Downloaded from https://www.everyspec.com ADS-79D-HDBK number of established CI algorithms. Engineering and scientific literature should also be searched for other promising feature extraction techniques . data collection to CI comparison FIGURE D-7. Example of typical signal processing steps from D.5.2.5 Obtain data to train and evaluate the CI. CI selection is application dependent, and the only way to ensure the CI is sufficient is to test the CI with data. In this example, we assume that technical obstacles to obtaining useful data are overcome and data sets are available for both known good IGBs and IGBs with known faults.
This data can be obtained in controlled laboratory tests, such as the test rig at the University of South Carolina, the Original Equipment Manufacturers or other service system commands and labs. Data from faulted components can be obtained from Seeded Fault Testing (See ADS-79D Appendix J) or from data collected from installed systems for which a CI has not been developed (a new fault or one lower on the priority list, for example). Test rigs can be used to train the CI, and test articles should be chosen based solely on the expected failure modes.
Seeding of faults is permissible but should concentrate on the actual failure mode regardless of the method by which the fault comes into existence. Thus, if the component is 88