Professional Documents
Culture Documents
1. Introduction
he monitoring of reliability data from life-test and field return is often modeled using a Weibull distribution. Recently, Woodall
T and Montgomery1 suggested that this area has potential for further research. Nelson2 presented the median chart, range chart,
location chart, and scale chart simultaneously to monitor a Weibull process. Bai and Choi3 used the weighted variance method
to construct the X and R control charts for skewed distributions such as exponential and Weibull. Hawkins and Olwell4 provided the
optimal design of cumulative sum (CUSUM) control chart for monitoring the scale parameter Weibull data with fixed shape
parameter. Ramalhoto and Moriais5 studied the Shewhart control chart for monitoring the scale parameter of a Weibull process with
fixed and variable sampling intervals. Chang and Bai6 used the weighted standard deviations to construct the X , CUSUM and
exponentially weighted moving average (EWMA) control charts for skewed distributions. Xie et al.7 developed a t-chart for monitoring
Weibull distributed time between failures based on probability limit method. Furthermore, a tr-chart was proposed to monitor time
between r failures. Here, the Erlang distribution was used to model the time until the occurrence of r failures in a Poisson process.
Borror et al.8 investigated the robustness of CUSUM control chart for a Weibull process. Chan and Cui9 proposed the skewness
correction method to construct the X and R control charts for skewed distributions. Nichols and Padgett10 used a bootstrap method
with pivotal quantities to monitor Weibull percentiles.
Batson et al.11 studied the individual and moving range (I-MR) control charts using the power transformation method to transform
the Weibull data to a normal distribution. Chen and Cheng12 investigated the effect of non-normality on the control limits of the X
chart. Pascual13 presented the EWMA control chart for monitoring the Weibull shape parameter. Pascual and Zhang14 proposed
the MR control chart by taking the natural logarithm of the Weibull distribution and then setting control limits based on the range
value of random samples from the resulting smallest extreme population for monitoring the Weibull shape parameter. Pascual and
Nguyen15 provided the average run length (ARL) of the MR control chart based on single-point samples from a smallest extreme value
distribution for a Weibull process. Pascual16 investigated the run length properties of the individuals and moving ratio control charts
for monitoring a Weibull process. Pascual and Li17 studied the MR chart to monitor the Weibull shape parameter under type II
censoring. Faraz et al.18 presented Shewhart-type Z and S2 control charts for monitoring the scale and shape parameters of a Weibull
process by taking a method to transform the Weibull to a normal distribution.
The Shewhart-type control charts for rational subgroups have used some methods to transform the Weibull data to a standard
normal distribution and subsequently monitor the process and the individuals and MR control charts for individual measurements.
It is needed to compare their performances. In addition, Chen et al.19 presented a MaxEWMA control chart to simultaneously monitor
Department of Industrial Management, National Taiwan University of Science and Technology, Taipei, Taiwan
*Correspondence to: Fu-Kwun Wang, Department of Industrial Management, National Taiwan University of Science and Technology, Taipei, Taiwan.
†
E-mail: fukwun@mail.ntust.edu.tw
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
the process mean and variability for rational subgroups. We will extend this chart to monitor a Weibull process with individual
measurements.
In this paper, we present the comparison study of the three different transformation methods and discuss their possible
applications in monitoring a Weibull process with individual measurements. Additionally, we compare the average run lengths
of the combined I-MR control charts20 and the MaxEWMA control chart for a Weibull process with individual measurements.
The rest of the paper is organized as follows. In the next section, we review the Weibull distribution and its properties. Three
methods to transform the Weibull data to a normal distribution are described. A simulation study is conducted to compare
the transformation performance in terms of bias and mean square error (MSE). Section 4 presents a MaxEWMA control chart
for a Weibull process with individual measurements. Two illustrative examples are given in Section 5, and concluding remarks
are provided in the last section.
