• Statistical Methods for Reliability Data (Meeker & Escobar, 1998) remains a foundational and well-respected text for analyzing failure-time and survival data

• Along with the text, the authors developed an S-Plus software package to utilize the methods for industry data

• Today, R is the most popular statistical computing language in the world - largely supplanting S

• This presentation introduces tools under-development for use with the current and future versions of the SMRD to

  • Analyzing industry and, laboratory test data
  • Simplify reliability/survivability instruction in the classroom

1. R Package SMRD

   • Implements methods from the text to reproduce results from the text
   • Implements methods from the text for use on industry data
   • Interactive Shiny gadgets to instantly perform reliability analyses and generate reports

2. Statistical Methods for Reliability Data in R

   • Expanded package documentation and example vignettes
   • In-depth examples corresponding to each chapter in the SMRD text

3. R package teachingSMRD

   • Examples from the SMRD reproduced as interactive shiny apps
   • Automatic generation of assigments and solutions for SMRD exercises

• A statistical programming environment for data analysis and graphics

• Developed by Ross Ihaka and Robert Gentleman at the University of Auckland

• Open-source implementation of the 'S' language created by Becker et. al. at Bell Labs

• A pre-eminent tool for statistics and data science

• One of the fastest growing technical computing languages in the world

  • Used for data processing and visualization, computational statistics, and natural language processing etc.

  • Heavily used by Google, Facebook, Twitter, Microsoft, etc.

• In R, the fundamental unit of shareable code is called a package

• Packages bundle together code, data, documentation, and tests to easily share analysis methods with others

• Currently 10773 packages are available on the Comprehensive R Archive Network (CRAN)

• Many more available from the Bioconductor and GitHub repositories

• The huge variety of packages is a key reason why R is so popular

  • Chances are that someone has already solved a problem that you're working on

  • You can benefit from their work by downloading their package

• Meeker developed a large collection of FORTRAN subroutines as part of contracted efforts at Bell Labs and Iowa State

• Meeker & Escobar wrapped the FORTRAN code into an S-Plus package called SPLIDA (S-Plus LIfe Data Analysis)

• SPLIDA serves as the companion software for Statistical Methods for Reliability Data 1st ed.

• Meeker began an effort to translate SPLIDA into R under the name RSplida

  • Not user-friendly - couldn't be installed as a traditional R package

  • Difficult to use with modern IDE's (i.e. RStudio, Visual Studio, Eclipse, etc.)

• (2015) Freels & Meeker sign MOU to share FORTRAN code for purpose of developing an R package

• Aim to publish SMRD to the CRAN in 2018

• Remaining tasks to be completed before publishing (% complete)

  • (90%) Update older R & S-Plus idioms to modern equivalents
  • (10%) Translate FORTRAN code over C++
  • (75%) Update graphics objects
  • (75%) Document datasets
  • (15%) Document exported functions
  • (75%) Update for modern use-cases - literate programming/interactivity
  • (50%) Ensure compatibility with modern dependencies

• Multiple failure modes
• Censored observations (right, left, and interval censoring)
• Truncated observations (right, left, and interval truncation)
• Failure data with explanatory variables (failure-time regression)
• Repeated measures degradation data (linear & non-linear mixed effects)
• Repairable system failure data (recurring events)
• Physical/performance degradation data
• Failure data with prior information (Bayesian reliability)
• Reliability growth test data
• Reliability test simulations

• Organizations often use different terms to describe the same event

  • 'Failure' = 'Failed' = 'Fail' = 'dead' = 'died'
  • 'right' = 'rcensored' = 'suspended' = 'alive'
  • 'left' = 'doa' = 'lcensored'
  • 'interval' = 'int' = 'icensored' = 'grouped'

• Many applications force users to recode these events

• SMRD allows for flexible event definitions to utilize the data as-is

• Event definitions can even be mixed

• SMRD event definitions easily mapped to survival numeric definitions

• SMRD includes over 120 fully-documented datasets

• For importing external data, SMRD leverages several other R packages

• Excel files

  • XLConnect, readxl, xlsx

• CSV/TSV files

  • base, utils, readr, data.table

• Info, Minitab, S, SAS, SPSS, Stata, Systat and Weka files

  • foreign, HMISC

Literate programming: integrating text with snippets of executable code in documents & presentations

• In R, literate programming is supported by the knitr and rmarkdown packages

  • Weave code from multiple languages, \(\LaTeX\)-typeset equations, and text together in one document
  • Run and compile code, \(\LaTeX\), text simultaneously
  • Everything is stored in a single file
  • Output to any of a number of presentation or document formats

• With SPLIDA, many results were returned simultaneously

  • Graphics
  • Numeric values
  • Tables of results
  • Text summaries

• For GUI-based software tools, presenting multiple results is GOOD

• For tools emphasizing reproducible research and literate programming, presenting multiple results simultaneously is BAD

  • SMRD built to support literate programming - go from data to report fast
  • Ensure that specific results can be produced and called where desired

• This example demonstrates some of the SMRD functions to analyze the shockabsorber dataset used throughout the text

• Many of the methods in the package require a life.data-class object

shock.ld <- frame.to.ld(frame = shockabsorber,
                        response.column = 1,
                        failure.mode.column = 2,
                        censor.column = 3,
                        time.units = 'Kilometers')

• Since SPLIDA was written as a GUI, many functions to produce results and graphics already existed

• Thus, once the life.data object has been created, many different plots and numeric results can be produced, each requiring only a single line of code

plot(shock.ld)
plot(shock.ld, distribution = 'lognormal')

mlehazplot(shock.ld,  distribution = 'lognormal', param.loc = 'topleft')
mleprobplot(shock.ld, distribution = 'weibull', param.loc = 'topleft')

simple.contour(shock.ld, distribution = 'sev', threeD = T, original.par = F)
simple.contour(shock.ld, distribution = 'sev', show.confidence = F, zoom = 1.75)

tab <- print(mlest(shock.ld, distribution = 'weibull'))$mle
xarray(table = tab)

\[ \begin{array}{rrrrr} \hline & MLE & Std.Err. & 95\% Lower & 95\% Upper \\ \hline mu & 10.23 & 0.11 & 10.01 & 10.45 \\ sigma & 0.32 & 0.07 & 0.20 & 0.50 \\ Weibull (eta) & 27718.72 & 3046.02 & 22347.77 & 34380.49 \\ Weibull (beta) & 3.16 & 0.73 & 2.01 & 4.97 \\ \hline \end{array} \]

• Teaching generates a lot of teaching files

• teachingSMRD is designed to manage your course content and make it easier to use SMRD in the classroom

  • Automatically generate assignments and solutions for SMRD exercises
  • Call up worked exercises in-class
  • Render examples as interactive shiny applications
  • Teach R programming skills and course content simultaneously

• Two package versions

  • Student version (published to the CRAN, solutions not included 👎)
  • Faculty version (not published, solutions included 👍)

Four Example Scenarios

• Viewing an interactive example from the text with shiny

• Generating a 'fill-able' assignment of multiple exercises

• Automatically generating a solution set for an assignment (.html and .pdf)

• Using shiny gadgets to instantly analyze a data set and generate a report

New reliability tools are under development for R

• R package SMRD - implements the methods presented in the Statistical Methods for Reliability Data text

• R package teachingSMRD - makes it easier to use SMRD in the classroom

• Statistical Methods for Reliability Data in R - extends the SMRD package documentation for in-depth examples advanded use-cases

Plan is to publish the packages to the CRAN late 2018