Research Article

(wileyonlinelibrary.com) DOI: 10.1002/qre.1832 Published online 3 July 2015 in Wiley Online Library

Non-normal Capability Indices for the Weibull

and Lognormal Distributions
Manuel R. Piña-Monarrez*† Jesús F. Ortiz-Yañez and
Manuel I. Rodríguez-Borbón
Because the normal process capability indices (PCIs) Cp, Cpu, Cpl, and Cpk represent the times that the process standard
deviation is within the specification limits; then, based on and by using the direct relations among the parameters of
the Weibull, Gumbel (minimum extreme value type I) and lognormal distributions, the Weibull and lognormal PCIs
are derived in this paper. On the other hand, because the proposed PCIs Pp, Ppu, Ppl, and Ppk were derived as a function
of the mean and standard deviation of the analyzed process, they have the same practical meaning with those of the
normal distribution. Results show that the proposed PCIs could be used as the standard Cp, Cpu, Cpl, and Cpk if a
short-term variance is analyzed. An application to a set of simulated data is presented. Copyright © 2015 John Wiley
& Sons, Ltd.

Keywords: non-normal capability index; Weibull distribution; lognormal distribution; minimum extreme value distribution

1. Introduction
rocess capability indices (PCIs) were first formulated for normal behavior.1 However, because any capable and controlled

P process is subjected to noise factors and products are more complex and multifunctional, the process behavior generally is
not normal.2 In particular, we notice that, because of the effect of the noise factors, the process presents the well-known
underdispersion phenomenon3 (variance is lesser than the mean, implying variation coefficient less than one4). On the other hand,
although several methods to estimate the non-normal PCIs had been proposed (see among others the works of Tang and Than,5
Gunter,6 Clements,7 and Castagliola8), until now, there is not a general accepted formulation to estimate them.1 In particular, methods
proposed by the Weibull distribution can be found in the works of Chang et al.,9 Vännman,10 Albing,11 and Shauly12; unfortunately,
because they do not quantify the times that the observed variance fits within the specification limits as the proposed method does,
they are inefficient. In particular, until now, significant difference on the estimated performance PCI (Ppk) remains. Nevertheless,
because the Weibull and lognormal distributions are flexible to model different process behaviors (Figures 1 and 2), they have been
extensively used to model the non-normal processes.2
On the other hand, although the Weibull distribution is generated by a non-homogeneous Poisson process (additive effect)
(the work of Rinne,13 Section 4.3) and the lognormal distribution is generated by a geometric Brownian motion (multiplicative
effect) (the works of Luenberger,14 and Marathe and Ryan15) here, because there are direct relationships among the parameters
of the Weibull, Gumbel, and lognormal distributions, their long–term PCIs: process capability (Pp), upper process capability (Ppu),
lower process capability (Ppl), and process performance capability (Ppk), are derived from the logarithm of the data. Finally,
because of the natural relationship among the Weibull, the Gumbel, and the lognormal parameters, let us determine the PCIs
in such that they reflect the times that the process variance is within the specification limits, the PCIs could be seen as those
of the normal distribution. In fact, if a short-term data variance is analyzed, they could be seen as the well-known short-term
PCIs: process capability (Cp), upper process capability (Cpu), lower process capability (Cpl), and process performance capability
(Cpk) indices.
The paper structure is as follows: Section 2 presents the Weibull and Gumbel (minimum extreme value type I) relationships.
Section 3 gives the lognormal analysis. In Section 4, the generalities of the PCIs are presented. In Section 5, the Weibull and the
lognormal PCIs are derived. Section 6 gives the conclusions. The paper ends in Section 7 with the references.