2. Three transformation methods with their control limits for individuals and moving range
charts
The probability density function of the two-parameter Weibull distribution is given by the following:
βx β1 ðθx Þβ
f ðx; θ; βÞ ¼ e (1)
θβ
where x > 0, θ > 0, and β > 0. The Weibull distribution reduces to an exponential distribution if β = 1 and a Rayleigh distribution if
β1
β = 2. Furthermore, the hazard function hðx Þ ¼ βxθβ is constant if β = 1, decreasing if β < 1, and increasing if β > 1.
Table I. Bias and MSE of different methods under θ = 50 and β = 0.5, 0.75, 1.0, 1.5, and 2.0
μ σ
n θ β Method-1 Method-2 Method-3 Method-1 Method-2 Method-3
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
Some methods such as the simple power transformation called method-1, the inverse erf transformation called method-2, and the
Box–Cox transformation called method-3 are used to transform the Weibull data to a normal distribution. It should be noted that the
inverse erf transformation method is based on the conversion of a normal distributed variable into a uniform one and then the
conversion of this one into a Weibull one.18
Method 1. Batson et al.11 studied the individuals and MR control charts using the power transformation method for a Weibull
hprocess. That is, y1 = xi
0.2777β
and y 1 eN μ1;0 ; σ 21;0 , where μ1,0 = E(y1) = Γ(1.2777)θ0.2777β, σ 21;0 ¼ Var ðy 1 Þ ¼
Γð1:5554Þ Γ ð1:2777Þ2 θ0:5554β , and Γ() = gamma function. The mean of MR1,i = |y1,i y1,i 1| is obtained as MR 1 ¼
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ½Γð1:5554ÞΓð1:2777Þ2 θ0:2777β
pffiffi ¼ 0:315563θ0:2777β .
π
The limits of the individuals and MR control charts11 are established as follows:
8
> 3 3
>
> LCLY 1 ¼ Y 1 MR 1 ¼ Γð1:2777Þθ
0:2777β
0:09724773Γð0:2777Þθ0:2777β ¼ 0:061856θ0:2777β
< d2 1:128
CLY 1 ¼ Γð1:2777Þθ0:2777β ¼ 0:901119θ0:2777β (2)
>
> 3 3
>
: UCLY 1 ¼ Y 1 þ MR 1 ¼ Γð1:2777Þθ0:2777β þ 0:09724773Γð0:2777Þθ0:2777β ¼ 1:740382θ0:2777β
d2 1:128
and
CLMR1 ¼ MR 1 ¼ 0:315563θ0:2777β (3)
UCLMR1 ¼ MR 1 D4 ¼ 0:315563θ0:2777β 3:267 ¼ 1:030944θ0:2777β
Table II. Bias and MSE of different methods under θ = 100 and β = 0.5, 0.75, 1.0, 1.5, and 2.0
μ σ
n θ β Method-1 Method-2 Method-3 Method-1 Method-2 Method-3
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
Method 2. Faraz et al.18 converted random Weibull observations to observations with a standard normal distribution, N(0,1). That
pffiffiffi h i z
is, y 2 ¼ 2erf1 1 2eðx=θÞ and y2 ~ N(0, 1), where erf ðz Þ ¼ p1ffiffiπ ∫z eu du. They then constructed the Shewhart-type
β 2
Z and S2 control charts for rational subgroups. When individual measurements are collected, the individuals and MR
control charts are needed. The mean of MR2,i = |y2,i y2,i 1| can be easily obtained as MR 2 ¼ p2ffiffiπ.
