Original Date: 03/08/1999
Revision Date: 01/18/2007
Information : Statistical Characterization of Failure Data
Condition based maintenance (CBM) is the philosophy of repairing/replacing a part or component based on observed objective evidence. Integral to this philosophy is the early detection and recognition of impending mechanical failures. Cost benefits of CBM, especially for safety or increased equipment usage, are also an important aspect of reliable maintenance programs. One of the Applied Research Laboratory at the Pennsylvania State University’s (ARL Penn State’s) contributions to CBM involves using statistical characterization of failure data to evaluate and develop early warning alarms for impending mechanical failures.
The ARL Penn State built upon previous research to determine the effectiveness of using wide band signal processing algorithms (e.g., the continuous wavelet, transform-based, diagnostic algorithms adapted to the problem of gear-tooth crack detection). After initial trials with the wavelet algorithms proved successful on generated transitional data, the ARL Penn State continued its research on related but separate paths. In particular, the laboratory addressed the importance of fault-detection threshold settings and false-alarm performance. A combination of analytical and empirical data analyses were performed to establish a methodology for deriving user- specified, false-alarm performance for the wavelet design fault monitor.
The ARL Penn State transferred this technology to NASA Ames Research Center’s applied research wind tunnel platforms, where a similar analysis will be performed on the Center’s data. This application represents the value of advanced sponsored research for the purposes of benefitting government and industry. Future work will be directed toward searching for best processing algorithms for early detection, detection performance, and computational efficiencies; developing discrete wavelet transform research on acoustical emission data; analyzing wavelet coefficient data with neural networks; addressing shaft torsion analysis data; developing multisensor fusion algorithms; and supporting reasoning algorithm development.
For more information see the
Point of Contact for this survey.