An Adaptive Exponentially Weighted Moving Average Control Chart for Monitoring Process Variances

Size: px
Start display at page:

Download "An Adaptive Exponentially Weighted Moving Average Control Chart for Monitoring Process Variances"

Transcription

1 An Adaptive Exponentially Weighted Moving Average Control Chart for Monitoring Process Variances Lianjie Shu Faculty of Business Administration University of Macau Taipa, Macau Abstract The exponentially weighted moving average (EWMA) control chart is efficient in detecting small changes in process parameters but less efficient when the changes are relatively large, due to what is known as the inertia problem. To diminish the inertia, an adaptive EWMA (AEWMA) chart has been proposed for monitoring process locations to improve over the traditional EWMA charts. The basic idea of the AEWMA scheme is to dynamically weight the past observations according to a suitable function of the current prediction error. This paper extends the idea of the AEWMA chart for monitoring process locations to the case of monitoring process dispersion. A Markov chain model is established to analyze and design the suggested chart. It is shown that the AEWMA dispersion chart performs better than the EWMA and other dispersion charts in terms of its ability to perform relatively well at both small and large changes in process dispersion. Keywords: Average Run Length; Statistical Process Control; Markov Chain; Normalizing transformation; Dispersion Charts 1

2 1 Introduction The exponentially weighted moving average (EWMA) control chart has received a great deal of attention in the quality control literature since it was first introduced by Roberts (1959). Many researchers have contributed to the theory and practical use of the EWMA scheme as a monitoring tool, primarily to detect shifts in the mean level of a normal process, see, for example, Robinson and Ho (1978), Crowder (1987, 1989), Lucas and Saccucci (1990), and Apley and Lee (2003). In addition to detecting changes in the process mean, it is also important to keep the process variance in control since an increase of process variance results in an increased number of defective units while a decrease of process variance implies an improvement of process capability (Acosta-Mejia et al. 1999). Recently, there is gradually increasing attention given to the use of EWMA charts as a tool for monitoring process variability. A sample of this research include Wortham and Ringer (1971), Sweet (1986), Ng and Case (1989), Crowder and Hamilton (1992), MacGregor and Harris (1993), Lu and Reynolds (1999), Acosta-Mejia et al. (1999), and Chen et al. (2001). To detect changes in process dispersion, rational subgroups of observations are often formed, and the subgroup variance or range is then computed for estimating the underlying process variation. Basically, the dispersion charts can be grouped into two categories based on if a transformation is performed to normalize the observations. The first category directly applies the traditional control charts to the subgroup variance or range for monitoring changes in process dispersion without normalizing transformations. For example, Ng and Case (1989) proposed an EWMA dispersion chart based on the subgroup range. Note that the distribution of subgroup variance is not normal but rightly skewed, which is inconsistent with the normality assumption often made in statistical process control (SPC). For this reason, some authors proposed using nonlinear functions to transform the sample variance into a normal or an approximately normal distribution so that traditional control charts can be easily applied to the normalized data. This results in the second class of dispersion charts. One of the widely discussed normalizing transformations is based on the logarithmic function of the subgroup variance, ln(s 2 ), due to its simplicity and efficiency. This transformation was first recommended by Box, Hunter, and Hunter (1978) for making inferences about variance of normally distributed data and was later developed by Crowder and Hamilton (1992) and Chang and Gan (1995) for the problem of 2

3 monitoring process dispersion. Chen et al. (2001) recently discussed using the EWMA chart of ln(s 2 ) for joint monitoring process location and dispersion. The essential motivation of using a normalizing transformation is that the transformed data are much more normally distributed than the subgroup variance itself, as pointed out in Crowder and Hamilton (1992). Furthermore, after the transformation, the variance of the normalized observations is often independent of the true process variance and remains constant over time. A change in the process variance can be translated into a shift in the mean level of the normalized variable. As a result, monitoring process variances is essentially equivalent to monitoring mean levels of the transformed data. Due to these properties, the normalizing transformation has been considered to be an efficient technique for monitoring process dispersion. The literature on the efficiency and robustness of the EWMA chart for monitoring process means indicates that it can provide greater sensitivity to small shifts than the Shewhart chart, but is not as effective as the Shewhart chart when the shifts in the process mean level are relatively large. Yashchin (1995) investigated the estimation efficiency of the EWMA scheme in terms of an inertia function. He showed that the inertia increases as the magnitude of the mean shift increases and thus the EWMA statistic with a small smoothing constant is not efficient in estimating abrupt mean changes of moderate and large magnitudes. To overcome the inertia problem, Yashchin (1995) proposed a modified EWMA statistic, EWMA-C, which can dynamically adjust the weight on past observations according to a function of the prediction error. Capizzi and Mastrotto (2003) developed a control chart based on the EWMA-C statistic for monitoring process means, which has been referred to as the adaptive EWMA (AEWMA) chart. Although the inertia problem of the EWMA chart can be counteracted in part by using the combined Shewhart-EWMA chart, the introduction of Shewhart limits only ameliorates this problem, as pointed out in Capizzi and Mastrotto (2003). In contrast, the AEWMA chart, which can be viewed as a smooth combination of the Shewhart and EWMA charts, can diminish the inertia problem. It has been shown that the AEWMA chart is able to offer an overall good detection performance against mean shifts of different sizes and performs better than the combined Shewhart-EWMA chart. Like the EWMA charts for monitoring process means, the EWMA-type dispersion charts such as the EWMA charts based on ln(s 2 ) also suffer from the inertia problem when detecting 3

4 large changes in process dispersion. Motivated by this, this paper extends the idea of AEWMA charts for monitoring process means to the case of monitoring process variability. A Markov chain model is developed to analyze the performance of the suggested chart. This simplifies the performance analysis without running a large number of simulations. The rest of the paper is organized as follows. In the next section, the AEWMA chart for monitoring process dispersion is introduced. Then, a two-stage design procedure is recommended for AEWMA dispersion charts. Next, the average run length (ARL) performance is compared with other control charts. After that, the performance of the AEWMA charts based on different normalizing transformations is discussed. Finally, some concluding remarks are given. 2 The AEWMA Chart Before discussing the AEWMA dispersion chart, the AEWMA chart for monitoring process means is first reviewed. Let µ 0 and σ 0 be the established standard values for the normal process mean and standard deviation, respectively, and denote Z t = (X t µ 0 )/(σ 0 / n), where X t is the tth subgroup mean of size n, i.e., X t = n i=1 X ti/n. The EWMA statistic for monitoring process means is Q t = (1 λ)q t 1 + λz t = Q t 1 + λ(z t Q t 1 ) where λ is a smoothing constant with 0 < λ 1. Yashchin (1995) showed that the EWMA scheme has optimality property when estimating mean shifts of small magnitudes within the class of linear estimators. However, the best estimating procedures cannot be linear if they are to adapt to changes of relatively large magnitude because the inertia increases as the magnitude of the shift increases. form To overcome the inertia problem, Yashchin (1995) suggested an AEWMA statistic of the Q t = Q t 1 + φ(e t ), where e t = Z t Q t 1 is the prediction error, and φ( ) is a score function motivated by some robust procedures such as Huber s function (Huber 1981) and Welsch s function (Holland and Welsch 1977). Note that, when e t 0, the AEWMA statistic can be rewritten as Q t = (1 φ(e t) e t )Q t 1 + φ(e t) e t Z t. 4

5 Clearly, the AEWMA statistic can adapt the weight on the past estimate Q t 1 according to the current prediction error to detect shifts of different sizes in a balanced way. Capizzi and Masarotto (2003) suggested using it for monitoring process means. For the sake of simplicity, this paper will limit the discussions on the following Huber s score function, e + (1 λ)γ, e < γ φ γ (e) = λe, e γ e (1 λ)γ, e > γ, while other score functions can also be similarly discussed. Note that when γ, φ γ (e) = λe, and the AEWMA statistic reduces to the EWMA scheme. Moreover, when λ = 1 or γ = 0, the AEWMA statistic performs essentially like a Shewhart statistic. Therefore, the AEWMA statistic can be considered as a smooth combination of the Shewhart statistic and the EWMA statistic. The following considers extensions of the AEWMA scheme to the case of monitoring process variability. Throughout the remaining of this paper, the following assumptions are made: (i) process observations are collected from an independent, and identically distributed (iid) normal process N(µ t, σ 2 t ); (ii) the only concern is to detect increases in the process variance, and the process mean is considered to be in control, i.e., µ t µ 0 ; and (iii) before an unknown time τ, σt 2 = σ0 2, and after τ, σ2 t = σ 2 > σ0 2. The objective is to detect an upward change in the process variance as soon as possible. For monitoring increases in the process standard deviation, Crowder and Hamilton (1992) proposed an EWMA-type dispersion chart of the form D t = max[0, (1 λ)d t 1 + λy t ], (1) where Y t is an estimate of the process variance/standard deviation or a simple function of them. Note that the EWMA statistic is reset to zero whenever it falls below zero for the purpose of quickly catching up with an upward change in the process standard deviation. The initial value of D t is usually set to the target value of zero, i.e., D 0 = 0. The EWMA chart declares an alarm when D t exceeds the upper control limit h. In Crowder and Hamilton (1992), the logarithmic function of the subgroup variance is used, namely, Y t = ln(st 2 /σ0 2 ), where n St 2 i=1 = (X ti X t ) 2. n 1 5

