Downloaded from https://www.everyspec.com ADS-79D-HDBK expected to always fail in a spalling failure mode, then any method can be used to create the spall. In situations where multiple methods can be used to seed a fault, the method that induces failure the fastest should be used. It is recommended that a minimum of 3 test articles be used per failure mode to train the CI. Test rigs that are intended to measure prognostic accuracy and only enhance maintenance procedures can be tested in the same way as training diagnostic CIs.
The minimum recommended number of test articles for a seeded fault prognostics test should be determined by identifying the number of expected failure mechanisms (methods to seed a fault) for the given failure mode. Thus the minimum number of recommended test articles is the number of failure mechanisms (which determine the critical failure mechanism) plus two additional tests of the critical failure mechanism. D.5.2.6 Code the algorithms and test performance. After selecting a number of candidate algorithms for the CI, the algorithms are converted to software, typically through the use of engineering development software packages such as MatLab™.
These programs are easily configured to read the data files obtained in Section 5.2.5 and run through the algorithm calculations. The output of the calculations is then easily portrayed in graphs for use by the engineers and analysts in determining the performance of the algorithms. The first performance metric of interest is the detectability of the CI, or its ability to correlate with both the existence of the fault and its increasing severity over time. In the process of obtaining data for the CIs, the testing or data collection should strive to collect the sensor data of the fault as well as the physical dimensions or other characteristics of the fault (examples:
crack length, pressure drop) in order to correlate the CI value with the fault severity. Figure D-8 shows an example of such a detailed data collection. The values of the fault (crack size) are measured at specific intervals in the data collection (shown as the vertical lines in the graph to the left). It is obvious from the graphical depiction that the fault and CI exhibit closely correlated behavior. In this case, the correlation was done with a simple linear calculation. The CI should also be able to detect the fault within the limits specified by engineering analysis, and do so with a high degree of confidence.
If a specific crack length is known to be the threshold beyond which catastrophic damage occurs, then the objective would be to develop a CI with the capability of detecting the crack prior to reaching that threshold value. In the top portion of Figure D-9, the CI varies with the fault progression, but the general behavior of the CI alert would not provide a high confidence level of the fault‘s existence prior to reaching the threshold value (top horizontal line). FIGURE D-8. An example correlation of fault dimension and CI value 89