A comparative study of prediction techniques

Abstract

2) Issue-interval Technique 1 underestimated much of the time, especially in the early months of the program.(3) Technique 1 appears to be more accurate than the other techniques for engine parts and communications (Groups 1 and 5).(4) Technique 6 is preferred among the service-life techniques, but fewer than 10 percent of the sample line items show evidence of age-related demands.(5) One of the upper-bound techniques (4) seems to be preferred for gunnery and accessories (Groups 3 and 6) with a switch to Technique 1 if the demand rate stabilizes.(6) Upper-bound Technique 2 seems adequate for airframes (Group 2). (7) Upper-bound Technique 3 seems preferred for fire-control components (Group 4).A great deal of work is yet to be done. These seven prediction techniques were applied to data on failures during the checkout of 27 components of an air-to-air missile, and results of that study need to be synthesized with the results reported here. A study has also been made which uses the engineering estimate of a service-life factor delineated early in the program, before any demand data have been collected. One solution to the demandprediction problem may involve a routine which originates with the first projection of an engineering estimate and systematically uses two or three of these techniques, weighted in such a way as to take advantage of evolving information about part changes, program changes, maintenance policy changes, etc. We do not expect any of these studies to be the last word on de mand-predi ction techniques .