Manufacturing and Industrial Engineering Department, Engineering and Technology Institute, Universidad Autónoma de Ciudad Juárez, Ciudad Juárez, Chihuahua
C.P. 32310, México
*Correspondence to: Manuel R. Piña-Monarrez, Manufacturing and Industrial Engineering Department, Engineering and Technology Institute, Universidad Autónoma
de Ciudad Juárez, Ciudad Juárez, Chihuahua C.P. 32310, México.
1321

†
E-mail: manuel.pina@uacj.mx

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
 M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

Figure 1. Weibull distribution behavior for η = 4600

Figure 2. Lognormal behavior

2. Weibull and Gumbel relationships

The two parameters Weibull distribution with scale parameter η and shape parameter β, defined first by Wallody in 1939,16 is
given by


β t β1 t β
f ðt Þ ¼ expðηÞ (1)
η η

The corresponding reliability function is given by


t β
RðtÞ ¼ exp  (2)
η
thus, the cumulative function is


t β
F ðt Þ ¼ 1  exp  (3)
η

and the corresponding risk function, which is a time-dependent function, is given by


β t β1
hðt Þ ¼ (4)
η η

On the other hand, in order to make a direct comparison between the PCIs of the Weibull and the normal distribution, let us first
1322

show how the logarithm of a Weibull variable follows a Gumbel distribution (minimum extreme value type I).13

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

Theorem
If a random variable t follows a Weibull distribution [t ~ W(β, η)], then its logarithm x = ln(t) follows a Gumbel distribution [x ~ G(μEV,σ EV)].

Proof
Let F(ln(t)) = P(ln(t) ≤ ln(T)) be the cumulative function of x = ln(t), with T representing the failure time value. In terms of x, F(ln(t)) =
Pr[ln(t) ≤ x]; F(x) = Pr[t ≤ exp(x)]. Thus, by substituting t = exp(x), in Equation (3), F(x) is
"
 #
expðx Þ β
F ðx Þ ¼ 1  exp  (5)
η

Finally, based on the relations between the Weibull and Gumbel parameters (the work of Crowder et al.,17 Section 2.4, pp 17) given by
μEV ¼ lnðηÞ (6)
1
σ EV ¼ (7)
β

and by taking W = ((x  μEV)/σ EV), Equation (5) is given by F(x) = 1  exp[exp{(x  ln(η)) · β}], which in terms of W is given by

F ðx Þ ¼ 1  expexp
w
(8)

from Equation (8), the reliability function is

RðtÞ ¼ 1  F ðtÞ ¼ expexp

w
(9)
and the density function is given by
dRðt Þ  
¼  expexp expw ¼ expwexp
w w
f ðx Þ ¼  (10)
dt
clearly, Equation (10) in terms of W is
 
  xμ
EV
xμEV σ EV
1 σ EV exp
f ðx Þ ¼ exp (11)
σ EV

Equations (10) or (11) are known as the Gumbel distribution (minimum extreme value type I) with hazard function given by


1 x  μEV
hðx Þ ¼ exp (12)
σ EV σ EV

On the other hand, by using the moment method (by NIST/SEMATECH18 Section 1.3.6.6.16), the parameters of Equations (10) or
(11) are given by
μEV ¼ E ðx Þ ¼ μY þ γσ EV (13)
pffiffiffi
6
σ EV ¼ σY (14)
π
where μY and σ Y are the mean and standard deviation of the logarithm of the failure times. Finally, from Equations (13) and (14) (the
work of McCool,19 Section 6.7), an estimation of μY and σ Y is given by
^ Y ¼ μEV  γσ EV
μ (15)
π
σ^ Y ¼ pffiffiffi σ EV (16)
6

In Equation (15), γ is the Euler’s constant (γ = 0.577216…).

On the other hand, the parameters defined in Equations (15) and (16), based on Equations (6) and (7), can be seen as a direct
estimation of the lognormal parameters based on the Weibull or Gumbel distribution. The lognormal distribution is as follows.