6 and The mean and variance of Y t = ln(st 2 /σ0 2 ) are approximated by µ Y = ln( σ2 t σ0 2 ) 1 n 1 1 3(n 1) (n 1) 4 (2) σ 2 Y = 2 n (n 1) (n 1) (n 1) 5, (3) respectively. From Equations (2) and (3), it is clear that the variance of Y t is independent of the true process variance σ 2 t and remains unchanged over the time, and that an increase in σ 2 t will lead to an increase in the mean level of Y t. More importantly, the transformed observations are much more normally distributed than the subgroup variance. These properties motivate the use of logarithmic transformation rather than the subgroup variance for monitoring process dispersion. Like the EWMA location chart, the EWMA dispersion chart in Equation (1) also encounters the inertia problem when reacting to large changes in process variability. problem, an AEWMA scheme can be applied to Y t, given by To diminish this W t = max[0, W t 1 + φ(e t )], (4) where e t = Y t W t 1 and W 0 = 0. The AEWMA chart signals an out-of-control situation when W t exceeds the upper control limit h. Decreases in process variability are also important. A one-sided lower chart for detecting decreases in process variability can be similarly defined as follows: L t = min[0, L t 1 + φ(e t )], To detect both increases and decreases in process dispersion, one can make the combined use of the two one-sided AEWMA charts. The one-sided upper AEWMA chart is chosen for discussion since it responds faster to increases in process variability, which is the main concern of this paper. Due to its simplicity and efficiency of the logarithmic transformation, the remaining of this paper will mainly focus on AEWMA charts based on Y t = ln(st 2 ) while the performance of AEWMA charts using other normalizing transformations will be also briefly discussed. 6

7 γ (σ Y ) 2 20 γ (σ Y ) λ λ (a) σ/σ 0 = 1.2 (b) σ/σ 0 = 2.0 Figure 1: Contour Plot of the Out-of-Control ARL Values of the AEWMA Chart based on ln(s 2 ) for n = 5 (In-Control ARL=200). 3 Design of AEWMA Charts The design of an AEWMA chart is a three-degree-of-freedom problem, which involves the selection of parameters λ, γ, and h. To get some implications on the choice of chart parameters, the parameter effects of λ and γ are first investigated. Then, a two-stage procedure is proposed to design an AEWMA chart for the purpose of providing an overall good detection performance at both a small increase and a large increase in the process standard deviation. 3.1 Effects of Parameters The choice of λ and γ has strong impacts on the performance of AEWMA charts. It is relatively easy to investigate the effect of λ and γ alone on the chart performance, which has been discussed in Yashchin (1995) and Capizzi and Mastrotto (2003). It is known that small values of λ and/or large values γ improve the detection performance of AEWMA charts at small mean shifts while large values of λ and/or small values of γ values improve the detection performance at large mean shifts. However, the joint effect of λ and γ on the run length performance is less clear since it depends on both parameters. To better understand the overall effect, Figure 1 shows contour 7

8 plots of the out-of-control ARL values of AEWMA charts based on ln(s 2 ) as a function of λ and γ for n = 5 when the zero-state in-control ARL is 200. From Figure 1 (a), when detecting a small increase (20%) in the process standard deviation, the ARL decreases as λ decreases or γ increases. From Figure 1 (b), when detecting a large increase (100%) in the process standard deviation, the ARL decreases as λ increases or γ decreases. Furthermore, some implications on the joint choice of AEWMA charting parameters aimed at balancing the sensitivity to both small and large changes in process dispersion can be observed from Figure 1. First, too small values of γ and too large values of λ should be avoided for this purpose. From Figure 1 (a), when γ is too small, say γ 0.5σ Y, the AEWMA chart provides a very poor performance at small increases in the process standard deviation, regardless of the values of λ. From Figure 1 (b), when λ is larger than 0.5, the out-of-control ARL of the AEWMA chart is nearly independent of the choice of γ. This implies that the AEWMA chart would not benefit from the introduction of parameter γ but suffer from the increased design complexity, if too large values of λ were used. Therefore, the selection of too small values of γ or too large values of λ cannot balance the protection against changes of different sizes. Consider the special cases when λ is chosen as large as λ = 1 and γ is set as small as γ = 0. As mentioned earlier, the AEWMA chart reduces to a Shewhart chart in these cases. Clearly, the Shewhart chart cannot provide a balanced performance at both small and large changes in process dispersion. Second, using a small value of λ and an appropriate value of γ can provide the AEWMA chart an overall good detection performance for both small and large changes in process dispersion. As shown from Figures 1 (a) and (b), the parameters within the range 0.1 λ 0.4 and 1.5σ Y γ 3σ Y seem to provide an acceptable detection performance for the AEWMA chart against both small and large changes in the process standard deviation. 3.2 Design Procedure Capizzi and Masarotto (2003) proposed a two-stage procedure to jointly search the optimal parameters of an AEWMA chart for efficiently detecting a small shift and a large shift in the process mean. Similarly, a two-stage procedure can be followed to design an AEWMA chart for monitoring process dispersion. However, note that Capizzi and Masarotto s procedure is a joint searching procedure. It requires extensive computation work and is thus complicated. Rather, 8

9 a two-stage sequential searching procedure is employed here for finding the nearly optimal parameters. The sequential searching procedure requires less computation work while providing a relatively good performance at both small and large changes in the process standard deviation. In Phase I of the two-stage sequential searching procedure, the optimal λ value of an EWMA chart for detecting a specified small change in the process standard deviation is first determined. Due to the attractive property that the parameter γ can be managed to improve the detection performance at large changes by a large percentage while causing a slight loss in the detection performance at small changes (refer to Table 4 for details and Capizzi and Masarotto 2003), this optimal λ value for the EWMA chart can be also considered approximately optimal for the AEWMA chart. In Phase II, based on the optimal λ value obtained in Phase I, the γ value is obtained to optimize the detection performance of the AEWMA chart at a specified large change in the process standard deviation. Details of the two-stage design procedure for the AEWMA chart are summarized below: (i) Choose a desired in-control ARL, and a small change (σ 1 ) and a large change (σ 2 ) in the process standard deviation. (ii) For the specified in-control ARL, find the value of λ giving the minimum ARL of an EWMA chart at σ 1. Some of the optimal λ values can be found in Crowder and Hamilton (1992). (iii) Select a small positive constant α (e.g., α = 0.05) to control the efficiency loss at σ 1 due to the introduction of the parameter γ. (iv) Based on the in-control ARL and the optimal λ value obtained in Step (ii), search the optimal γ value having the minimum ARL at σ 2 subject to the condition that the percentage increase in the out-of-control ARL at σ 1 is at most α. Tables 1 to 3 report a list of parameters of AEWMA charts based on ln(s 2 ) for different in-control ARL values. These parameters are adequate for most practical purposes, and are obtained using the forgoing strategy through a Markov chain approximation shown in Appendix A. As an example, suppose that the desired in-control ARL is 200, and that the small and large changes in process dispersion to be detected are σ 1 = 1.2σ 0 and σ 2 = 1.7σ 0. When the sample 9

10 size is n = 5, from Table 2 in Crowder and Hamilton (1992), the optimal value of λ of an EWMA chart at σ 1 = 1.2σ 0 is λ = 0.05, giving the minimum ARL of If α = 0.05, the optimal γ value is searched on the range 1σ Y γ 3σ Y with a step size of 0.025σ Y for minimizing the ARL at σ 2 = 1.7σ 0 subject to the condition that the out-of-control ARL of the AEWMA chart at σ 1 = 1.2σ 0 is no more than (i.e., ( ) 18.10). The value of γ = 1.6σ Y gives the minimum ARL at σ 2 = 1.7σ 0 of 3.44 and an out-of-control ARL of at σ 1 = 1.2σ 0. Thus, the combination (λ = 0.05, γ = 1.6σ Y, h = ) is considered to be optimal for detecting dispersion changes on the range [1.2σ 0, 1.7σ 0 ]. The same approach can be used to determine the optimal parameters of AEWMA charts for other cases. It can be observed from Tables 1 to 3 that a large value of λ or a small value of γ is more sensitive to large changes while a small value of λ or a large value of γ is more sensitive to small changes. This is consistent with the results in Yashchin (1995) and Capizzi and Masarotto (2003). Furthermore, examination of Tables 1 to 3 indicates a tendency that for a fixed in-control ARL, to detect changes in process dispersion on the same interval [σ 1, σ 2 ], the optimal values of λ and γ tend to increase as the sample size n increases. For example, from Table 1, to detect the changes in process dispersion on the interval [1.4σ 0, 2σ 0 ], the optimal λ value increases from 0.29 to 0.44 while the optimal γ value increases from 1.125σ Y to 1.325σ Y as the sample size increases from n = 3 to 8. This can be explained by changes in the mean and variance of Y t. Based on Equations (2) and (3), note that the mean of Y t tends to increase while the standard deviation of Y t tends to decrease as n increases. This in turn results in an increased signal-to-noise ratio (i.e., µ Y /σ Y ). Thus, given the same detection interval [σ 1, σ 2 ], the λ value chosen for optimizing the detection performance at the same σ 1 value will be larger for larger n values as a large value of λ is more sensitive to large changes. On the other hand, the increased signal-to-noise ratio of Y t due to increase of n requires the selection of a larger γ value in order not to cause much loss in the detection efficiency at σ 1. 4 Performance Comparisons In this section, the run length distribution of the AEWMA chart based on ln(s 2 ) is first discussed. This discussion will provide clear insights into the ARL performance. Run length is the number of observations until the first out-of-control signal triggered by a control chart. As 10