The limits of the individuals and MR control charts are established as follows:
8
> 3 3 2
>
> LCLY 2 ¼ Y 2 d MR 2 ¼ 0 1:128pffiffiffi ¼ 3:001008
< 2 π
CLY 2 ¼ Y 2 ¼ 0 (4)
>
> 3 3 2
>
: UCLY 2 ¼ Y 2 þ MR 2 ¼ 0 þ pffiffiffi ¼ 3:001008
d2 1:128 π
and
8 2
>
< CLMR2 ¼ MR 2 ¼ pffiffiffi ¼ 1:128379
π (5)
> 2
: UCLMR2 ¼ MR 2 D4 ¼ pffiffiffi3:267 ¼ 3:686415
π
Method 3. Hernandez and Johnson21 showed that the Weibull distribution using the Box–Cox transformation method is an
0:2654β
1 0:2654β
1
approximate normal distribution. That is, y 3 ¼ x0:2654β and y 3 eN μ3;0 ; σ 23;0 , where μ3;0 ¼ E ðy 3 Þ ¼ 0:9034θ
0:2654β and
Table III. Comparison of the average run length of the MaxEWMA control chart and the combined I-MR charts for mean changes
under ARL0 = 250 and 500
ARL0 = 250
Mean
changes Combined I-MR MaxEWMA λ = 1.0, MaxEWMA λ = 0.3, MaxEWMA λ = 0.2, MaxEWMA λ = 0.1,
(μ1–μ0) M = 2.88, R = 5.5 UCL = 3.09 UCL = 1.2981 UCL = 1.0300 UCL = 0.7089
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
2 0:2654β
21:008θ β
σ 23;0 ¼ Var ðy 3 Þ ¼ 1:008θ0:2654β
. Moreover, the mean of MR3,i = |y3,i y3,i 1| can be obtained as MR 3 ¼ pffiffi ¼
β π
0:2654β
1:137406θ
β . Thus, the limits of the individuals and MR control charts can be established as follows:
8
>
> 3 0:9034θ0:2654β 1 3 1:137406θ0:2654β 0:3789027θ0:2654β 3:767898
>
> LCLY 3 ¼ Y 3 MR 3 ¼ ¼
>
> d2 0:2654β 1:128 β β
<
0:9034θ 0:2654β
1
CLY 3 ¼ Y 3 ¼ (6)
>
> 0:2654β
>
>
>
> 3 0:9034θ0:2654β 1 3 1:137406θ0:2654β 6:428935θ0:2654β 3:767898
: UCLY 3 ¼ Y 3 þ MR 3 ¼ þ ¼
d2 0:2654β 1:128 β β
and
8
>
> 1:137406θ0:2654β
< CLMR3 ¼ MR 3 ¼
β (7)
>
> UCL 1:137406θ0:2654β 3:715905θ0:2654β
: MR3 ¼ MR 3 D 4 ¼ 3:267 ¼
β β
3. Comparison study
The samples were generated from the Weibull distribution with different values of the parameters theta and beta with three different
sample sizes. To evaluate the performance of these transformation methods, the bias and MSE were calculated using 10,000
Table IV. Comparison of the average run length of the MaxEWMA control chart and the combined I-MR charts for sigma changes
under ARL0 = 250 and 500
ARL0 = 250
Sigma changes Combined I-MR MaxEWMA λ = 1.0, MaxEWMA λ = 0.3, MaxEWMA λ = 0.2, MaxEWMA λ = 0.1,
(σ1/σ0) M = 3.8, R = 4.05 UCL = 3.09 UCL = 1.2981 UCL = 1.0300 UCL = 0.7089
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
replications for the parameter values(θ, β) = (100,0.5), (100,0.75), (100,1.0), (100,1.5), (100,2.0), (50,0.5), (50,0.75), (50,1.0), (50,1.5), and
(50,2.0) and the sample sizes n = 25, 50, and 100. All simulation programs are written by R language.22
Simulation results are given in Tables I, II. The findings are as follows:
1) For μ under β = 0.5, 0.75, and 1.0 with θ =50 and 100 for the sample sizes n = 25, 50, and 100, method-1 has lower bias and MSE
than method-2 and method-3. For μ under β = 1.5 and 2.0 with θ = 50 and 100 for the sample sizes n = 25, 50, and 100, method-
2 has lower bias and MSE than method-1 and method-3.
2) For σ under β = 0.5, 0.75, 1.0, 1.5, and 2.0 with θ = 50 and 100 for the sample sizes n = 25, 50, and 100, method-2 has the lowest
bias except the cases of β = 1.0 and 2.0 with θ = 50. For σ under β = 0.5, 0.75, 1.0, 1.5, and 2.0 with θ = 50 and 100 for the sample
sizes n = 25, 50, and 100, method-2 has lowest MSE except the cases of β = 0.5 and 0.75 with θ = 50 and 100.