3. Lognormal distribution
The lognormal distribution is based on a Brownian motion (BM) process. A standard BM is a type of Markov stochastic process with a
1323

drift rate of zero and a variance of one. The drift rate of zero means that the expected value at any future time is equal to the current
value, and the variance of one means that the variance of the change in a time interval of length T is equal to T.15 Because BM could

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
 M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

be negative, then the lognormal distribution is based on a variant of BM, which is non-negative and known as geometric Brownian
motion. A geometric Brownian motion is given by
Sðt Þ ¼ S0 expX ðtÞ (17)

where X(t) = σB(t) + μt is a BM and S(0) = S0 > 0 is the initial value. Taking logarithms yields back the BM; X(t) = ln(S(t)/S0) = ln(S(t))  ln(S0).
Thus, ln(S(t)) = ln(S0) + X(t) is normal with mean μt + ln(S0), and variance σ 2t. Thus, for each t, S(t) has a lognormal distribution.
On the other hand, if Y ~ N(μ,σ 2), then
X ¼ expY (18)
2
is a non-negative random variable having a lognormal distribution. Thus, because Y ~ N(μ,σ ), then Y = ln(X) follows a normal
distribution. The lognormal density function is given by

 !
1 1 lnðx Þ  μY 2
f ðx Þ ¼ pffiffiffiffiffiffi exp  (19)
xσ Y 2π 2 σY

where μY is the mean of the logarithms and σ Y is the shape parameter. Equation (19), as was shown in Figure 2, is a flexible and time-
dependent distribution with an increasing risk function. However, Y = ln(X) follows a normal distribution given by

 !
1 1 Y  μY 2
f ðx Þ ¼ pffiffiffiffiffiffi exp  (20)
σ Y 2π 2 σY

On the other hand, based on Equation (20), the reliability function is given by
∞
Rðx Þ ¼ ∫z ΦðZ Þdz (21)

the cumulative distribution is given by


lnðx Þ  μY
F ðx Þ ¼ 1  Rðx Þ ¼ Φ (22)
σY
and the corresponding risk function is given by
f ðx Þ ΦðZ Þ
hðx Þ ¼ ¼ (23)
Rðx Þ xσ Y Rðx Þ

In Equations (21), (22), and (23), Φ stands for the normal distribution. Finally, based on the observed data, the lognormal
parameters are estimated as
1 Xn
μY ¼ lnðx i Þ (24)
n i¼1

1 Xn
σ 2Y ¼ ðlnðx i Þ  μY Þ2 (25)
n i¼1

4. Generalities of the process capabilities indices

Process capability is referred as the inherent ability of a process to produce homogeneous parts for a sustained time period under the
given quality conditions. The most commonly used PCIs, which measure the potential process performance (process consistency in the
short term), as the times that the process standard deviation is within the tolerance specification, are the Cp given by Juran20 and the
Cpk given by Kane,21 with Cpk = min(Cpu, Cpl), where U is the upper specification limit and L is the lower specification limit. Because the Cp
index is related to the upper (USL) and lower (LSL) specification limits, the location (mean) is not considered. The Cp is given by
ðUSL  LSLÞ
Cp ¼ (26)
6σ
where σ is the population standard deviation, which when it is unknown, it is estimated by the sample standard deviation (σ̂ ). On the
other hand, the expected proportion of nonconforming products by taking the population mean, as
ðUSL þ LSLÞ
μ¼ (27)
2
(note that μ is symmetrical), is estimated by E(p) = Pr(X < LSL) + Pr(X > USL), which in terms of the normal distribution is given by



LSL  μ USL  μ
E ðpÞ ¼ Pr Z < þ Pr Z > (28)
σ σ
1324

On the other hand, the Cpk index, which is more sensitive to departures from normality than Equation (26), for normal behavior is
given by

