Volume 1, Number 2, Fourth Quarter 1993

RAC Journal

Volume 1, No. 2, 4th Quarter 1993

Opinion - Let's get rid of the term "Random Failure"

By: Kam L. Wong, Kambea Industries In the Spring 1993 issue of the RAC Newsletter Mr. Anthony Glaser attempted to clarify the meaning of a random failure. He feels that if one can estimate the approximate time of the occurrence of a failure then the failure is not random. From that explanation, I would add a corollary that if one can predict an MTBF for the next failure, the failure is not random as the approximate time of failure is known. Yes, there is a statistical variation on exactly when the next failure will occur. But there is a statistical variation of the wear-out life of an item too. In fact, in terms of hours the standard deviation of a wear-out life is a lot larger than the standard deviation of an MTBF. The standard deviation of the life of an integrated circuit might be in the order of 100,000 hours. Now, what is random?

Let's face it. Any failure is random. If not, a failure can not be called a failure. It is simply an anticipated point of life termination, such as using up a tank of gas in one's car. We would not call that a gas failure. True, in failures we are dealing with a stochastic process of random variables. But, in practice we don't need to be that esoteric or to perpetuate the "random" mystique.

The bad connotations of calling a failure random are that: (1) as Mr. Glaser had explained, many people think that no cause can be assigned to a random failure. Consequently, no effort would be conducted to eliminate such failures; and (2) people associate the concept of random failure to constant failure rate. The truth is that failure occurrence in electronics is a non-stationary stochastic process. In general the failure rate of electronics decreases with respect to operating time (a more accurate way to put it would be with respect to cumulation of operating stresses). The decreasing failure rates have been displayed in practically all of the well kept failure versus age data.

I believe the concept of random failures was originally introduced to justify the use of the exponential law to explain failure occurrence, borrowing the statistical concept of random occurrences of rare events such as soldiers kicked by horses and thus providing a mathematical sophistication to boost the status of reliability engineering. We don't need such a sky hook anymore. The exponential law is in truth not applicable to electronics. If you disagree with me on that, please feel free to call me at (310) 372-4533 to get supporting data sources. I propose that we simply consider that all failures are random and drop the term "random" completely. We can classify failures into two categories: either failures with assignable causes (ACF); or failures with no assignable causes (NCF).

Back to
Journal Index

User Guide

Products
and Services

Contacting
RAC

*Related
*WWW Sites

Home

From the Reliability Analysis Center Webserver. Comments or questions can be submitted using e-mail or telephone at (315) 337-9933. (Last update: December 1, 1993, jmc)