11 Table 1: Optimal Parameters for AEWMA Charts based on ln(s 2 ) When the In-Control ARL is Equal to 200 n = 3 n = 5 n = 8 σ 1 /σ 0 σ 2 /σ 0 λ γ (σ Y ) h λ γ (σ Y ) h λ γ (σ Y ) h Note: γ is given in the unit of standard deviation of Y t, σ Y. Table 2: Optimal Parameters for AEWMA Charts based on ln(s 2 ) When the In-Control ARL is Equal to 370 n = 3 n = 5 n = 8 σ 1 /σ 0 σ 2 /σ 0 λ γ (σ Y ) h λ γ (σ Y ) h λ γ (σ Y ) h Note: γ is given in the unit of standard deviation of Y t, σ Y. 11

12 Table 3: Optimal Parameters for AEWMA Charts based on ln(s 2 ) When the In-Control ARL is Equal to 500 n = 3 n = 5 n = 8 σ 1 /σ 0 σ 2 /σ 0 λ γ (σ Y ) h λ γ (σ Y ) h λ γ (σ Y ) h Note: γ is given in the unit of standard deviation of Y t, σ Y. shown in Appendix A, the run length distribution of the AEWMA chart based on ln(s 2 ) can be approximated by a Markov chain. Next, the ARL performance of the AEWMA chart based on ln(s 2 ) is compared with the traditional EWMA and combined Shewhart-EWMA charts. Moreover, the AEWMA chart is compared with the change-point CUSUM (CP-CUSUM) chart when the process mean µ 0 is known. 4.1 Run Length Distribution of AEWMA Charts As an illustrative example, Figure 2 plots the zero-state in-control probability mass function (PMF) of run length from T = 1 to 50 for the AEWMA chart of ln(s 2 ) with λ = 0.1, γ = 2σ Y, and n = 5. The control limit is set at h = to provide a zero-state in-control ARL of 200. The run length distribution of the traditional EWMA chart of ln(s 2 ) with the same values of λ = 0.1 and n = 5 is also plotted for comparison. From Figure 2, in the in-control situation, the false alarm rate curve of the AEWMA chart is close to that of the EWMA chart except a little bit large false alarm rate at short run lengths T < 3. For run lengths T > 3, the run length distribution curves of both charts are almost the same. 12

13 6 x 10 3 EWMA AEWMA 5 4 Probability Run Length Figure 2: In-Control Run Length Distributions of the AEWMA and EWMA Charts EWMA AEWMA EWMA AEWMA Probability 0.03 Probability Run Length Run Length (a) σ/σ 0 = 1.2 (b) σ/σ 0 = 2.0 Figure 3: Out-of-Control Run Length Distributions of the AEWMA and EWMA Charts. 13

14 Figures 3 (a) and (b) plot the zero-state out-of-control run length PMF of both charts under increases in the process standard deviation of sizes of 20% and 100%, respectively. As can be seen from Figure 3 (a), when detecting a small increase in the process standard deviation, the signal probability of the AEWMA chart is larger than the EWMA chart at short run lengths T 3 and almost the same at T > 3. Therefore, there is negligible difference in the detection power between the AEWMA and EWMA charts for signalling small changes in process variability. On the other hand, when detecting a large increase in the process standard deviation, the AEWMA chart has a much higher signal probability than the EWMA chart at the very short run length T = 1, as can be observed from Figure 3 (b). This implies a better sensitivity of the AEWMA chart to large increases in process variability than the EWMA chart. These observations indicate a good property of the AEWMA chart. That is, the AEWMA chart can be designed to have performance close to the EWMA chart at a small increase in process variability but better performance than the EWMA chart at a large increase in process variability. 4.2 Comparison with EWMA Charts Table 4 compares the ARL of AEWMA charts with the traditional EWMA chart for monitoring increases in the process standard deviation. Both zero-state and steady-state ARL results are provided. The AEWMA and EWMA charts are designed based on λ = 0.1 and n = 5. All charts are designed to have a zero-state in-control ARL of 200. The same values of control limits were used for computing both zero-state and steady-state out-of-control ARL values. Note that when γ 3σ Y, the zero-state/steady-state ARL performance of the AEWMA chart is the same as that of the EWMA chart. This is because when γ takes a large value, the estimation errors are very unlikely to exceed γ, and the AEWMA scheme performs essentially the same as the regular EWMA scheme. However, it can be observed from Table 4 that when γ decreases, the AEWMA chart can perform substantially better than the EWMA chart for detecting large increases in process dispersion and perform closely to the EWMA charts at small increases in process dispersion. For example, when γ decreases from 3σ Y to 1.5σ Y, the percentage decrease of the zero-state out-of-control ARL at σ = 3σ 0 is around 30.6% (( )/1.83) while the percentage increase in the ARL at σ = 1.2σ 0 is 6.55% (( )/18.25) only. To get some visual insights on this, Figure 4 further plots the ARL curves of the AEWMA chart 14

15 with γ = 1.5σ Y and the EWMA chart. The ARL scale is chosen to be logarithmic. From Figure 4, it is clear that the AEWMA chart performs closely to the EWMA chart at small changes in the process standard deviation but better than the EWMA chart at relatively large changes. The above observations demonstrate the benefit of introducing the additional parameter γ for the AEWMA chart. In particular, the parameter γ can be managed to improve the performance of an AEWMA chart at larger changes in process dispersion with minor loss in the efficiency in detecting smaller changes in process dispersion. This agrees with the expectation of the superiority of AEWMA estimators over conventional EWMA statistics in catching up with large changes in the process mean level. From Table 4, it is observed that the zero-state ARL is larger than the steady-state ARL. The difference between the two ARL results depends on both values of γ and λ. From Table 4, it is clear that the difference between the zero and steady-state ARL values decreases as γ decreases. Moreover, it is also well known that this ARL difference decreases as λ increases (Lucas and Saccucci 1990). Note that when γ = 0 or λ = 1, the AEWMA chart reduces to a Shewhart chart. In both cases, the steady-state ARL would be the same as the zero-state ARL. As can be seen from Table 4, when γ = 0.5σ Y ( 0.4 for n = 5), the steady-state ARL is almost the same as the zero-state ARL. In practice, however, the difference is minor and not critical since both ARL results lead to qualitatively the same conclusions. Either one suffices for most practical purposes. 4.3 Comparison with Shewhart-EWMA Charts To improve the performance of the traditional EWMA charts for monitoring large process means, the combined Shewhart-EWMA chart has been suggested (Lucas and Saccucci 1990; Albin, Kang and Shea 1997). This scheme is achieved by adding Shewhart limits to an EWMA scheme. It is also straightforward to extend this idea to the dispersion monitoring problem. Figure 5 compares the ARL between the AEWMA and Shewhart-EWMA charts based on ln(s 2 ) for n = 5. The zero-state in-control ARL is maintained as 200. The considered AEWMA chart is designed for σ 1 = 1.2σ 0 and σ 2 = 2σ 0, and α = 0.05 (λ = 0.05, γ = 1.6σ Y, and h = as shown in Table 1). For the combined Shewhart-EWMA chart, the same value of λ = 0.05 is used, and two different Shewhart control limits of SCL = 2σ Y and SCL = 3σ Y are considered here. 15

16 Table 4: ARL Values of AEWMA Charts based on ln(s 2 ) When λ = 0.1 and n = 5 AEWMA γ (σ Y )= EWMA σ/σ 0 Type h= Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state Zero-state Steady-state

17 200 EWMA AEWMA 200 EWMA AEWMA ARL ARL σ /σ σ /σ 0 (a) Zero-State (b) Steady-State Figure 4: ARL Comparisons between AEWMA and EWMA Charts base on ln(s 2 ) When λ = 0.1 and n = 5. Simulations were used to obtain the ARL of the combined Shewhart-EWMA chart of ln(s 2 ). From Figure 5, it is clear that the AEWMA chart often performs closely to or better than the Shewhart-EWMA charts for a wide range of changes in process dispersion. This result is consistent with that observed in the situation of monitoring process means in Capizzi and Mastrotto (2003). As they pointed out, the Shewhart-EWMA chart can only ameliorates the inertial problem of the EWMA chart in detecting large changes in process means while the AEWMA chart can diminish this inertia problem. Therefore, the AEWMA chart performs better than the Shewhart-EWMA chart for monitoring process means. 4.4 Comparison with CP-CUSUM Charts When µ 0 is Known The above comparisons assume that the target mean is unknown. When the target mean is known and is always in control, a more accurate estimator T 2 t = n i=1 (X ti µ 0 ) 2 n can be used to estimate the process variance. Although both St 2 and Tt 2 are unbiased estimators of σt 2, the former does not utilize the process mean information but rather estimate it. The mean estimation results in a loss of one degree of freedom in the distribution of St 2. Acosta-Mejia et 17

18 200 Shewhart EWMA (SCL=2σ Y ) Shewhart EWMA (SCL=3σ Y ) AEWMA 50 ARL σ /σ 0 Figure 5: Zero-State ARL Comparisons between AEWMA and Shewhart-EWMA Charts base on ln(s 2 ) When λ = 0.1 and n = CP CUSUM (K=1.193) CP CUSUM (K=1.848) AEWMA 50 ARL σ /σ 0 Figure 6: Zero-State ARL Comparisons between AEWMA and CP-CUSUM Charts When µ 0 is Known and n = 5. 18