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN



USL  μ
^ μ ^  LSL
C pk ¼ min C pu; C pl ¼ min ; (29)
3^σ σ
3^

In Equation (29), μ̂ is the sample mean, and σ̂ is the sample standard deviation of the analyzed process.
On the other hand, the long term PCIs, Pp, Ppu, Ppl, and Ppk, represent the overall process variation and are used when data are not
normal. Several methods have been proposed to estimate them. Among others, we have Clement’s and the graphical method. The
graphical method is the most used method22 and is based on the fact that for normal behavior ±3σ(0.998650102, 0.001349898) =
95%. Then the Pp, analogous to Equation (26), is given by
USL  LSL
Pp ¼ (30)
x0:99865  x0:00135

where x0.99865 and x0.00135 are the value of the 0.9986 and 0.00135 quantiles of the analyzed distribution. The corresponding Ppk
analogous to Equation (29) is given by

Ppk ¼ min Ppu ; Ppl (31)

and because the median is the preferable central value for skewed distributions, the equivalent Cpu and Cpl are given by


USL  Median
Ppu ¼ (31a)
x0:00135  Median


Median  LSL
Ppl ¼ (31b)
Median  x0:99865

Finally, although these non-normal PCIs are used in practice, because their values depend on the used method and because they
could not be interpreted as the times that the process standard deviation is within the specification limits, their application is
suboptimal. Thus, here, based on the direct relationships between the Weibull, Gumbel (minimum type I), and lognormal
distributions, the long-term PCIs Pp, Ppu, Ppl and Ppk for the Weibull and lognormal distributions are proposed as in the next section.

5. Weibull and lognormal process capability indices

Because both the Gumbel and lognormal distributions use the logarithm of the observed data, then based on the Weibull and
Gumbel relationships discussed in Section 2, the Weibull and lognormal PCIs are derived as in the next sections.

5.1. Weibull process capabilities indices

Let us begin this section by mentioning that the behavior of the Gumbel distribution defined in Equation (11) and represented here in
Figure 3 is similar to that of the normal distribution defined in Equation (20) and represented here in Figure 4.
From Figures 3 and 4, we note that both distributions have a Gaussian form and they only have position and scale parameters. On
the other hand, from Equations (15), (16), (24), and (25), we observe that μY and σ Y are the mean and standard deviation of the

1325

Figure 3. Gumbel distribution with constant μ

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
 M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

Figure 4. Normal distribution

logarithm of the analyzed process, thus, although in logarithm metric, they are analogous to those of the normal distribution. Based
on this fact, the proposed Weibull capability index Cp defined in Equation (26), by using (7) and (16), is given by

½lnðUSLÞ  lnðLSLÞ β½lnðUSLÞ  lnðLSLÞ

C pW ¼ pffiffiffi ¼ pffiffiffi (32)
6πσ EV 6π

Note, in Equation (32), that π/√6 = 1.28255 was used in order to make the denominator reflect the times that the logarithm of the
standard deviation is within the rank of the numerator, as it is made in the normal case. The analogous Cpk index, in the proposed
Weibull capability index, is given by

pffiffiffi pffiffiffi 

β 6½lnðUSLÞ  μ
^ Y  β 6½ μ^ Y  lnðLSLÞ
C pkW ¼ min C pu ; C pl ¼ min ; (33)
3π 3π

In Equation (33), μ̂Y is as it was defined in Equation (15). Observe that in this Weibull case, the population mean defined in Equation
(27) is given by
½lnðUSLÞ þ lnðLSLÞ
μY ¼ (34)
2

