I have always understood that the initial steep descent is due to poor manufacturing in the general sense. Specific causes include: faulty components that passed test (less than perfect testing); poor assembly (dry joints internal to ICs and at board level); and similar items. Techniques to improve this initial reliability generally involve taking the assembly past that part of the curve: burn-in, possibly with accelerated ageing through vibration and temperature cycling.

My first thought was about design testing and modification to remove design errors. However this addresses initial unreliability of all instances. I suspect that the question relates to specific instances. I note though that instances of a new design is likely to exhibit a lower reliability than a mature design with extensive usage (i.e. the design has a 'bath tub' reliability curve in addition to each manufactured instance).

This is a personal view and not necessarily the view of my employer.

Robin

Dear Robin,

Thank you for your reply. Yes, we can use the reasons you appoint in order to improve some technological processes to reach more reliability for man-made things. But, for instance, a human just born has more chance to die than those who live a year. This rule holds for practically everything which was created.

The question does not addressed to a specific case. I have already wrote the paper concerning this subject, see my recent preprint here. The question arises because I plan to proceed the consideration in the same way, but can't find any related research results. It is mauvais ton to omit other researches results and I wish just hear about it.

The failure distributions to characterize the probability of failure over the lifetime of a component. What you describe is the "Infant mortality" zone of a bathtub curve. Many types of components typically will have a bathtub-shaped failure rate life distribution. It has proven to be particularly appropriate for electronic equipment and systems. This distribution is commonly characterized by a decreasing failure rate during the early portion of a component's life, a constant failure rate during the useful portion of its life, and an increasing failure rate during the wear-out portion of its life.

As mentionned above by Mr Cook, "infant mortality" is normally the result of poor design, the use of substandard components, or lack of adequate controls in the manufacturing process. When these mistakes are not caught by quality control inspections, an early failure is likely to result. Early failures can be eliminated by a “burn in” period during which time the equipment is operated at stress levels closely approximating the intended actual operating conditions. The equipment is then released for actual use only when it has successfully passed through the “burn-in” period. A 48 hour “burn-in” is usually adequate to “cull out” a large proportion of the infant mortality failures.

Chris Ar

Thank you for your response. This is really interesting info for me. I hope in our project we'll find some analogue of "burn-in" process for so called "black swan" risks.

BR

Sergei

I am not an expert in the human or 'living thing' side of your question but it does give rise to some interesting thoughts. These may also be relevant when we consider technical systems implemented using artificial intelligence or machine learning.

The key difference between living and technical systems, for want of better terms, is technical systems are fully formed immediately following manufacture while for living systems, birth is an just event during the creation process. Indeed for a living system there may be no finished state as degradation starts before the creation is complete. Many sub-systems are not complete at birth. A good example is the immune system. It needs stimulating for it to learn how to defend against illnesses. Hence the overall system is vulnerable to illness until such systems have developed. Hence infant mortality.

There is also a major difference in the prevention of infant mortality. You cannot 'burn-in' a living entity. Different species have different ways of addressing infant mortality. For humans, modern living conditions and medicine have improved matters. For species subject to predators, e.g. zebra and antelopes, they have evolved to flee from predators very soon after birth.

Hence, my view at present, is that the 'burn-in' equivalent will be very different for different types of system.

This is attributed to the infant mortality where debugging is required for correcting the materials and/or settings.