3) The sample size affected the bias and MSE in all cases. When the sample size is increased, the bias and MSE of all three methods
are decreased.
Based on the previous findings, method-1 is recommended forβ < 1, and method-2 is recommended forβ ≥ 1 to transform the
Weibull data to a normal distribution for monitoring a Weibull process with individual measurements.
Table V. Comparison of the average run length of the MaxEWMA control chart and the combined I-MR charts for mean and sigma
changes under ARL0 = 250 and 500
ARL0 = 250
Mean/sigma Combined I-MR MaxEWMA λ = 1.0, MaxEWMA λ = 0.3, MaxEWMA λ = 0.2, MaxEWMA λ = 0.1,
changes M = 3.10, R = 4.22 UCL = 3.09 UCL = 1.2981 UCL = 1.0300 UCL = 0.7089
0.0/1.0 250.777 250.055 250.136 250.055 250.075
0.5/1.5 17.300 10.165 5.034 3.702 2.127
0.5/2.0 6.481 3.400 2.946 2.512 1.820
0.5/3.0 2.980 1.910 1.771 1.690 1.506
0.5/4.0 2.153 1.449 1.412 1.388 1.327
1.0/1.5 16.795 6.146 1.869 1.378 1.041
1.0/2.0 6.257 3.241 1.678 1.364 1.074
1.0/3.0 2.922 1.817 1.455 1.317 1.122
1.0/4.0 2.131 1.425 1.307 1.247 1.134
2.0/1.5 14.408 2.419 1.018 1.001 1.000
2.0/2.0 5.408 1.970 1.042 1.005 1.000
2.0/3.0 2.707 1.554 1.088 1.028 1.001
2.0/4.0 2.047 1.344 1.109 1.052 1.006
3.0/1.5 10.123 1.378 1.000 1.000 1.000
3.0/2.0 4.207 1.364 1.000 1.000 1.000
3.0/3.0 2.397 1.317 1.008 1.001 1.000
3.0/4.0 1.919 1.247 1.023 1.005 1.000
ARL0 = 500
Mean/sigma Combined I-MR MaxEWMA λ = 1.0, MaxEWMA λ = 0.3, MaxEWMA λ = 0.2, MaxEWMA λ = 0.1,
changes M = 3.3, R = 4.53 UCL = 3.2904 UCL = 1.3823 UCL = 1.0968 UCL = 0.7549
0.0/1.0 516.467 500.025 500.260 500.025 500.261
0.5/1.5 24.051 13.704 6.348 4.538 2.460
0.5/2.0 7.870 4.789 3.409 2.859 2.003
0.5/3.0 3.292 2.080 1.911 1.812 1.594
0.5/4.0 2.298 1.522 1.479 1.450 1.378
1.0/1.5 23.589 7.891 2.127 1.498 1.060
1.0/2.0 7.622 3.789 1.829 1.450 1.096
1.0/3.0 3.228 1.967 1.534 1.372 1.145
1.0/4.0 2.273 1.493 1.355 1.285 1.155
2.0/1.5 21.055 2.841 1.027 1.001 1.000
2.0/2.0 6.660 2.186 1.056 1.008 1.000
2.0/3.0 2.989 1.650 1.105 1.034 1.001
2.0/4.0 2.180 1.398 1.126 1.061 1.007
3.0/1.5 15.561 1.498 1.000 1.000 1.000
3.0/2.0 5.235 1.450 1.001 1.000 1.000
3.0/3.0 2.642 1.372 1.010 1.001 1.000
3.0/4.0 2.040 1.285 1.027 1.006 1.000
MaxEWMA, maximum exponentially weighted moving average; UCL, upper control limit; I-MR, individual and moving range.
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
5. Illustrative examples
Two examples are used to illustrate the applicability of the proposed MaxEWMA control chart.
Example 1 Padgett and Spurrier24 presented Shewhart-type charts for percentile distribution on the strengths of the carbon fibers.