5.1.1. An application. Suppose that for a Weibull process behavior, the customer’s specification limits are 50 ± 30 (USL = 80 and LSL = 20);
thus, according to Equation (34), the mean to monitor the process (in logarithm scale) is μY = [ln(80) + ln(20)]/2 = 3.688879, implying a
mean of μ = 40. Note that with this requirement, in order to have a Pp = 1, from Equation (27) or (32), the process standard deviation of
the logarithms must be lesser or equal to σ̂Y = [ln(80)  ln(20)]/6 = 0.231049; this implies from Equations (7) and (16) that the shape
parameter of the process is β = 5.550985 and from Equations (6) and (15) that the corresponding scale parameter is η = 40.104. It is to
say that the process must be monitored with mean μ = exp(3.688879) = 40 and standard deviation of σ = exp(0.231049) = 1.259921 or
directly in logarithm scale with μY = 3.688879 and σ Y = 0.231049 (with these logarithm data, the normal statistical process control could
be applied directly). Finally, observe that by applying with this data Equations (32) and (33), the estimated PCIs are equal to 1
(Pp = Ppu = Ppl = Ppk = 1) as expected.
On the other hand, in order to compare the results with those of the graphical method defined in Equations (30), (31), (31a), (31b),
by solving Equation (2) for t, we obtain x0.99865 = 12.16916, x0.00135 = 56.22156, and x0.5 = 37.45397. Thus, Pp = 1.358, Ppu = 2.257,
Ppl = 0.692, and Ppk = 0.692. Because Minitab routine has incorporated Equations (30), (31), (31a), (31b), it was used to generate a
sample of n = 20. A routine result and the simulated Weibull parameters are shown in Figure 5.
From Figure 5, we observe that the use of the median in Equation (31a) and (31b) biases the Ppk index. Clearly, the conclusion is
that the lower the β value be, the greater the difference between μY and the median, and as a consequence, the bias on the estimated
Ppk value is increased. To see this numerically, suppose that the customer requirement were 5000.5 ± 4999.5 (exponential behavior;
the variance equals the mean), (USL = 10,000 and LSL = 1); thus, according to Equations (34) and (27), the process parameters to
monitor the process would be μY = [ln(10,000) + ln(1)]/2 = 4.60517 (or μ = 100), with σ Y = [ln(10,000)  ln(1)]/6 = 1.535057 (or σ =
4.641589). Thus, by applying Equations (7) and (16), β = 0.835506, and by using Equations (6) and (15), η = 100.6909. With these
Weibull parameters, by solving Equations (2) to 0.5 (R(t) = 0.5), the median is x0.5 = 64.9348, which presents significant difference with
the estimated μ = 100.
On the other hand, for these Weibull parameters, x0.99865 = 0.03704, and x0.00135 = 964.91; thus, the corresponding PCIs given by
1326

Equations (30), (31), (31a), (31b) are Pp = 10.363, Ppu = 11.039, Ppl = 0.985, and Ppk = 0.985. The corresponding Minitab routing for these
PCIs is given in Figure 6.

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

Figure 5. Minitab routine for proposed method. LSL, lower specification limits; USL, upper specification limits

Figure 6. Minitab routine for β > 1. LSL, lower specification limits; USL, upper specification limits

Finally, observe that in the proposed method, by applying Equations (32) and (33), the estimated PCIs are equal to 1
(Pp = Ppu = Ppl = Ppk = 1) as expected. As a conclusion, we have that in order to correct the bias generated by the skewness of the Weibull
behavior, the mean of the logarithm of the analyzed data must be used in the PCIs, as it was proposed in Equations (32) and (33).

5.2. Lognormal process capabilities indices

Because from Equation (18), Y = ln(X) is normal, Y ~ N(μ,σ 2); then by using Equations (24) and (25), the proposed PCIs are

½lnðUSLÞ  lnðLSLÞ
PpL ¼ (35)
6σ Y

0h  i 1
ln USL  μy ½ μ  ln ðLSL Þ
C pkL ¼ min@ ; Y A (36)
3σ Y 3σ Y

where μY and σ Y are the mean and standard deviation of the logarithm of the observed data defined in Equations (24) and (25).

