Identify Suspensions
Units that have not failed by the failure mode under investigation are called suspensions or censored units. Suspensions have either not failed at all or have failed by an entirely different failure mode. Suspensions are categorised based on how their ages compare to the length of service (or age) that the component has so far attained. In engineering, suspensions generally refer to units with true times to failures greater than the oldest age for the failure mode under consideration. However, other types of suspensions exist and are categorised based on age:
• Early suspensions. Units whose failure age is less than the earliest failure age for the failure mode in question. Early suspensions have little effect on the Weibull probability plot. Also known as left-censored data, early suspensions are not often found in engineering data. During a medical prevention study, left-censored data is created when a person joins the study after learning that he or she already has the disease. Because contraction of the disease occurred prior to joining the prevention study, this occurrence has an age that is less than the first failure (occurrence) that develops during the course of the study.
• Intermediate suspensions. Units that have random failure ages for failure modes other than the failure mode in question. Intermediate suspensions, also known as random suspensions or progressive suspensions, tend to shift the Weibull line somewhere between the early and late suspensions.
• Late suspensions. Units whose failure age is greater than the oldest failure age for the failure mode in question. Late suspensions may reduce the slope of the Weibull probability plot. Also known as right-censored data, late suspensions are a concern in engineering data. During life testing, right-censored data is created by the removal of a part before failure. While it is known that the part operated successfully for a given period of time, the length of time it may have continued to operate is unknown.
Although not weighted as much as failures, all identified suspensions must be included in the sample data set. Because suspensions have no effect on adjusted ranks or median ranks until after they occur, the procedure is to rank the data with the suspensions first and then to adjust the ranks. While adding suspensions generally has little effect on the slope (β), it does tend to increase characteristic life (η). Thus, failing to include suspensions can yield results that are too pessimistic.