The dataset consisted of samples of size 5 from 20 inspection periods. It was known that the first 10 periods were
obtained from a stable process Weibull (θ = 3.2 and β = 4.8) with mean = 2.9312, while the later 10 periods had shifted
to Weibull (θ = 2.4 and β = 3.2) with mean = 2.1496. That is, the process mean was shifted from 2.9312 to 2.1496. One
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
observation is randomly chosen from each of the 20 sampling periods. These values and their corresponding
computed Ui, Vi, Z1,i, Z2,i, and Mi statistics with λ = 0.1 are given in Table VI.
Suppose that the desired stable-process ARL0 = 250. The control limit is obtained as 0.7089. Figure 1 displays the MaxEWMA chart
for the carbon-fiber data. Monitoring starts in Period 1. This figure identifies samples 15 and 17 as out of control, where samples 15
and 17 are related to the mean. Because U15 = 2.369 and U17 = 3.939, these two values suggest that the process mean has
decreased.
Example 2 Xie et al.7 presented a process with the time between failures in hours. The first 20 observations are collected from a
historical process. The estimates of the Weibull parameters are θ = 9.42 and β = 1.03. The last 10 observations were
simulated from the process mean shifted with Weibull (θ = 53.41 and β = 0.92), which means that the process has
improved. These values and their corresponding computed Ui, Vi, Z1,i, Z2,i, and Mi statistics with λ = 0.05 are given in
Table VII.
Suppose that the desired stable-process ARL0 = 100. The control limit is obtained as 0.4494. Figure 2 displays the MaxEWMA control
chart for monitoring time between failures. This figure identifies samples 24 ~ 30 as out of control. Because U14 = 5.946, U15 = 3.711,
U16 = -0.820, U17 = 0.895, U18 = 4.749, U19 = 1.828, and U20 = 1.996, these seven values suggest that the process mean has increased.
This finding confirmed that the process mean has shifted (improved).
Table VI. Individual observations and their corresponding statistics with λ = 0.1
Period xi y2,i = Ui Vi Z1,i Z2,i Mi = max{|Z1,i|, |Z2,i|}
UCL=0.7089
Figure 1. Maximum exponentially weighted moving average chart for monitoring the strengths of carbon fibers under ARL0 = 250. UCL, upper control limit
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
Table VII. Individual observations and their corresponding statistics with λ = 0.05
Period xi y2,i = Ui Vi Z1,i Z2,i Mi = max{|Z1,i|, |Z2,i|}
1 30.02 1.788 0.365 0.225 0.225 0.225
2 1.44 1.105 1.296 0.158 0.278 0.278
3 22.47 1.363 1.179 0.218 0.323 0.323
4 1.36 1.139 1.196 0.151 0.367 0.367
5 3.43 0.531 0.623 0.116 0.317 0.317
6 13.2 0.697 0.364 0.146 0.320 0.320
7 5.15 0.214 0.014 0.128 0.304 0.304
8 3.83 0.449 1.077 0.099 0.235 0.235
9 21 1.271 0.664 0.157 0.257 0.257
10 12.97 0.678 0.734 0.183 0.207 0.207
11 0.47 1.700 1.004 0.089 0.247 0.247
12 6.23 0.051 0.294 0.082 0.249 0.249
13 3.39 0.540 0.557 0.051 0.209 0.209
14 9.11 0.304 0.040 0.064 0.197 0.197
15 2.18 0.846 0.253 0.018 0.200 0.200
16 15.53 0.887 0.773 0.062 0.228 0.228
17 25.72 1.555 0.836 0.136 0.175 0.175
18 2.79 0.680 0.968 0.096 0.215 0.215
19 1.92 0.928 1.205 0.044 0.144 0.144
20 4.13 0.391 0.618 0.023 0.106 0.106
21 70.47 3.387 0.365 0.191 0.119 0.191
22 17.07 1.002 1.296 0.231 0.177 0.231
23 3.99 0.417 1.179 0.199 0.227 0.227
24 176.06 5.946 1.196 0.486 0.276 0.486
25 81.07 3.711 0.623 0.648 0.231 0.648
26 2.27 0.820 0.364 0.574 0.238 0.574
27 15.63 0.895 0.014 0.590 0.226 0.590
28 120.78 4.759 1.077 0.799 0.161 0.799
29 30.81 1.828 0.664 0.850 0.186 0.850
30 34.19 1.996 0.734 0.907 0.140 0.907
Figure 2. Maximum exponentially weighted moving average chart for monitoring time between failures under ARL0 = 100. UCL, upper control limit
