When you analyze how long a population of equipment lasts, the tidy case is that every unit has failed and you know its exact age at failure. Real field data almost never looks like that. At the moment you run the analysis, many units are still happily running, and others were removed for reasons unrelated to failure. Those survivors carry real information, they have lasted at least as long as their current age, but you do not know their failure age because they have not failed yet. That is censored data, and handling it correctly is one of the most important skills in life-data analysis. This page explains what censored data is, why right-censored survivors must be treated specially, and how ignoring them biases life estimates low.
Censored data in reliability in one line: Censored data are life-data records for units whose exact failure time is not known, most commonly right-censored units, also called suspensions or survivors, that were still running when the study ended or were removed before failing. These records tell you the unit lasted at least as long as its observed age, which is genuine information that proper reliability methods incorporate rather than discard. Ignoring survivors and analyzing only the units that failed makes the population look worse than it is, biasing life estimates such as MTBF and Weibull parameters toward shorter lives.
The defining feature of censored data is incomplete knowledge of a failure time. In the most common case, right censoring, a unit is still operating when the analysis is performed, or it was taken out of service for a reason unrelated to the failure mode being studied, such as a planned upgrade or an accident. For that unit you do not know when it would have failed, but you do know something valuable: it survived at least up to its current age without failing. A unit that has run for a long time without failing is evidence that long lifetimes are achievable, and that evidence is exactly what a life analysis needs, even though there is no failure time attached to it.
These survivor records go by several names, most often suspensions or censored units, and they are extremely common in field data. At any given moment, a fleet contains many units that are younger than the ages at which failures occur, plus older units that simply have not failed yet, and all of them are censored. It is entirely normal for the majority of a fleet's records to be suspensions rather than failures, particularly for reliable equipment where failures are rare. Treating this as missing or unusable data would throw away most of the information the fleet contains.
It helps to distinguish right censoring, by far the most common, from its relatives. Left censoring means a unit had already failed before the observation started, so you know only that its life was shorter than the observation age, and interval censoring means a failure occurred somewhere between two inspections without the exact time being known. Field reliability work is dominated by right censoring, the running survivors, but recognizing the type matters because each is incorporated differently. In every case the principle is the same: the record constrains where the true failure time lies, and a correct analysis uses that constraint rather than pretending the record does not exist.
The tempting shortcut is to analyze only the units that have actually failed, because those have clean failure times, and to leave the survivors out entirely. This is a serious mistake, and it always biases the result in the same direction: toward shorter life. The reason is selection. The units that have failed so far are, by definition, the ones that failed early enough to be seen, while the long-lived units are still running and would be discarded. Fitting a life distribution to only the failures effectively asks how long the short-lived members lasted, ignoring all the evidence that many units are lasting much longer, so the estimated typical life comes out too small.
Concretely, quantities like mean time between failures, a Weibull characteristic life, or a computed B10 will all be understated if survivors are dropped. The analysis sees a cluster of early failure ages and no counterbalancing evidence of the survivors that outlived them, so it concludes the population is less durable than it truly is. For decisions that depend on these numbers, sizing spares, setting maintenance intervals, rating a design life, this pessimistic bias is not conservative in a helpful way; it distorts the trade-offs and can trigger unnecessary replacements or overstocking based on a life estimate the equipment beats in practice.
The correct treatment is to include the censored records in the fit using a method built for them, which reliability software and standard life-data techniques provide. Maximum-likelihood estimation, and probability plotting with the appropriate adjustments for suspensions, both incorporate each survivor as the information that its true failure time lies beyond its observed age. The fit then balances the observed failures against the survivors' evidence of longevity, producing life parameters that reflect the whole population rather than just its early failures. The practical rule is simple and firm: never fit a life distribution to failures alone when survivors exist, always include the suspensions.
Field reliability analysis lives on censored data because operating fleets are, at any instant, mostly survivors. To use them you need each unit's current age or accumulated runtime and a clear flag for whether that age represents a failure or a suspension, and getting both right is where operational records matter. A control or SCADA system that tracks runtime hours gives the age for every unit, failed or not, and its event and status logs mark which units have failed and when, so the raw material for a properly censored analysis is present rather than having to be reconstructed from memory.
A cloud SCADA platform such as Merobix helps by maintaining this runtime and event history across many units and dispersed sites in one place, which is precisely what a censored life analysis needs: the accumulated age of every survivor alongside the failure ages of the units that did fail. Without a central record it is easy to capture only the dramatic failures and lose track of the quiet survivors, which is the very omission that biases the analysis low. Keeping the full population, survivors included, visible ensures the suspensions that outnumber the failures are available to be counted.
There is also a data-hygiene angle worth flagging too. A record must correctly distinguish a genuine failure of the mode being studied from a unit removed for an unrelated reason, because a planned replacement or an unrelated fault is a suspension, not a failure, and mislabeling it as a failure biases the estimate the other way. The timestamped, categorized event history a monitoring platform keeps makes that distinction traceable, so an analyst can separate true failures from suspensions rather than guessing. Handled this way, the everyday runtime and event data a SCADA system already collects becomes exactly the correctly censored life data that unbiased reliability estimates require.
Right censoring means a unit's failure time is known only to be greater than its observed age, typically because the unit was still running when the study ended or was removed before failing for an unrelated reason. These survivors, also called suspensions, tell you the unit lasted at least as long as its current age, which is real information. Right censoring is by far the most common form of censoring in field reliability data.
Because analyzing only failed units selects for the short-lived members of the population while discarding the long-lived survivors that have not failed yet, the fit sees only early failure ages and no evidence of longevity. As a result estimates like MTBF, Weibull characteristic life, and B10 come out too small, making the equipment look less durable than it is. Including the suspensions balances the failures against the survivors' evidence and removes this pessimistic bias.
Standard life-data methods incorporate each censored unit as the information that its true failure time lies beyond its observed age, rather than dropping it. Maximum-likelihood estimation and probability plotting with suspension adjustments both do this, letting the fit balance observed failures against survivors. Reliability software handles the bookkeeping, so the practical requirement is simply to record every unit's age and whether it represents a failure or a suspension, and to include the suspensions in the analysis.
Merobix reads your field devices into a cloud SCADA - the real thing behind these terms, live in days from any browser.