US 6,748,341 B2Granted

Method and device for machinery diagnostics and prognostics

OwnerGeorge E. Crowder, Jr.inventor

Life of this patent

Estimating the term…

Loading patent…

Abstract

Machinery diagnostics and prognostics is an emerging engineering field that seeks to accurately determine the operational health of a machine without waiting for the machine to fail, performing maintenance that may not yet be required or, in the worst case, performing unnecessary maintenance that inadvertently causes other problems and hastens machine health deterioration. Accurate prediction of machine health (operability) enables operators to base machine maintenance on the machine's actual condition, in contrast to the common practice of time-based maintenance (e.g., perform maintenance every 100 hours). The “just-in-time” methodology of the machine health monitoring system (MHMS) of the present invention translates into significant cost savings by providing early warning of impending failures and thus reducing unanticipated catastrophic machine failures through preventative maintenance techniques (but no more or no less than is required) while simultaneously keeping false alarm rates low. The MHMS couples proprietary analysis, modeling and pattern recognition techniques to provide accurate machine health predictions. The MHMS's combination of technologies provides accurate, easy-to-understand, early indication of potential problems, enabling efficient and timely repair. In addition, MHMS technology can continuously “learn” from its own machine health monitoring experience, so that system accuracy and false alarm rate improve over time.

A computer readable medium having instructions encoded thereon that cause a computer system to perform an accurate machine health prediction upon receipt of raw monitored machine operating variables data from said machine, the instructions on the computer readable medium comprising:

(a) a data filter comprising a filtering algorithm operable for eliminating “spikes” in said raw monitored machine operating variables data while retaining legitimate “ups and downs” in the raw monitored machine operating variables data, thereby improving the predictive performance of transfer function models (TFMs);
(b) a TFM estimator comprising a multivariate non-linear transfer function estimation algorithm that uses historical or real-time calibration data comprised of values of machine operating variables from one or more normally operating machines to define normal monitored machine operating variables and to construct statistical multivariate non-linear TFMs of normal monitored machine operating variables;
(c) a TFM modelbase comprising (i) a plurality of TFMs;
and (ii) a set of sampled normal monitored machine operating variable values and operational condition values employed to develop said plurality of TFMs;
(d) a predictor operable for identifying the best TFMs in the TFM modelbase for the monitored machine operating variables data and the current set of real-time machine operational conditions, and uses said monitored machine operating variables to predict one-step-ahead values for said monitored machine operating variables;
(e) a comparator that enables the computer to compare current monitored machine operating variable values to the value of the monitored machine operating variables predicted by the appropriate TFMs for the same variables, and producing the algebraic differences therebetween, said differences defining residuals;
(f) a prognosticator component that receives the residuals for each monitored machine operating variable from said comparator and conducts statistical tests on said residuals to categorize each monitored machine operating variable as normal or abnormal, and employ the results of said statistical tests on each monitored machine operating variable to calculate an output, the value of the output being a measure of the overall probability of machine abnormality.

4 more elements

Abstract

Claims2