Field failure rate may not be what you think

Abstract

Ramp up, commercialization or roll-out are all common terms for one stage of a project when it goes from a low level production rate to a high rate. During this time, it is common for new problems to arise and the time to failure remain unknown. When shipping systems without operating time clocks or serialization, only the quantities shipped and quantities replaced are known. Weibull modeling generated from such roll-out data can easily be misleading. This paper will show some common errors with these model attempts that can be avoided. The roll-out process itself is part of the problem. Often, this is a hurried phase of limited time that is followed by a longer and fairly steady production rate. Even when ramping with a constant failure rate situation, it takes more than six months for the Weibull model data to settle down and look constant. Add a second failure mode, one that occurs in addition to the constant failure rate and the result is a complex Weibull curve that doesn't reflect either mode well. This easily happens when the operating environment varies from customer to customer. Several examples will make this graphically clear. Lastly, some conclusions are presented and some suggestions that will help separate failure modes during the ramp up. These suggestions will help obtain better estimates for planning warranty costs and determining repair support necessary. The problem described is real; examples from the disk drive industry are cited where ramp up and multiple failure modes are intertwined [1, 2, 3].