19 al. (1999) proposed a CP-CUSUM chart based on T 2 t statistic derived from the sequential probability ratio test (SPRT), given by C t = max[0, C t 1 + T 2 t σ 2 0 k]. An out-of-control signal is triggered when C t exceeds a threshold h C. The CP-CUSUM chart using a SPRT reference value k = ln(σ2 /σ 2 0 ) 1 1/(σ 2 /σ 2 0 ) is proved to be optimal at detecting a variance of size σ 2 in Moustakides (1986). Figure 6 compares the zero-state ARL between the AEWMA chart based on ln T 2 t and the CP-CUSUM chart when µ 0 is known and n = 5. Two reference values of k = and k = are considered for the CP-CUSUM chart for providing an optimal detection performance at σ = 1.2σ 0 and σ = 2σ 0, respectively. The corresponding control limits are set as h C = and h C = 8.16 for providing an in-control zero-state ARL of 200. The parameters for the AEWMA chart are chosen as λ = 0.05, γ = 1.7σ Y, and h = aimed at optimizing the detecting performance on the interval [1.2σ 0, 2σ 0 ], which are obtained based on the foregoing design strategy. From Figure 6, it can be observed that a single CUSUM chart cannot perform well at both a small change and a large change in process dispersion while the AEWMA chart can provide a relatively good detection performance at both small and large changes. The AEWMA chart always provides an ARL either the shortest or close to the shortest for all the changes considered here. 5 AEWMA Charts based on Other Transformations The logarithmic transformation is simple to implement but it only provides an approximate normal distribution. Other more complicated transformation could be employed to generate even more normally distributed observations. One such transformation is based on the inverse normal distribution. Quesenberry (1995) showed that statistic ( )) (n 1)S P S,t = Φ (G 1 2 t has a standard normal distribution, where G ( ) is the cumulative distribution function of a chi-square distribution with n 1 degrees of freedom, and Φ 1 is the inverse of a standard σ

20 normal distribution. Besides these transformations, Wilson and Hilferty (1931) proposed a transformation based on the cubic root of a chi-square distribution to normalize the sample variance. They showed that 3 χ 2 n is approximately normally distributed with mean 1 2/(9n) and variance 2/(9n). Thus, χ S,t = ( (S 2 t σ 2 0 ) 1/3 ( ) ) / (n 1) 9(n 1) will have an approximate standard normal distribution when σ = σ 0. Acosta-Mejia et al. (1999) showed that the CUSUM charts based on χ S and P S have almost the same performance. That is the evidence that the distribution of χ S,t is almost identical to a standard normal distribution. Acosta-Mejia et al. (1999) compared the performance of CUSUM charts based on the three different normalizing transformations: ln(s 2 ), χ S and P S. They showed that the CUSUM charts based on χ S and P S have similar performance and both perform better than the CUSUM chart based on ln(s 2 ). However, note that the statistic ln(s 2 ) is not standardized into zero mean and unit variance while P S,t and χ S,t are standardized in Acosta-Mejia et al. (1999). Therefore, their comparison result is a little bit unfair. To be fair, ln(st 2 /σ0 2 ) is also standardized by using its mean and variance given by Equations (2) and (3) respectively in the following performance comparison of AEWMA charts based on different transformations. Table 5 reports the ARL values of AEWMA charts based on the above three standardized transformations. The ARL values of the AEWMA charts based on χ S and P S are also approximated using a Markov chain approach, details of which are shown in Appendix B. From Table 5, for the same value of γ, the AEWMA charts based on χ S and P S have nearly the same performance. This is consistent with that of Acosta-Mejia et al. (1999). Moreover, it can be observed that the performance of the AEWMA chart based on the standardized ln(s 2 ) is competitive to the charts based on χ S and P S. Clearly, the AEWMA chart based on the standardized ln(s 2 ) performs better than the AEWMA charts based on χ S and P S when γ = 1.5 but worse when γ = 2.5. When γ = 2, the AEWMA chart based on the standardized ln(s 2 ) performs better than the AEWMA charts based on χ S and P S for small increases in the process standard deviation σ 1.3σ 0 but worse for σ > 1.3σ 0. 20

21 Table 5: Zero-State ARL Values of AEWMA Charts based on Other Normalizing Transformation Functions When λ = 0.1 and n = 5 Using ln(s 2 ) Using χ S Using P S γ = σ/σ 0 h = Concluding Remarks This paper extends the idea of AEWMA charts for monitoring process means to the case of monitoring process dispersion. A Markov chain model is established to evaluate the performance of AEWMA dispersion charts, which simplifies the analysis without a large number of simulation replicates. A two-stage design procedure is suggested for the AEWMA charts aimed at providing a balanced detection performance against changes in the process standard deviation of different sizes. The comparison results show that the suggested control chart offers a more balanced protection performance against changes of different sizes in process dispersion than the traditional EWMA chart, the combined Shewhart-EWMA chart, and the CP-CUSUM chart. This paper mainly focus on the AEWMA chart based on the logarithmic transformation of the subgroup variance due to its simplicity and efficiency while other different transformations can also be used in the AEWMA chart. Although detecting decreases in process dispersion is not discussed in detail in this paper, one can easily extend the AEWMA chart to this case. Furthermore, joint monitoring of process location and dispersion has attracted a lot of attentions 21

22 recently in SPC research (Chen et al. 2001; Reynolds and Stoumbos 2004). The proposed technique can be developed for this purpose. Some research work is doing in this way. Appendix A: ARL Computations for AEWMA Charts of ln(s 2 ) Similar to the method of Brook and Evans (1972) and Lucas and Saccucci (1990), the incontrol region [0, h] can be divided into sub-regions to obtain a discretized Markov chain for approximating the run length distribution of an AEWMA chart based on ln(s 2 ). Let m be the number of sub-intervals along the W -axis on the interval [0, h]. Thus, the width of each subinterval is ω = (2h)/(2m 1), except that the width of the first one is ω/2. These sub-intervals are labelled as i = 1, 2,, m, and are approximated by their mid values, (i 1)ω. Let P (i, j) be the transition probability of W t in Equation (4) from state i to state j. For i = 1, 2,..., m and j 1, P (i, j) = Pr{ W t in state j W t 1 in state i} = Pr{(j 1)ω 0.5ω < (i 1)ω + φ(e t ) (j 1)ω + 0.5ω} = Pr{(i 1)ω + φ 1 [(j i 0.5)ω] < Y t (i 1)ω + φ 1 [(j i + 0.5)ω]} = Pr{a 1 < Y t a 2 }, where a 1 = (i 1)ω + φ 1 [(j i 0.5)ω], and a 2 = (i 1)ω + φ 1 [(j i + 0.5)ω]. The function φ 1 ( ) is the inverse of the Huber score function given by (Capizzi and Masarotto 2003) φ 1 γ (v) = v (1 λ)γ, v/γ v + (1 λ)γ v < λγ λγ v λγ v > λγ. Note that S 2 t /σ 2 0 has a gamma distribution with shape parameter α = n

23 and scale parameter 2σ 2 β = (n 1)σ0 2. It follows that Pr{a 1 < Y t a 2 } = F (exp(a 2 ), α, β) F (exp(a 1 ), α, β), where F (, α, β) is the cumulative distribution function of a gamma distribution with shape parameter α and scale parameter β defined above. For i = 1, 2,..., m and j = 1, P (i, j) = F (exp(a 2 ), α, β). Let the matrix R m m be the transition probability matrix, which contains the probabilities of going from one transient state to other states. Then, the probability mass function of run length T and the ARL are given by Pr(T = t) = p ini (R t 1 R t ) 1, and ARL = p ini (I R) 1 1, respectively, where p ini is any initial probability vector of states, and 1 is a column vector of ones. In the zero-state analysis, the initial probability vector p ini corresponds to W 0 = 0. To obtain the steady-state ARL, Lucas and Saccucci (1990) suggested using the cyclical steady-state probability vector, p ss, which is solved from p ss = P 1 p ss subject to 1 p ss = 1, where R (I R)1 P 1 = The cyclical steady-state ARL of the AEWMA chart is obtained in a similar way, given by ARL = p ss (I R) 1 1. The ARL of the EWMA chart base on ln S 2 was obtained using integral equation approach in Crowder and Hamilton (1992). Note that the EWMA scheme is a special case of the AEWMA scheme with γ +. Thus, the above Markov chain for the AEWMA chart can be modified to compute the ARL of the EWMA chart by taking a sufficiently large value of γ. Within our investigations, both the Markov chain approach and the integral equation approach can produce nearly the same results. In this paper, the Markov chain approach was used to evaluate the ARL of an EWMA chart of ln S 2. 23

24 Appendix B: ARL Computations for AEWMA Charts based on χ S and P S The transition probability matrix of the AEWMA charts based on χ S and P S can be obtained in a way similar to that of the AEWMA chart based on ln(s 2 ), which is summarized below. For i = 1, 2,..., m and j 1, the transition probability of the AEWMA chart based on χ S statistic is given by P (i, j) = Pr{a 1 < χ S,t a 2 } = F (b 2, α, β) F (b 1, α, β) where and b 1 = b 2 = (a 1 (a 2 2 9(n 1) + 2 9(n 1) + ( 1 ( 1 ) ) 3 2 9(n 1) ) ) (n 1) The values of a 1, a 2, α, β and F (, α, β) are defined as above. When j = 1, P (i, j) = F (b 2, α, β). For i = 1, 2,..., m and j 1, the transition probability of the AEWMA chart based on P S statistic is given by P (i, j) = Pr{a 1 < P S,t a 2 } = F (b 4, α, β) F (b 3, α, β), where and ( ) b 3 = G 1 Φ(a1 ) n 1 ( ) b 4 = G 1 Φ(a2 ). n 1 G 1 ( ) represents the inverse of a chi-square distribution with n 1 degrees of freedom. When j = 1, P (i, j) = F (b 4, α, β). 24

