README.md

In lxy009/weibullpp: weibullpp

weibullpp

Checkout the project's site hosted on github pages.

weibullpp is an R package designed for reliability engineering. It is meant to add to community of packages in R for survival analysis. See CRAN Task View: Survival Analysis for other packages. Another goal is to offer an open-source alternative to commercial software such as Reliasoft's Weibull++. To this end, many plots and calculations available in Weibull++ will be made available in the package.

As of version 0.3, the package is restricted to the following life data analysis right censoring weibull and exponential distribution standard errors estimated by Fisher Information method

Life Data Analysis

The lifedata() function will take the input times and censoring status to create a lifedata object. There are two required inputs: time-to-event and censoring status, respectively. Status should be a 0 for right censored, otherwise 1. An optional 3rd parameter for units of time values is available not used in any way. Example:

x = lifedata(c(90.7, 114.8, 12.0, 144.35, 199.8), c(0,1,1,1,0), 'hours')

To fit data to a distribution, use lifedata(). The result is a fitted_life_data object. This object constains the original data, the distribution fitted, the fitted values, the log-likelihood, standard errors, and goodness-of-fit values, if possible. The default method is MLE for parameter estimates. Standard errors estimated from Fisher matrix. Note: you may see warnings outputted by the optimization routine when fitting Weibulls. Examples:

y =  fit_data(x) #fit weibull distribution by default
y_exp=fit_data(x, dist = 'exponential') #fit exponential distribution
y =  fit_data(x, method = 'rry') #fit weibull distribution using rank regression on y instead of MLE

The package provides several plot types of the fitted distribution. These include: probability on linearized scales, CDF (unreliability), reliability, PDF, and failure rate (hazard function). The first input to the plot() function should be a fitted_life_data object. The second input is a string to denote the plot type. Linearized scale plots are specific to the fitted distribution and attempt to follow reliability conventions.

The first 3 plots include points representing the calculated median ranks of failure data. It is referred to as the "standard ranking method" in Weibull++.

Examples:

plot(y) #default to linearized scale plot
plot(y, type = 'failure') #probability of failure over time
plot(y, type = 'reliability') #reliability over time
plot(y, type = 'pdf') #pdf of fitted distribution

For styling purposes, an optional theme input can accept values of either 'base_r' or 'weibull++. The latter produces a plot that looks similar to those produced by Weibull++.

plot(y, theme = 'weibull++') #linearized scale plot with Weibull++ styling

Graphical parameters can be passed when theme = base_r. For plots that have both points and lines, graphical parameters are passed as separate lists. All other parameters can be passed as usual.

plot(y, line_par = list(lwd = 2, col = 'red'), point_par = list(pch = 16, col = 'blue', cex = 0.6), main = 'Weibull paper')

Histograms, pie charts, and timeline charts are also available. These functions can take either lifedata objects or fitted_life_data objects. NOTE: histogram functionality is in its early infancy and does not respect them input nor allow graphical parameters to be passed Examples:

hist(y, type = 'failure') #histogram of failure points, input is fitted_life_data_object
hist(y, type = 'suspension') #histogram of suspension points
pieplot(y)
pieplot(y, theme = 'weibull++')
timeline(y)
timeline(y, theme = 'weibull++')

Calculation of metrics on fitted data is done through the calculate(...) function. See example for available options.

fit = fit_data(x, dist = 'weibull') #input to calculate is the result of fit_data
calculate(fit, 'reliability', 100) #probability of surviving to t=100
calculate(fit, 'failure', 100) #probability of failure before t=100
calculate(fit, 'mean life', NA) #mean time to failure
calculate(fit, 'failure rate',100) #aka hazard function; instantaneous number of failures per unit time at t = 100
calculate(fit, 'reliable life', 0.75) #time when reliability is 75%
calculate(fit, 'bx life', 0.75) #time when prob. of failure is 75%
calculate(fit, 'cond reliab', 100, 50 ) #reliability within next 100 if survived to 50
calculate(fit, 'cond fail', 100, 50) #probability of failing within next 100 if survived to 50

Recurrent Event Analysis

This is a skeleton in what we hope the actual functionality will be.

At this time, only really set up for 1 unique ID. General Renewal Process (GRP) is the assumed parameteric model with power law model.

Input: times, ID, and status. Optional Input: time units. Status and time units input should be similar to Life Data Analysis.

x = recurrentdata(time, id, status, units)
plot(x)
y = mle_recurrent(x, type = 1)

Plotting data will result in a step plot of Number of occurences vs time. mle_recurrent will default to type I but also accepts type II model.

lxy009/weibullpp documentation built on May 21, 2019, 9:16 a.m.