6. Conclusions
To monitor a Weibull process with individual measurements, the power transformation, inverse erf function, and Box–Cox
transformation are studied to transform the Weibull data to a normal distribution. The simulation results indicate that the inverse
erf function is recommended for a Weibull process with β ≥ 1 and the power transformation method is recommended for Weibull
process with β < 1. We extended the work of Chen, Cheng, and Xie19 where a MaxEWMA control chart is proposed to monitor the
process mean and variability for rational subgroups. It combines the usual individuals chart and the moving range chart into a single
chart. The MaxEWMA control chart enables a simultaneous monitoring of the process mean and variability for a Weibull process with
individual measurements. Regarding the ARL value, the proposed control chart outperforms the combined I-MR charts in detecting
the process mean changes, sigma changes, and both mean and sigma changes.
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
The proposed chart can be extended to other distributions such as the Burr XII or Poisson-exponential distribution. To develop the
economic design of the MaxEWMA control chart is a future work.
References
1. Woodall WH, Montgomery DC. Some current directions in the theory and application of statistical process monitoring. Journal of Quality Technology
2014; 46(1):78–94.
2. Nelson PR. Control charts for Weibull processes with standards given. IEEE Transactions on Reliability 1979; 28(4):283–287. doi:10.1109/
TR.1979.5220605.
3. Bai DS, Choi IS. X and R control charts for skewed populations. Journal of Quality Technology 1995; 27(2):120–131.
4. Hawkins DM, Olwell DH. Cumulative Sum Charts and Charting for Quality Improvement. Springer: New York, NY, 1998; 1998.
5. Ramalhoto MF, Moaris M. Shewhart control charts for the scale parameter of a Weibull control variable with fixed and variable sampling intervals.
Journal of Applied Statistics 1999; 26(1):129–160. doi:10.1080/02664769922700.
6. Chang YS, Bai DS. Control charts for positively skewed populations with weighted standard deviations. Quality and Reliability Engineering
International 2001; 17(5):397–406. doi:10.1002/qre.427.
7. Xie M, Goh TN, Ranjan P. Some effective control chart procedures for reliability monitoring. Reliability Engineering and System Safety 2002;
77(2):143–150. doi:10.1016/S0951-8320(02)00041-8.
8. Chan LK, Cui HJ. Skewness correction X and R charts for skewed distributions. Naval Research Logistics 2003; 50(6):555–573. doi:10.1002/nav.10077.
9. Borror CM, Keats JB, Montgomery DC. Robustness of the time between events CUSUM. International Journal of Production Research 2003; 41(15):
3435–3444. doi:10.1080/0020754031000138321.
10. Nichols MD, Padgett WJ. A bootstrap control chart for Weibull percentiles. Quality and Reliability Engineering International 2006; 22(2):141–151.
doi:10.1002/qre.691.
11. Batson RG, Jeong Y, Fonseca DJ, Ray PS. Control charts for monitoring field data. Quality and Reliability Engineering International 2006; 22(7):
733–755. doi:10.1002/qre.725.
12. Chen H, Cheng Y. Non-normality effects on the economic-statistical design of Xbar charts with Weibull in-control time. European Journal of
Operational Research 2007; 176(2):986–998. doi:10.1016/j.ejor.2005.08.022.
13. Pascual F. EWMA charts for the Weibull shape parameter. Journal of Quality Technology 2010; 42(4):400–416.
14. Pascual F, Zhang H. Monitoring the Weibull shape parameter by control charts for the sample range. Quality and Reliability Engineering International
2011; 27(1):15–25. doi:10.1002/qre.1099.