25 After obtaining the transition probability matrix, it is straightforward to compute both the zero-state and steady-state ARL values of the AEWMA charts based on χ S and P S. References [1] Acosta-Mejia, C. A., Pignatiello, J. J. Jr. and Rao, B. V. (1999). A Comparison of Control Charting Procedures for Monitoring Process Dispersion. IIE Transactions 31, pp [2] Albin, S. L., Kang, L. and Shea, G. (1997). An X and EWMA Chart for Individual Observations. Journal of Quality Technology 29, pp [3] Apley, D. W. and Lee, H. C. (2003). Design of Exponentially Weighted Moving Average Control Charts for Autocorrelated Processes With Model Uncertainty. Technometrics 45, pp [4] Box, G. E. P., Hunter, W. and Hunter J. (1978). Statistics for Experiments. John Wiley & Sons, New York, NY. [5] Brook, D. and Evans, D. A. (1972). An Approach to the Probability Distribution of CUSUM Run Length. Biometrika 59, pp [6] Capizzi, G. and Masarotto, G. (2003). An Adaptive Exponentially Weighted Moving Average Control Chart. Technometrics 45, pp [7] Chang, T. C. and Gan, F. F. (1995). A Cumulative Sum Control Chart for Monitoring Process Variance. Journal of Quality Technology 27, pp [8] Chen, G. M, Cheng, S. W. and Xie, H. S. (2001). Monitoring Process Mean and Variability with One EWMA Chart. Journal of Quality Technology 33, pp [9] Crowder, S. V. (1987). A Simple Method for Studying Run Length Distributions of Exponentially Weighted Moving Average Control Charts. Technometrics 29, pp [10] Crowder, S. V. (1989). Design of Exponentially Weighted Moving Average Schemes. Journal of Quality Technology 21, pp

26 [11] Crowder, S. V. and Hamilton, M. (1992). Average Run Lengths of EWMA Controls for Monitoring a Process Standard Deviation. Journal of Quality Technology 24, pp [12] Holland, P. W. and Welsch, R. E. (1977). Robust Regression Using Iteratively Reweighted Least-Squares. Communications in Statistics: Theory and Methods 6, pp [13] Huber, P. J. (1981). Robust Statistics. John Wiley & Sons, New York, NY. [14] Lu, C. W. and Reynolds, M. R. Jr. (1999). Control Charts for Monitoring the Mean and Variance of Autocorrelated Processes. Journal of Quality Technology 31, pp [15] Lucas, J. M. and Saccucci, M. S. (1990). Exponentially Weighted Moving Average Control Schemes: Properties and Enhancements. Technometrics 32(1), pp [16] MacGregor, J. F. and Harris, T. J. (1993). The Exponentially Weighted Moving Variance. Journal of Quality Technology 25, pp [17] Moustakides, G. V. (1986). Optimal Stopping Times for Detecting Changes in Distributions. Annals of Statistics 14, pp [18] NG, C. H. and Case, K. E. (1989). Development and Evaluation of Control Charts Using Exponentially Weighted Moving Averages. Journal of Quality Technology 21, pp [19] Quesenberry, C. P. (1995). On Properties of Q Charts Variables. Journal of Quality Technology 27, pp [20] Reynolds, M. R. Jr. and Stoumbos, Z. G. (2004). Control Charts and The Efficient Allocation of Sampling Resources. Technometrics 46, pp [21] Roberts, S. W. (1959). Control Chart Tests Based on Geometric Moving Averages. Technometrics 1, pp [22] Robinson, P. B and Ho, T. Y. (1978). Average Run Lengths of Geometric Moving Average Charts by Numerical Methods. Technometrics 20, pp [23] Sweet, A. L. (1986). Control Charts Using Coupled Exponentially Weighted Moving Averages. IIE Transactions 18, pp

27 [24] Wilson, E. P. and Hilferty, M. M. (1931). The Distribution of Chi-square. Proceeding of the National Academy of Science 17, pp [25] Wortham, A. W. and Ringer, L. J. (1971). Control via Exponential Smoothing. The Logistics Review 7, pp [26] Yashchin, E. (1995). Estimating the Current Mean of a Process Subject to Abrupt Changes. Technometrics 37, pp

Exponentially Weighted Moving Average Control Charts for Monitoring Increases in Poisson Rate

Exponentially Weighted Moving Average Control Charts for Monitoring Increases in Poisson Rate Exponentially Weighted Moving Average Control Charts for Monitoring Increases in Poisson Rate Lianjie SHU 1,, Wei JIANG 2, and Zhang WU 3 EndAName 1 Faculty of Business Administration University of Macau

More information

The Robustness of the Multivariate EWMA Control Chart

The Robustness of the Multivariate EWMA Control Chart The Robustness of the Multivariate EWMA Control Chart Zachary G. Stoumbos, Rutgers University, and Joe H. Sullivan, Mississippi State University Joe H. Sullivan, MSU, MS 39762 Key Words: Elliptically symmetric,

More information

CUMULATIVE SUM CHARTS FOR HIGH YIELD PROCESSES

CUMULATIVE SUM CHARTS FOR HIGH YIELD PROCESSES Statistica Sinica 11(2001), 791-805 CUMULATIVE SUM CHARTS FOR HIGH YIELD PROCESSES T. C. Chang and F. F. Gan Infineon Technologies Melaka and National University of Singapore Abstract: The cumulative sum

More information

JINHO KIM ALL RIGHTS RESERVED

JINHO KIM ALL RIGHTS RESERVED 014 JINHO KIM ALL RIGHTS RESERVED CHANGE POINT DETECTION IN UNIVARIATE AND MULTIVARIATE PROCESSES by JINHO KIM A dissertation submitted to the Graduate School-New Brunswick Rutgers, The State University

More information

THE DETECTION OF SHIFTS IN AUTOCORRELATED PROCESSES WITH MR AND EWMA CHARTS

THE DETECTION OF SHIFTS IN AUTOCORRELATED PROCESSES WITH MR AND EWMA CHARTS THE DETECTION OF SHIFTS IN AUTOCORRELATED PROCESSES WITH MR AND EWMA CHARTS Karin Kandananond, kandananond@hotmail.com Faculty of Industrial Technology, Rajabhat University Valaya-Alongkorn, Prathumthani,

More information

Self-Starting Control Chart for Simultaneously Monitoring Process Mean and Variance

Self-Starting Control Chart for Simultaneously Monitoring Process Mean and Variance International Journal of Production Research Vol. 00, No. 00, 15 March 2008, 1 14 Self-Starting Control Chart for Simultaneously Monitoring Process Mean and Variance Zhonghua Li a, Jiujun Zhang a,b and

More information

A problem faced in the context of control charts generally is the measurement error variability. This problem is the result of the inability to

A problem faced in the context of control charts generally is the measurement error variability. This problem is the result of the inability to A problem faced in the context of control charts generally is the measurement error variability. This problem is the result of the inability to measure accurately the variable of interest X. The use of

More information

The Efficiency of the VSI Exponentially Weighted Moving Average Median Control Chart

The Efficiency of the VSI Exponentially Weighted Moving Average Median Control Chart The Efficiency of the VSI Exponentially Weighted Moving Average Median Control Chart Kim Phuc Tran, Philippe Castagliola, Thi-Hien Nguyen, Anne Cuzol To cite this version: Kim Phuc Tran, Philippe Castagliola,

More information

An Investigation of Combinations of Multivariate Shewhart and MEWMA Control Charts for Monitoring the Mean Vector and Covariance Matrix

An Investigation of Combinations of Multivariate Shewhart and MEWMA Control Charts for Monitoring the Mean Vector and Covariance Matrix Technical Report Number 08-1 Department of Statistics Virginia Polytechnic Institute and State University, Blacksburg, Virginia January, 008 An Investigation of Combinations of Multivariate Shewhart and

More information

Weighted Likelihood Ratio Chart for Statistical Monitoring of Queueing Systems

Weighted Likelihood Ratio Chart for Statistical Monitoring of Queueing Systems Weighted Likelihood Ratio Chart for Statistical Monitoring of Queueing Systems Dequan Qi 1, Zhonghua Li 2, Xuemin Zi 3, Zhaojun Wang 2 1 LPMC and School of Mathematical Sciences, Nankai University, Tianjin

More information

First Semester Dr. Abed Schokry SQC Chapter 9: Cumulative Sum and Exponential Weighted Moving Average Control Charts

First Semester Dr. Abed Schokry SQC Chapter 9: Cumulative Sum and Exponential Weighted Moving Average Control Charts Department of Industrial Engineering First Semester 2014-2015 Dr. Abed Schokry SQC Chapter 9: Cumulative Sum and Exponential Weighted Moving Average Control Charts Learning Outcomes After completing this

More information

A Simulation Study Of The Impact Of Forecast Recovery For Control Charts Applied To ARMA Processes

A Simulation Study Of The Impact Of Forecast Recovery For Control Charts Applied To ARMA Processes Journal of Modern Applied Statistical Methods Volume 1 Issue 2 Article 43 11-1-2002 A Simulation Study Of The Impact Of Forecast Recovery For Control Charts Applied To ARMA Processes John N. Dyer Georgia

More information

Control charting normal variance reflections, curiosities, and recommendations

Control charting normal variance reflections, curiosities, and recommendations Control charting normal variance reflections, curiosities, and recommendations Sven Knoth September 2007 Outline 1 Introduction 2 Modelling 3 Two-sided EWMA charts for variance 4 Conclusions Introduction

More information

Confirmation Sample Control Charts

Confirmation Sample Control Charts Confirmation Sample Control Charts Stefan H. Steiner Dept. of Statistics and Actuarial Sciences University of Waterloo Waterloo, NL 3G1 Canada Control charts such as X and R charts are widely used in industry

More information

Mathematical and Computer Modelling. Economic design of EWMA control charts based on loss function

