Weibull distribution



Release:

4.6  •  5.0  •  5.1  •  5.2  •  5.3

The Weibull distribution is often used to represent failure time in reliability models. It is similar in shape to the gamma distribution, but tends to be less skewed and tail-heavy. It is a continuous distribution over the positive real numbers.

Weibull(10, 4) → Weibull graph.jpg

Functions

Both parameters most be positive, i.e., $ shape, scale > 0 $. The «scale» parameter is optional and defaults to 1.

Weibull(shape, scale, over)

The distribution function. Use this to define a chance variable or other uncertain quantity as having a Weibull distribution.

You can use the optional «over» parameter to generate independent and identically distributed distributions over one or more indicated indexes.

DensWeibull(x, shape, scale )

The density on $ x\ge 0 $ is given by:

$ p(x) = {{shape}\over{scale}} \left({x\over{scale}}\right)^{shape-1} \exp\left(-(x/{scale})^{shape}\right) $

CumWeibull(x, shape, scale )

The Weibull distribution has a cumulative density on $ x\ge 0 $ given by:

$ F(x) = 1 - \exp\left({-\left({x\over{scale}}\right)^{shape}}\right) $

and F(x) = 0 for x < 0.

CumWeibullInv(p, shape, scale )

The inverse cumulative distribution, or quantile function. Returns the «p»th fractile/quantile/percentile.

$ F^{-1}(p) = scale * \left( \ln\left( 1\over{1-p} \right)\right)^{1/shape} $

Statistics

The theoretical statistics (i.e., without sampling error) for the Weibull distribution are as follows. I use $ \alpha = 1/shape $ and $ \beta = scale $.

  • Mean = $ \beta \Gamma\left( 1 + \alpha\right) $
  • Mode = $ \left\{ \begin{array}{ll} \beta \left( 1 - \alpha \right)^\alpha & \alpha>1 \\ 0 & \alpha \leq 1 \end{array}\right. $
  • Median = $ \beta \left( \ln 2\right)^\alpha $
  • Variance = $ \beta^2 \Gamma(1+2\alpha) - \left(\beta \Gamma(1+\alpha)\right)^2 $

Parameter Estimation

Suppose you have sampled historic data in Data, indexed by I, and you want to find the parameters for the best-fit Weibull distribution. The parameters can be estimated using a linear regression as follows:

Index bm := ['b', 'm'];
Var Fx := (Rank(Data, I) - 0.5)/Size(I);
Var Z := Ln(-Ln(1 - Fx));
Var fit := Regression(Z, Array(bm, [1, Ln(Data)]), I, bm);
Var shape := fit[bm = 'm'];
Var b := fit[bm = 'b'];
Var scale := Exp(-b/shape);
[shape, scale]

See Also

Comments


You are not allowed to post comments.