E. J. Sharp, Boeing Phantom Works, Seattle, WA; R. J. Weiss, Boeing, Auburn, WA
There is a large push in the aerospace industry to create a system that allows for electronic submittal and analysis of cerification data to the final customer from a supplier or test house. Analysis of this data is expected to be automated to look for trends and assignable cause variation within a process. However, one can not always expect a current and uninterrupted database from which to derive knowledge on how to control the process. Engineers would then rely on the real-time data to identify shifts or trends, but with some products, such as titanium forgings or castings, the amount of variation can be quite large, which can inflate control limits for processes with only a small number of data points. This work is aimed at using existing production data to test how control calculations for small pieces of a larger production run compare the the control calculations for the overall set of data. Randomizing this "tested occurrence" to happen at any given point within a historical data set will simulate the beginning of tracking a process. Comparing the results of these random beginning points to the overall results will help to show what kind of accuracy and control abiltiy is to be expected once an electronic data submittal system is brought online. Further, using a known problematic data set can help to also test the possibility that a marked process deviation can be deciphered during the early stages of process monitoring.
Summary: For some time now, the topic of electronic submittal and analysis of certification data for aerospace production parts has been discussed. One challenge facing the implementation of such a system would be in analyzing data from a supplier that has just come online with the system. The data may populate very slowly especially if the needed parts are produced at a low rate, are sampled at a low rate, or are divided among many suppliers. As such, tracking process changes may prove difficult without waiting sufficient time to gather extended data. With this work, I plan to experiment with known data sets to show the ability of predicting and using reliable control limits for smaller data sets, such as those found at the beginning of the process of electronic compilation. I hope to show a reliable means for choosing and utilizing control limits as a process matures.