Mathematical and Computer Modelling. Economic design of EWMA control charts based on loss function Mathematical and Computer Modelling 49 (2009) 745 759 Contents lists available at ScienceDirect Mathematical and Computer Modelling journal homepage: www.elsevier.com/locate/mcm Economic design of EWMA

More information

Nonparametric Multivariate Control Charts Based on. A Linkage Ranking Algorithm

Nonparametric Multivariate Control Charts Based on. A Linkage Ranking Algorithm Nonparametric Multivariate Control Charts Based on A Linkage Ranking Algorithm Helen Meyers Bush Data Mining & Advanced Analytics, UPS 55 Glenlake Parkway, NE Atlanta, GA 30328, USA Panitarn Chongfuangprinya

More information

Robustness of the EWMA control chart for individual observations

Robustness of the EWMA control chart for individual observations 1 Robustness of the EWMA control chart for individual observations S.W. Human Department of Statistics University of Pretoria Lynnwood Road, Pretoria, South Africa schalk.human@up.ac.za P. Kritzinger Department

More information

Likelihood Ratio-Based Distribution-Free EWMA Control Charts

Likelihood Ratio-Based Distribution-Free EWMA Control Charts Likelihood Ratio-Based Distribution-Free EWMA Control Charts CHANGLIANG ZOU Nankai University, Tianjin, China FUGEE TSUNG Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong

More information

Modified cumulative sum quality control scheme

Modified cumulative sum quality control scheme Journal of Engineering and Technology Research Vol. 2(12), pp. 226-236, December 2010 Available online at http:// www.academicjournals.org/jetr ISSN 2006-9790 2010 Academic Journals Full Length Research

More information

Performance of Conventional X-bar Chart for Autocorrelated Data Using Smaller Sample Sizes

Performance of Conventional X-bar Chart for Autocorrelated Data Using Smaller Sample Sizes , 23-25 October, 2013, San Francisco, USA Performance of Conventional X-bar Chart for Autocorrelated Data Using Smaller Sample Sizes D. R. Prajapati Abstract Control charts are used to determine whether

More information

Optimal Design of Second-Order Linear Filters for Control Charting

Optimal Design of Second-Order Linear Filters for Control Charting Optimal Design of Second-Order Linear Filters for Control Charting Chang-Ho CHIN School of Mechanical and Industrial Systems Engineering Kyung Hee University Yongin-si, Gyeonggi-do 446-701 Republic of

More information

Regenerative Likelihood Ratio control schemes

Regenerative Likelihood Ratio control schemes Regenerative Likelihood Ratio control schemes Emmanuel Yashchin IBM Research, Yorktown Heights, NY XIth Intl. Workshop on Intelligent Statistical Quality Control 2013, Sydney, Australia Outline Motivation

More information

A Modified Poisson Exponentially Weighted Moving Average Chart Based on Improved Square Root Transformation

A Modified Poisson Exponentially Weighted Moving Average Chart Based on Improved Square Root Transformation Thailand Statistician July 216; 14(2): 197-22 http://statassoc.or.th Contributed paper A Modified Poisson Exponentially Weighted Moving Average Chart Based on Improved Square Root Transformation Saowanit

More information

An exponentially weighted moving average scheme with variable sampling intervals for monitoring linear profiles

An exponentially weighted moving average scheme with variable sampling intervals for monitoring linear profiles An exponentially weighted moving average scheme with variable sampling intervals for monitoring linear profiles Zhonghua Li, Zhaojun Wang LPMC and Department of Statistics, School of Mathematical Sciences,

More information

arxiv: v1 [stat.me] 14 Jan 2019

arxiv: v1 [stat.me] 14 Jan 2019 arxiv:1901.04443v1 [stat.me] 14 Jan 2019 An Approach to Statistical Process Control that is New, Nonparametric, Simple, and Powerful W.J. Conover, Texas Tech University, Lubbock, Texas V. G. Tercero-Gómez,Tecnológico

More information

Directionally Sensitive Multivariate Statistical Process Control Methods

Directionally Sensitive Multivariate Statistical Process Control Methods Directionally Sensitive Multivariate Statistical Process Control Methods Ronald D. Fricker, Jr. Naval Postgraduate School October 5, 2005 Abstract In this paper we develop two directionally sensitive statistical

More information

The Non-Central Chi-Square Chart with Double Sampling

The Non-Central Chi-Square Chart with Double Sampling Brazilian Journal of Operations & Production Management Volume, Number, 5, pp. 1-37 1 The Non-Central Chi-Square Chart with Double Sampling Antonio F. B. Costa Departamento de Produção, UNESP Guaratinguetá,

More information

Variations in a manufacturing process can be categorized into common cause and special cause variations. In the presence of

Variations in a manufacturing process can be categorized into common cause and special cause variations. In the presence of Research Article (wileyonlinelibrary.com) DOI: 10.1002/qre.1514 Published online in Wiley Online Library Memory-Type Control Charts for Monitoring the Process Dispersion Nasir Abbas, a * Muhammad Riaz

More information

A Multivariate EWMA Control Chart for Skewed Populations using Weighted Variance Method

A Multivariate EWMA Control Chart for Skewed Populations using Weighted Variance Method OPEN ACCESS Int. Res. J. of Science & Engineering, 04; Vol. (6): 9-0 ISSN: 3-005 RESEARCH ARTICLE A Multivariate EMA Control Chart for Skewed Populations using eighted Variance Method Atta AMA *, Shoraim

More information

Distribution-Free Monitoring of Univariate Processes. Peihua Qiu 1 and Zhonghua Li 1,2. Abstract

Distribution-Free Monitoring of Univariate Processes. Peihua Qiu 1 and Zhonghua Li 1,2. Abstract Distribution-Free Monitoring of Univariate Processes Peihua Qiu 1 and Zhonghua Li 1,2 1 School of Statistics, University of Minnesota, USA 2 LPMC and Department of Statistics, Nankai University, China

More information

Multivariate Process Control Chart for Controlling the False Discovery Rate

Multivariate Process Control Chart for Controlling the False Discovery Rate Industrial Engineering & Management Systems Vol, No 4, December 0, pp.385-389 ISSN 598-748 EISSN 34-6473 http://dx.doi.org/0.73/iems.0..4.385 0 KIIE Multivariate Process Control Chart for Controlling e

More information

Monitoring General Linear Profiles Using Multivariate EWMA schemes

Monitoring General Linear Profiles Using Multivariate EWMA schemes Monitoring General Linear Profiles Using Multivariate EWMA schemes Changliang Zou Department of Statistics School of Mathematical Sciences Nankai University Tianjian, PR China Fugee Tsung Department of

More information

Zero-Inflated Models in Statistical Process Control

Zero-Inflated Models in Statistical Process Control Chapter 6 Zero-Inflated Models in Statistical Process Control 6.0 Introduction In statistical process control Poisson distribution and binomial distribution play important role. There are situations wherein

More information

Analysis and Design of One- and Two-Sided Cusum Charts with Known and Estimated Parameters

Analysis and Design of One- and Two-Sided Cusum Charts with Known and Estimated Parameters Georgia Southern University Digital Commons@Georgia Southern Electronic Theses & Dissertations Graduate Studies, Jack N. Averitt College of Spring 2007 Analysis and Design of One- and Two-Sided Cusum Charts

More information

Design and Implementation of CUSUM Exceedance Control Charts for Unknown Location

Design and Implementation of CUSUM Exceedance Control Charts for Unknown Location Design and Implementation of CUSUM Exceedance Control Charts for Unknown Location MARIEN A. GRAHAM Department of Statistics University of Pretoria South Africa marien.graham@up.ac.za S. CHAKRABORTI Department

More information

Faculty of Science and Technology MASTER S THESIS

Faculty of Science and Technology MASTER S THESIS Faculty of Science and Technology MASTER S THESIS Study program/ Specialization: Spring semester, 20... Open / Restricted access Writer: Faculty supervisor: (Writer s signature) External supervisor(s):

More information

Optimal SPRT and CUSUM Procedures using Compressed Limit Gauges

Optimal SPRT and CUSUM Procedures using Compressed Limit Gauges Optimal SPRT and CUSUM Procedures using Compressed Limit Gauges P. Lee Geyer Stefan H. Steiner 1 Faculty of Business McMaster University Hamilton, Ontario L8S 4M4 Canada Dept. of Statistics and Actuarial

More information

THE CUSUM MEDIAN CHART FOR KNOWN AND ESTIMATED PARAMETERS

THE CUSUM MEDIAN CHART FOR KNOWN AND ESTIMATED PARAMETERS THE CUSUM MEDIAN CHART FOR KNOWN AND ESTIMATED PARAMETERS Authors: Philippe Castagliola Université de Nantes & LS2N UMR CNRS 6004, Nantes, France (philippe.castagliola@univ-nantes.fr) Fernanda Otilia Figueiredo

More information

On the performance of Shewhart-type synthetic and runs-rules charts combined with an chart

On the performance of Shewhart-type synthetic and runs-rules charts combined with an chart On the performance of Shewhart-type synthetic and runs-rules charts combined with an chart Sandile Charles Shongwe and Marien Alet Graham Department of Statistics University of Pretoria South Africa Abstract

More information

COMPARISON OF MCUSUM AND GENERALIZED VARIANCE S MULTIVARIATE CONTROL CHART PROCEDURE WITH INDUSTRIAL APPLICATION

COMPARISON OF MCUSUM AND GENERALIZED VARIANCE S MULTIVARIATE CONTROL CHART PROCEDURE WITH INDUSTRIAL APPLICATION Journal of Statistics: Advances in Theory and Applications Volume 8, Number, 07, Pages 03-4 Available at http://scientificadvances.co.in DOI: http://dx.doi.org/0.864/jsata_700889 COMPARISON OF MCUSUM AND