5.2.1. An application. Suppose that for a lognormal process behavior, the customer’s specification limits are 50 ± 30 (USL = 80 and
LSL = 20); thus, according to Equation (34), the mean of monitoring the process (in logarithm scale) is μY = [ln(80) + ln(20)]/2 = 3.688879.
1327

The observed mean is μ = 40. Note that with this requirement, in order to have a Pp = 1, from Equation (35), the process standard deviation

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
 M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

Figure 7. Process capability indices for lognormal analysis. LSL, lower specification limits; USL, upper specification limits

of the logarithms must be lesser or equal to σ̂ Y = [ln(80) + ln(20)]/6 = 0.231049. By using these data in Equation (35), the Pp is
PpL = [ln(80)  ln(20)]/6*0.231049 = 1.1486. Because Ppu = 1.2442 and Ppl = 1.053, then the corresponding Ppk is Ppk = 1.053.
On the other hand, by simulating n = 20 lognormal data with μY = 3.63119 and σ̂ Y = 0.201154 and by applying Equations (30), (31),
(31a), (31b) with x0.99865 = 69.038, x0.00135 = 20.65, and x0.5 = 37.758, the PCIs are Pp = 1.24, Ppu = 1.35, Ppl = 1.04, and Ppk = 1.04.
These PCIs using Minitab routine are shown in Figure 7.
By comparing the estimated values of Figure 7 with those of the proposed method, we observe that the use of the x0.99865 and
x0.00135 percentiles, instead of the corresponding logarithm values, biases the PCIs; thus, the conclusion is that it is better to estimate
the PCIs as they were proposed in Equations (35) and (36).

6. Conclusions
Determining the PCIs by using the logarithm of the analyzed data, as it is proposed here, corrects by itself the bias generated by the
skewness of the Weibull or lognormal distributions. And because the proposed PCIs could be seen analogous to those of the normal
distribution, they represent in logarithm scale, the times that the logarithm deviation of the process fits within the customer’s
specification limits. In particular, it is important to note that the proposed PCIs could be used by practitioners for short-term and
long-term deviation, as the Cp, Cpk, Pp, and Ppk for the normal case are used. Specially, we remark that in the Weibull case, the mean
of the logarithm of the data, as it is carried out in Equations (32) and (33), should be equivalent to the use of the logarithm of the scale
parameter η. It is to say that, from the analysis, it seems to be possible to use ln(η) = μY, which implies that β = 1/σ Y, as a consequence;
Equations (35) and (36) could be used directly for both distributions, but much research must be performed.
Observe that in practice, the arithmetic mean of the customer is neither used in the analysis nor to be monitored. For example, in
the given specification of 50 ± 30, with USL = 80 and LSL = 20, the value to be monitored in the process is μ = 40 instead of the
arithmetic mean of 50. The same occurs with the standard deviation; thus, we must monitor the logarithm of the standard deviation
(σ Y = 0.201154) instead of the arithmetic standard deviation.

References
1. Safdar S, Ahmed E. Process capability indices for shape parameter of Weibull distribution. Open journal of Statistics 2014; 4:207–219. doi:10.4236/
ojs.2014.43020.
2. Liu P-H, Chen F-L. Process capability analysis of non-normal process data using the Burr XII distribution. International Journal of Advance
Manufacturing Technology 2006; 27:975–984. doi:10.1007/s00170-004-2263-8.
3. Blake M, Moshe A, Bradlow ET, Fader P. Count models based on Weibull interarrival times. Journal of Business & Economic Statistics 2008;
26:369–378. doi:10.1198/073500107000000278.
4. Winkelmann R. Duration dependence and dispersion in count-data models. Journal of Business & Economic Statistics 1995; 13:467–474. doi:10.1080/
07350015.1995.10524620.
5. Tang LC, Than SE. Computing process capability indices for non-normal data: a review and comparative study. Quality and Reliability Engineering
International 1999; 15:339–353. doi:10.1002/(SICI)1099-1638(199909/10)15:5<339::AID-QRE259>3.0.CO;2-A.
6. Gunter BH. The use and abuse of Cpk. Quality Progress 1994; 22:108–120.
7. Clements JA. Process capability calculations for non-normal distributions. Quality Progress 1989; 22:95–100.
8. Castagliola P. Evaluation of Non-normal process capability indices using Burr’s distributions. Quality Engineering 1996; 8:587–593. doi:10.1080/
08982119608904669.
1328

