Additional Resources > Windchill Risk and Reliability Practitioner’s Guide > Weibull Analysis > Gathering “Good” Life Data > Identify the Data Type
Identify the Data Type
When the precise failure or suspension time for each point in the data set is known, the data is point-by-point. Considered the standard type of data for Weibull analysis, point-by-point data is classified into occurrences (failures) and suspensions. For an occurrence, the failure age or event is recorded precisely at a point on the time scale (t). For a suspension, the removal of the unfailed unit is recorded precisely at a point on the time scale, even though its true time to failure is actually greater than the age attained so far (> t). Most controlled test data is point-to-point because the length of the testing period and the time of failures are known. When all failure times are known and good estimates can be made of suspension times, warranty data can also be classified as point-by-point.
When exact failure and suspension times are not known, the data is grouped by failure intervals (or number of units). Grouped data is considered dirty because it causes the uncertainty of the analysis to increase. When handled in monthly counts of failures without exact failure and suspension times, warranty data is considered grouped data. Terms used to better describe grouped data include:
• Interval data. Involves benign (or dormant) failure modes that are only found when the component or system is shut down and inspected at periodic intervals. When a benign failure mode is found upon first inspection, it is called a discovery. The true time to failure for the failed part is actually less than the age recorded at the first inspection (< t). A benign failure that occurs after the last inspection time (t1) but is not discovered until the next inspection time (t2) has a true time to failure greater than the previous inspection age but less than the detection age.
• Coarse data. Related to interval data, coarse data has less precise time to failures because the intervals between data collections are too long, perhaps even months rather than days or hours.
• Probit data. Also known as destructive inspection data, probit data is obtained when every part is inspected at every inspection due to the additional uncertainties that are related to detecting or finding failures during inspection. For probit data, each observation is either considered to be a suspension or a failure. For example, when bombs and missiles are tested (or eddy currents are inspected), they either do or do not work.
Because the type of life data determines which distribution type is best, Table 7-3 describes the selections that are commonly found in Weibull software for indicating how data points are collected.
Table 7-3. Data Types and Descriptions
Type
Description
Point-by-point
Provides for entering the failure and suspension data when the precise failure or suspension time is known for each point in the data set. When 20 or fewer of such data points exist, the standard method is to select the Weibull distribution and use median rank regression as the parameter estimation method.
Point-by-point/Inspect
Provides for entering the failure and suspension data when the data is specified in periodic inspection intervals. This classification also provides for defining the interval frequency.
Grouped, Probit 2
Provides for entering the failure and suspension data from repeated tests on the same units by occurrences. This method compares the cumulative number of failures to the number of inspected units at various points in time. When a new unit replaces a unit that failed in a previous inspection, it is added to the number of failed units as well as to the number of inspected units. This classification also provides for entering a varying number of inspected units at different ages.
Grouped, Probit 3
Provides for entering the failure and suspension data from non-repeated tests on varying sizes of units tested at different times by percentages. This method compares the cumulative percentage of failures to the number of inspected units at various points in time. Such tests are sometimes found in destructive inspections. Because the cumulative failure distribution is an increasing function in time, the cumulative percentage failed tends to increase with time for most destructive tests. However, considering the random nature of failures, this may not always be the case. This classification also provides for using the varying number of inspected units at different ages.
Grouped, Kaplan-Meier
Provides for entering the failure and suspension data when the exact failure time defines the intervals, which means that failures and suspensions occur at the end of the interval. This method can also be used for intervals that are not same, especially if actuarial corrections are used when entering the data. This method accurately estimates the cumulative distribution without making any distribution assumptions.
Interval MLE
Provides for entering the failure and suspension data in a generalised data format for when Maximum Likelihood Estimation (MLE) or Modified Maximum Likelihood (MMLE) is the parameter estimation method. (Refer to Specifying the Estimation Method.) Occurrence, suspension, discovery and intervals for the data set can be specified, and the interval can be defined.