More information

Module B1: Multivariate Process Control

Module B1: Multivariate Process Control Module B1: Multivariate Process Control Prof. Fugee Tsung Hong Kong University of Science and Technology Quality Lab: http://qlab.ielm.ust.hk I. Multivariate Shewhart chart WHY MULTIVARIATE PROCESS CONTROL

More information

CONTROL charts are widely used in production processes

CONTROL charts are widely used in production processes 214 IEEE TRANSACTIONS ON SEMICONDUCTOR MANUFACTURING, VOL. 12, NO. 2, MAY 1999 Control Charts for Random and Fixed Components of Variation in the Case of Fixed Wafer Locations and Measurement Positions

More information

ON CONSTRUCTING T CONTROL CHARTS FOR RETROSPECTIVE EXAMINATION. Gunabushanam Nedumaran Oracle Corporation 1133 Esters Road #602 Irving, TX 75061

ON CONSTRUCTING T CONTROL CHARTS FOR RETROSPECTIVE EXAMINATION. Gunabushanam Nedumaran Oracle Corporation 1133 Esters Road #602 Irving, TX 75061 ON CONSTRUCTING T CONTROL CHARTS FOR RETROSPECTIVE EXAMINATION Gunabushanam Nedumaran Oracle Corporation 33 Esters Road #60 Irving, TX 7506 Joseph J. Pignatiello, Jr. FAMU-FSU College of Engineering Florida

More information

A NONLINEAR FILTER CONTROL CHART FOR DETECTING DYNAMIC CHANGES

A NONLINEAR FILTER CONTROL CHART FOR DETECTING DYNAMIC CHANGES Statistica Sinica 20 (2010), 1077-1096 A NONLINEAR FILTER CONTROL CHART FOR DETECTING DYNAMIC CHANGES Dong Han 1, Fugee Tsung 2, Yanting Li 1 and Kaibo Wang 3 1 Shanghai Jiao Tong University, 2 Hong Kong

More information

CONTROL CHARTS FOR THE GENERALIZED POISSON PROCESS WITH UNDER-DISPERSION

CONTROL CHARTS FOR THE GENERALIZED POISSON PROCESS WITH UNDER-DISPERSION International J. of Math. Sci. & Engg. Appls. (IJMSEA) ISSN 0973-9424, Vol. 10 No. III (December, 2016), pp. 173-181 CONTROL CHARTS FOR THE GENERALIZED POISSON PROCESS WITH UNDER-DISPERSION NARUNCHARA

More information

Quality Control & Statistical Process Control (SPC)

Quality Control & Statistical Process Control (SPC) Quality Control & Statistical Process Control (SPC) DR. RON FRICKER PROFESSOR & HEAD, DEPARTMENT OF STATISTICS DATAWORKS CONFERENCE, MARCH 22, 2018 Agenda Some Terminology & Background SPC Methods & Philosophy

More information

Monitoring aggregated Poisson data with probability control limits

Monitoring aggregated Poisson data with probability control limits XXV Simposio Internacional de Estadística 2015 Armenia, Colombia, 5, 6, 7 y 8 de Agosto de 2015 Monitoring aggregated Poisson data with probability control limits Victor Hugo Morales Ospina 1,a, José Alberto

More information

Optimal Cusum Control Chart for Censored Reliability Data with Log-logistic Distribution

Optimal Cusum Control Chart for Censored Reliability Data with Log-logistic Distribution CMST 21(4) 221-227 (2015) DOI:10.12921/cmst.2015.21.04.006 Optimal Cusum Control Chart for Censored Reliability Data with Log-logistic Distribution B. Sadeghpour Gildeh, M. Taghizadeh Ashkavaey Department

More information

MCUSUM CONTROL CHART PROCEDURE: MONITORING THE PROCESS MEAN WITH APPLICATION

MCUSUM CONTROL CHART PROCEDURE: MONITORING THE PROCESS MEAN WITH APPLICATION Journal of Statistics: Advances in Theory and Applications Volume 6, Number, 206, Pages 05-32 Available at http://scientificadvances.co.in DOI: http://dx.doi.org/0.8642/jsata_700272 MCUSUM CONTROL CHART

More information

AUTOCORRELATED PROCESS MONITORING USING TRIGGERED CUSCORE CHARTS

AUTOCORRELATED PROCESS MONITORING USING TRIGGERED CUSCORE CHARTS QUALITY AND RELIABILITY ENGINEERING INTERNATIONAL Qual. Reliab. Engng. Int. 22; 8: 4 42 (DOI:.2/qre.492) AUTOCORRELATED PROCESS MONITORING USING TRIGGERED CUSCORE CHARTS LIANJIE SHU, DANIEL W. APLEY 2

More information

2.830J / 6.780J / ESD.63J Control of Manufacturing Processes (SMA 6303) Spring 2008

2.830J / 6.780J / ESD.63J Control of Manufacturing Processes (SMA 6303) Spring 2008 MIT OpenCourseWare http://ocw.mit.edu 2.830J / 6.780J / ESD.63J Control of Processes (SMA 6303) Spring 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/term

More information

Effect of sample size on the performance of Shewhart control charts

Effect of sample size on the performance of Shewhart control charts DOI 10.1007/s00170-016-9412-8 ORIGINAL ARTICLE Effect of sample size on the performance of Shewhart control s Salah Haridy 1,2 & Ahmed Maged 1 & Saleh Kaytbay 1 & Sherif Araby 1,3 Received: 20 December

More information

Run sum control charts for the monitoring of process variability

Run sum control charts for the monitoring of process variability Quality Technology & Quantitative Management ISSN: (Print) 1684-3703 (Online) Journal homepage: http://www.tandfonline.com/loi/ttqm20 Run sum control charts for the monitoring of process variability Athanasios

More information

Control charts are used for monitoring the performance of a quality characteristic. They assist process

Control charts are used for monitoring the performance of a quality characteristic. They assist process QUALITY AND RELIABILITY ENGINEERING INTERNATIONAL Qual. Reliab. Engng. Int. 2009; 25:875 883 Published online 3 March 2009 in Wiley InterScience (www.interscience.wiley.com)..1007 Research Identifying

More information

A New Model-Free CuSum Procedure for Autocorrelated Processes

A New Model-Free CuSum Procedure for Autocorrelated Processes A New Model-Free CuSum Procedure for Autocorrelated Processes Seong-Hee Kim, Christos Alexopoulos, David Goldsman, and Kwok-Leung Tsui School of Industrial and Systems Engineering Georgia Institute of

More information

Approximation of Average Run Length of Moving Sum Algorithms Using Multivariate Probabilities

Approximation of Average Run Length of Moving Sum Algorithms Using Multivariate Probabilities Syracuse University SURFACE Electrical Engineering and Computer Science College of Engineering and Computer Science 3-1-2010 Approximation of Average Run Length of Moving Sum Algorithms Using Multivariate

More information

Control charts continue to play a transformative role in all walks of life in the 21st century. The mean and the variance of a

Control charts continue to play a transformative role in all walks of life in the 21st century. The mean and the variance of a Research Article (wileyonlinelibrary.com) DOI: 10.1002/qre.1249 Published online in Wiley Online Library A Distribution-free Control Chart for the Joint Monitoring of Location and Scale A. Mukherjee, a

More information

AN ANALYSIS OF SHEWHART QUALITY CONTROL CHARTS TO MONITOR BOTH THE MEAN AND VARIABILITY KEITH JACOB BARRS. A Research Project Submitted to the Faculty

AN ANALYSIS OF SHEWHART QUALITY CONTROL CHARTS TO MONITOR BOTH THE MEAN AND VARIABILITY KEITH JACOB BARRS. A Research Project Submitted to the Faculty AN ANALYSIS OF SHEWHART QUALITY CONTROL CHARTS TO MONITOR BOTH THE MEAN AND VARIABILITY by KEITH JACOB BARRS A Research Project Submitted to the Faculty of the College of Graduate Studies at Georgia Southern

More information

DIAGNOSIS OF BIVARIATE PROCESS VARIATION USING AN INTEGRATED MSPC-ANN SCHEME

DIAGNOSIS OF BIVARIATE PROCESS VARIATION USING AN INTEGRATED MSPC-ANN SCHEME DIAGNOSIS OF BIVARIATE PROCESS VARIATION USING AN INTEGRATED MSPC-ANN SCHEME Ibrahim Masood, Rasheed Majeed Ali, Nurul Adlihisam Mohd Solihin and Adel Muhsin Elewe Faculty of Mechanical and Manufacturing

More information

A Study on the Power Functions of the Shewhart X Chart via Monte Carlo Simulation

A Study on the Power Functions of the Shewhart X Chart via Monte Carlo Simulation A Study on the Power Functions of the Shewhart X Chart via Monte Carlo Simulation M.B.C. Khoo Abstract The Shewhart X control chart is used to monitor shifts in the process mean. However, it is less sensitive

More information

Evaluation of X and S charts when Standards Vary Randomly *Aamir Saghir Department of Statistics, UAJK Muzaffrabad, Pakistan.

