 days in advance. Success of the prognostication
algorithm over such widely differing time scales suggests
that it is robust.
Links and References:
Parker, Jr., B.E.,
H.V. Poor, M.P. Carley, R.J. Ryan, and
M.J. Szabo, “Helicopter transmission diagnostics using
vibration signature analysis,” Proc. 50th Mtg. Soc. Machinery
Failure Prevention Technology, Mobile, AL, Apr. 22-26, 1996,
pp. 419-430.
Larson, E.C., D.P. Wipf, and B.E. Parker, Jr., “Gear
and bearing diagnostics using neural network-based amplitude
and phase demodulation,” Proc. 51st Mtg. Soc. Machinery
Failure Prevention Technology, Virginia Beach, VA, Apr. 14-18,
1997, pp. 511-521.
Parker, Jr., B.E., H.A. Ware, D.P. Wipf, W.R. Tompkins, B.R.
Clark, E.C. Larson, and H.V. Poor, “Fault diagnostics
using statistical change detection in the bispectral domain,”
Mechanical Systems & Signal Processing, Vol. 14, No. 4,
2000, pp. 561-570.
|
