Pseudo Failure Time Script

[#1]

Purpose

This script takes a repeated measures data table (a table with multiple rows per unit)

and turns it into a data table with 1 row per unit and a Pseudo Failure time.

The original data table must contain:

1) A column that identifies each unit,

2) a variable that measures time (can be use such as miles driven), and 3) a

response column.

Usage

Simply run this script by any one of these methods:

  Edit > Run Script

  Control-R

  Click "Run Script" button in tool bar

//*******************General information

//This script takes a repeated measures data table (a table with multiple rows per unit)

//and turns it into a data table with 1 row per unit and a Pseudo Failure time.

//The original data table must contain:

1) A column that identifies each unit,

2) a variable that measures time (can be use such as miles driven), and

3) a response column.

//*****************Initial Dialog

//The initial dialog allows you to select the response, time measure

//(can be time or a useage measure such as miles driven), and a unit identifier.

//You may also specify 1 group-by variable. If a

//group by variable is specified, the user can select a different transformation

//for each level of the grouping variable

//The radio box at the lower left allows the user to specify whether the plots

//should display a regression line or connect each point in the series for each

//unit. This affects the display graphs but does not alter how the internal calculations

//are performed

//************************Select Transformation window

//After the initial variables are specified, the script opens a window displaying the

//relationship between the time variable and the response. Transformations for time

//and response can be selected frop the drop-down lists at the side of the

//plots. The plots will update when transformations are

//selected to show the new relationship.

//If a group by variable is specified, there will be a separate plot for each

//level of the grouping variable.

//When the user has found a transformation(s) that causes a linear relationship for

//transformed response vs transformed time, he/she can select

//"Create Pseudo Failure Time Data"

//to create the new data table.

//For more information about the Select Transformation window, select the Help

//button in the dialog box.

//**********************Create Data Table Dialog

//This dialog box allows you to enter the failure threshold,

//specify additional covariates (optional), and specify a column in the data

//table that controls the size of the pseudo failure time.

//The resulting data table will have ONE ROW per Unit ID.

//Details

//Covariates for Failure Model (optional):

//   An optional list of covariates to be included in the failure

//   time model. These columns are copied from the original data table to the

//   Pseudo failure time table. They do not affect calculation

//   If there are multiple measures for each covariate, the

//   LAST VALUE is included in the final table. At this time, the script does

//   not support time-dependent covariates

//Maximum Predicted Time (optional)

//    An optional column containing maximum predicted times.

//    If the predicted failure time from the linear models is greater

//    than the value in this column, the pseudo failure time is the

//    maximum predicted time from this column.

//Censoring Column (optional)

//   A column in the data table that has a 0 for observations that failed during testing

//   and a 1 otherwise. If this column is specified, any unit that has a 1 for this column

//   will be considered a "hard failure", and the pseudo failure time will be the time of the

//   last measurement taken on this variable.

//Failure Threshold (required)

//  The value of the response for which the unit would have been considered Failed.

//***************Details on Inputs, outputs, and internal calculations

//Inputs:

//   * Data table, as described in assumptions

//       --Will use active table or script will prompt for open

//   * Failure Level: damage level at which the component is considered "failed"

//       --User will type into a labeled box

//   * Censor Time (optional): if predicted time exceedes this time, the component

//                  should be considered "censored"

//       --The data table will have a column with the maximum predicted time

//         the user will allow for each unit

//   * Transformation of time and/or response

//       --User will select from a list

//       --User will be shown a plot of transformed and predicted values

//          and allowed to change selection

//Outputs:

//    * Interactive plot. Allows the user to visually fit

//      linear transformation

//    * Data table containing the following:

//      --a row for each component

//      --"Predicted Time": a numeric column containing the predicted failure time

//      --a numeric column containing the original failure time

//      --"Pseudo Failure Time" a numeric column containing the pseudo failure time.

//         This is the time predicted by the linear model or,

//         if censoring variables are specified,

//         the appropriate time of failure. This variable is based on a formula that

//         assigns either the original

//         time or the predicted time based on the user's specifications.

//      --"Censored" A 0,1 variable indicating if the unit is censored (1) or not

//         censored (0). This is the variable that should be specified as the

//         censoring variable in the Life Distribution platform or other survival

//         model.

//      --"Estimate Type": an ordinal colum indicating if the row is censored in

//          one of the following ways

//        --failed during testing: 0

//          (In this case, the modeling time is set to the last time that

//            unit was measured).

//        --did not fail during testing, and predicted time makes sense: 1

//          (pseudo failure time = predicted time)

//        --predicted censoring time is longer than the user's specified time: 2

//          (in this case, the modeling time is set to the maximum time

//           specified by the user)

//        --there was only one row for the unit--prediction was impossible: 3

//          (in this case, the modeling time is to the last time the unit was measured ).

//        --in some cases, the calculation for predicted time will result in a time

//          that is too large for JSL to handle. In that case, Pseudo fialure is set to

//          the maximum time or to the last time the unit was measured if

//          there is no maximum time column

//     --covariates from the original data table. There is a dialog that allows the user

//          to select covariates that will be used in a later analysis.

//     * Time-dependent covariates are NOT SUPPORTED. The last values measured for each

//       component appears in the data table. 

//Calculations:

//    * Linear transformation of the input values

//    * Predicted failure time using the linear transformation

//      --the model calculates a separate line for each unit

//      --the script performs the calculation using JSL's matrix language

//Assumptions:

//     * The data table is in the following form:

//        --1 row for each measurement on each item (repeated measures)

//        --Contains a time column specifiying when each measure was taken

//        --Contains a censoring column (0,1) indicating if the component

//          failed during testing.

//          --If the last measure for the unit has censoring=0,

//             the new data table will use the failure time given in the table and not the

//             predicted failure time.

//     * There is an appropriate linear transformation that works for

//       all rows or all rows within a by-group (i.e. we do not have situation

//       where some rows require a

//       log transformation and others require a power transformation)

//Platforms Used:

//    * The Bivariate platform is used to create the plots that allow the user to select

//      the appropriate transformation. Fit Each Value is used to allow the user to see the

//      if the fit is approximately linear.

[monicah22]

Quesiton about the method to find the PredictedTime to failure based on the response variable.

How is the time to failure calculated if there is data for before and after failure. Say the device is good at 24 hours but bad at 96 hours.

What about if there is only data before failure, but the device is on a failing trajectory (it is drifting)? How is the time to failure calculated for that?

Thanks