AFR: problems of definition, calculation and measurement in a commercial environment

Most commercial enterprises want a meaningful yet simple method for determining product reliability. "Average", "annual" or "annualized" failure rates are selected because of their apparent simplicity. All involve some sort of "failure rate" averaging over some period of time. All are abbreviated AFR even though there are no accepted industry standard definitions for these terms or the processes and conditions for their measurement. If the failure rate is not constant in time, creating an average can easily generate meaningless results. A number of businesses use the term "failure rate" but are in fact calculating probabilities or time independent percentages. A basic premise for this paper is that the true underlying product failure rates are not constant in time. This paper presents a detailed discussion of different ways commercial industries calculate simple failure rate based metrics. It identifies some ambiguities in the definitions and resultant inaccuracies. It will help the reliability pundit understand the mathematical considerations, making him more aware of potential problems. This paper should help generalists and managers understand that the AFR's they calculate may be significantly different than those calculated by their suppliers and customers. This can result in substantially different numbers and conclusions. Averaging instantaneous failure rates should be done only to smooth data collected from a distribution known to have a constant failure rate.