9. Chang YS, Choi IS, Bai DS. Process capability index for skewed populations. Quality and Reliability Engineering International 2002; 18:383–393. Wiley
and sons, ltd. DOI: 10.1002/qre.489

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329
M. R. PIÑA-MONARREZ, J. F. ORTIZ-YAÑEZ AND M. I. RODRÍGUEZ-BORBÓN

10. Vännman K, Albing M. Process capability indices for one-sided specification intervals and skewed distributions. Quality and Reliability Engineering
International 2007; 23:755–765. Wiley and sons, ltd. DOI: 10.1002/qre.877
11. Albing M. Process capability indices for Weibull distributions and upper specification limits. Quality and Reliability Engineering International 2009;
25:317–334. Wiley and sons, ltd. DOI: 10.1002/qre.972
12. Shauly M, Parmet Y. Comparison of Pearson distribution system and response modeling methodology (RMM) as models for process capability
analysis of skewed data. Quality and Reliability Engineering International 2011; 27:681–687. Wiley and sons, ltd. DOI: 10.1002/qre.1232
13. Rinne H. The Weibull Distribution: A Handbook. Taylor & Francis Group: Boca Raton, FL, 2009.
14. Luenberger D. Investment Science. Oxford University Press: New York, 1995.
15. Marathe RR, Ryan SM. On the validity of the geometric Brownian motion assumption. The Engineering Economist 2005; 50:159–192. doi:10.1080/
00137910590949904.
16. Weibull W. A Statistical Theory of the Strength of Materials. Swedish Royal Institute for Engineering Research: Stockholm, 1939.
17. Crowder MJ, Kimber A, Sweeting T, Smith R. Statistical Analysis of Reliability Data. CRC Press: Boca Raton, Florida, 1994.
18. NIST/SEMATECH e-handbook of statistical methods, http://www.itl.nist.gov/div898/handbook/eda/section3/eda366g.htm [November 06 2014].
19. McCool JI. Using the Weibull Distribution: Reliability, Modeling and Inference. Wiley Series in Probability and Statistics: USA, 2012.
20. Juran JM. Juran’s Quality Control Handbook (3rd edn). McGraw-Hill: New York, 1974.
21. Kane VE. Process capability indices. Journal of Quality Technology 1986; 18:41–52.
22. Bissell D. Statistical Methods for SPC and TQM. Chapman and Hall: London, 1994.

Authors' biographies
Manuel R. Piña-Monarrez is a researcher professor of the postgraduate department at Universidad Autónoma de Ciudad Juárez,
Chihuahua, México. He completed his PhD degree in Science in Industrial Engineering in 2006 at Instituto Tecnológico de Ciudad
Juarez, Chihuahua, México. He had conducted research on robust design, multiple linear regression, reliability, and multivariate
process control. He is a member of the National Research System of the National Council on Science and Technology.
Jesús F. Ortiz-Yañez is a PhD student on the Science in Engineering Doctoral Program (DOCI), at Universidad Autónoma de Ciudad
Juárez, Chihuahua, México. He completed his master’s degree in Industrial Engineering in 2012. His PhD research is on reliability with
focus on the Weibull and lognormal comparisons.
Manuel I. Rodríguez-Borbón is a researcher professor of the postgraduate department at Universidad Autónoma de Ciudad Juárez,
Chihuahua, México. He completed his PhD degree in Doctor of Philosophy in 2011 at New Mexico State University, Las Cruces, NM,
USA (EUA). He focuses his researches on reliability and quality.

1329

Copyright © 2015 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016, 32 1321–1329