15. Pascual F, Nguyen D. Moving range charts for monitoring the Weibull shape parameter with single-observation samples. Quality and Reliability
Engineering International 2011; 27(7):905–919. doi:10.1002/qre.1180.
16. Pascual F. Individual and moving ratio charts for Weibull processes. In Stochastic Orders in Reliability and Risk, pp. 331–350. Springer: New York,
2013. DOI:10.1007/978-1-4614-6892-9_17.
17. Pascual F, Li S. Monitoring the Weibull shape parameter by control charts for the sample range of type II censored data. Quality and Reliability
Engineering International 2012; 28(2):233–246. doi:10.1002/qre.1239.
18. Faraz A, Saniga EM, Heuchenne C. Shewhart control charts for monitoring reliability with Weibull lifetimes. Quality and Reliability Engineering
International 2015; 31(8):1565–1574. doi:10.1002/qre.1692.
19. Chen G, Cheng SW, Xie H. Monitoring process mean and variability with one EWMA chart. Journal of Quality Technology 2001; 33(2):223–233.
20. Crowder SV. Computation of ARL for combined individual measurement and moving range charts. Journal of Quality Technology 1987; 19(2):
98–102.
21. Hernandez F, Johnson RA. The large-sample behavior of transformations to normality. Journal of the American Statistical Association 1980; 75(372):
855–861. doi:10.1080/01621459.1980.10477563.
22. R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing: Vienna, Austria, 2016.
23. Delignette-Muller ML, Dutang C. Fitdistrplus: an R package for fitting distributions. Journal of Statistical Software 2015; 64(4):1–34.
24. Padgett WJ, Spurrier JD. Shewhart-type charts for percentiles of strength distributions. Journal of Quality Technology 1990; 22(4):283–288.
Appendix
2
Because Ui = y2,i ~ N(0, 1) and V i ¼ Φ1 χ 21 y 2;i y 2;i1 eNð0; 1Þ are mutually independent.
X
i1
Now, Z1,i = (1 λ)Z1,i + λUi and Z2,i = (1 λ)Z2,i + λVi. We have that Z 1;i ¼ λ ð1λÞj Uij þ ð1 λÞi Z 1;0 and Z 2;i ¼
j¼1
X
i1
λ ð1λÞj V ij þ ð1 λÞi Z 2;0 .
j¼1
λ
h i
Suppose that Z1,0 = Z2,0 = 0. We have V Z 1;i ¼ V Z 2;i ¼ 2λ 1 ð1 λÞ2i .
The in-control distribution of Mi is established as follows:
F ðmÞ ¼ PðMi ≤mÞ ¼ P Z 1;i ≤m;Z 2;i j≤mÞ
8 2 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1ffi3 92
> u0 >
< 6 u u 2λ =
¼ P Z 1;i ≤m P Z 2;i ≤m ¼ 2Φ4mt@ h iA7 5 1
>
: λ 1 ð1 λÞ2i >
;
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,
F.-K. WANG
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u0 1ffi 2 v u0 1ffi3 8 2 v
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u0 1ffi3 9
u u >
< u >
=
u@ h 2 λ A 6 u
i ϕ 4mt h@ 2λ A 7 6
i 5 2Φ4mt h u@ 2λ A
i 51 7
f ðmÞ ¼ 4t
λ 1 ð1 λÞ2i λ 1 ð1 λÞ2i >
: λ 1 ð1 λÞ2i >
;
where φ[] is the probability density function of the standard normal distribution. With a given ARL0, the UCL value is determined by
ARL0 ¼ UCL 1 .
1 ∫ f ðmÞdm
0
Authors' biography
Fu-Kwun Wang received his PhD degree in Industrial Engineering from Arizona State University, Tempe, USA. Currently, he is a
Distinguished Professor and Chair in the Department of Industrial Management at the National Taiwan University of Science and
Technology, Taiwan. His fields of interest are reliability engineering, statistical process control, and predictive analytics
Copyright © 2016 John Wiley & Sons, Ltd. Qual. Reliab. Engng. Int. 2016,