Evaluation of X and S charts when Standards Vary Randomly *Aamir Saghir Department of Statistics, UAJK Muzaffrabad, Pakistan. Evaluation of and charts when tandards Vary Randomly *Aamir aghir Department of tatistics, UAJK Muzaffrabad, Pakistan. E-mail: aamirstat@yahoo.com. Abstract The study proposes control limits for and charts

More information

GOTEBORG UNIVERSITY. Department of Statistics ON PERFORMANCE OF METHODS FOR STATISTICAL SURVEILLANCE

GOTEBORG UNIVERSITY. Department of Statistics ON PERFORMANCE OF METHODS FOR STATISTICAL SURVEILLANCE GOTEBORG UNIVERSITY Department of Statistics RESEARCH REPORT 1994:7 ISSN 0349-8034 ON PERFORMANCE OF METHODS FOR STATISTICAL SURVEILLANCE by o Goran Akermo Statistiska institutionen Goteborgs Universitet

More information

On the Distribution of Hotelling s T 2 Statistic Based on the Successive Differences Covariance Matrix Estimator

On the Distribution of Hotelling s T 2 Statistic Based on the Successive Differences Covariance Matrix Estimator On the Distribution of Hotelling s T 2 Statistic Based on the Successive Differences Covariance Matrix Estimator JAMES D. WILLIAMS GE Global Research, Niskayuna, NY 12309 WILLIAM H. WOODALL and JEFFREY

More information

APPLICATION OF Q CHARTS FOR SHORT-RUN AUTOCORRELATED DATA

APPLICATION OF Q CHARTS FOR SHORT-RUN AUTOCORRELATED DATA International Journal of Innovative Computing, Information and Control ICIC International c 2013 ISSN 1349-4198 Volume 9, Number 9, September 2013 pp. 3667 3676 APPLICATION OF Q CHARTS FOR SHORT-RUN AUTOCORRELATED

More information

Monitoring Censored Lifetime Data with a Weighted-Likelihood Scheme

Monitoring Censored Lifetime Data with a Weighted-Likelihood Scheme Monitoring Censored Lifetime Data with a Weighted-Likelihood Scheme Chi Zhang, 1,2 Fugee Tsung, 2 Dongdong Xiang 1 1 School of Statistics, East China Normal University, Shanghai, China 2 Department of

More information

The Changepoint Model for Statistical Process Control

The Changepoint Model for Statistical Process Control The Changepoint Model for Statistical Process Control DOUGLAS M. HAWKINS and PEIHUA QIU University of Minnesota, Minneapolis, MN 55455 CHANG WOOK KANG Hanyang University, Seoul, Korea Statistical process

More information

Detection and Diagnosis of Unknown Abrupt Changes Using CUSUM Multi-Chart Schemes

Detection and Diagnosis of Unknown Abrupt Changes Using CUSUM Multi-Chart Schemes Sequential Analysis, 26: 225 249, 2007 Copyright Taylor & Francis Group, LLC ISSN: 0747-4946 print/532-476 online DOI: 0.080/0747494070404765 Detection Diagnosis of Unknown Abrupt Changes Using CUSUM Multi-Chart

More information

CONTROL CHARTS FOR MULTIVARIATE NONLINEAR TIME SERIES

CONTROL CHARTS FOR MULTIVARIATE NONLINEAR TIME SERIES REVSTAT Statistical Journal Volume 13, Number, June 015, 131 144 CONTROL CHARTS FOR MULTIVARIATE NONLINEAR TIME SERIES Authors: Robert Garthoff Department of Statistics, European University, Große Scharrnstr.

More information

Univariate and Multivariate Surveillance Methods for Detecting Increases in Incidence Rates

Univariate and Multivariate Surveillance Methods for Detecting Increases in Incidence Rates Univariate and Multivariate Surveillance Methods for Detecting Increases in Incidence Rates Michael D. Joner, Jr. Dissertation submitted to the Faculty of the Virginia Polytechnic Institute and State University

More information

Monitoring Expense Report Errors: Control Charts Under Independence and Dependence. Darren Williams. (Under the direction of Dr.

Monitoring Expense Report Errors: Control Charts Under Independence and Dependence. Darren Williams. (Under the direction of Dr. Monitoring Expense Report Errors: Control Charts Under Independence and Dependence by Darren Williams (Under the direction of Dr. Lynne Seymour) Abstract Control charts were devised to evaluate offices

More information

On ARL-unbiased c-charts for i.i.d. and INAR(1) Poisson counts

On ARL-unbiased c-charts for i.i.d. and INAR(1) Poisson counts On ARL-unbiased c-charts for iid and INAR(1) Poisson counts Manuel Cabral Morais (1) with Sofia Paulino (2) and Sven Knoth (3) (1) Department of Mathematics & CEMAT IST, ULisboa, Portugal (2) IST, ULisboa,

More information

Performance Analysis of Queue Length Monitoring of M/G/1 Systems

Performance Analysis of Queue Length Monitoring of M/G/1 Systems Performance Analysis of Queue Length Monitoring of M/G/1 Systems Nan Chen, 1 Yuan Yuan, 2 Shiyu Zhou 2 1 Department of Industrial and Systems Engineering, National University of Singapore, Singapore 2

More information

Glossary. The ISI glossary of statistical terms provides definitions in a number of different languages:

Glossary. The ISI glossary of statistical terms provides definitions in a number of different languages: Glossary The ISI glossary of statistical terms provides definitions in a number of different languages: http://isi.cbs.nl/glossary/index.htm Adjusted r 2 Adjusted R squared measures the proportion of the

More information

Chart for Monitoring Univariate Autocorrelated Processes

Chart for Monitoring Univariate Autocorrelated Processes The Autoregressive T 2 Chart for Monitoring Univariate Autocorrelated Processes DANIEL W APLEY Texas A&M University, College Station, TX 77843-33 FUGEE TSUNG Hong Kong University of Science and Technology,

More information

A multivariate exponentially weighted moving average control chart for monitoring process variability

A multivariate exponentially weighted moving average control chart for monitoring process variability Journal of Applied Statistics, Vol. 30, No. 5, 2003, 507 536 A multivariate exponentially weighted moving average control chart for monitoring process variability ARTHUR B. YEH 1, DENNIS K. J. LIN 2, HONGHONG

More information

SCIENCE & TECHNOLOGY

SCIENCE & TECHNOLOGY Pertanika J. Sci. & Technol. 24 (1): 177-189 (2016) SCIENCE & TECHNOLOGY Journal homepage: http://www.pertanika.upm.edu.my/ A Comparative Study of the Group Runs and Side Sensitive Group Runs Control Charts

More information

A New Bootstrap Based Algorithm for Hotelling s T2 Multivariate Control Chart

A New Bootstrap Based Algorithm for Hotelling s T2 Multivariate Control Chart Journal of Sciences, Islamic Republic of Iran 7(3): 69-78 (16) University of Tehran, ISSN 16-14 http://jsciences.ut.ac.ir A New Bootstrap Based Algorithm for Hotelling s T Multivariate Control Chart A.

More information

Control Charts for Monitoring the Zero-Inflated Generalized Poisson Processes

Control Charts for Monitoring the Zero-Inflated Generalized Poisson Processes Thai Journal of Mathematics Volume 11 (2013) Number 1 : 237 249 http://thaijmath.in.cmu.ac.th ISSN 1686-0209 Control Charts for Monitoring the Zero-Inflated Generalized Poisson Processes Narunchara Katemee

More information

A STUDY ON IMPROVING THE PERFORMANCE OF CONTROL CHARTS UNDER NON-NORMAL DISTRIBUTIONS SUN TINGTING

A STUDY ON IMPROVING THE PERFORMANCE OF CONTROL CHARTS UNDER NON-NORMAL DISTRIBUTIONS SUN TINGTING A STUDY ON IMPROVING THE PERFORMANCE OF CONTROL CHARTS UNDER NON-NORMAL DISTRIBUTIONS SUN TINGTING NATIONAL UNIVERSITY OF SINGAPORE 004 A STUDY ON IMPROVING THE PERFORMANCE OF CONTROL CHARTS UNDER NON-NORMAL

More information

Contents. Preface to Second Edition Preface to First Edition Abbreviations PART I PRINCIPLES OF STATISTICAL THINKING AND ANALYSIS 1

Contents. Preface to Second Edition Preface to First Edition Abbreviations PART I PRINCIPLES OF STATISTICAL THINKING AND ANALYSIS 1 Contents Preface to Second Edition Preface to First Edition Abbreviations xv xvii xix PART I PRINCIPLES OF STATISTICAL THINKING AND ANALYSIS 1 1 The Role of Statistical Methods in Modern Industry and Services

More information

Simultaneous Prediction Intervals for the (Log)- Location-Scale Family of Distributions

Simultaneous Prediction Intervals for the (Log)- Location-Scale Family of Distributions Statistics Preprints Statistics 10-2014 Simultaneous Prediction Intervals for the (Log)- Location-Scale Family of Distributions Yimeng Xie Virginia Tech Yili Hong Virginia Tech Luis A. Escobar Louisiana

More information

Likelihood-Based EWMA Charts for Monitoring Poisson Count Data with Time-Varying Sample Sizes

Likelihood-Based EWMA Charts for Monitoring Poisson Count Data with Time-Varying Sample Sizes Likelihood-Based EWMA Charts for Monitoring Poisson Count Data with Time-Varying Sample Sizes Qin Zhou 1,3, Changliang Zou 1, Zhaojun Wang 1 and Wei Jiang 2 1 LPMC and Department of Statistics, School

More information

Rejoinder. 1 Phase I and Phase II Profile Monitoring. Peihua Qiu 1, Changliang Zou 2 and Zhaojun Wang 2

Rejoinder. 1 Phase I and Phase II Profile Monitoring. Peihua Qiu 1, Changliang Zou 2 and Zhaojun Wang 2 Rejoinder Peihua Qiu 1, Changliang Zou 2 and Zhaojun Wang 2 1 School of Statistics, University of Minnesota 2 LPMC and Department of Statistics, Nankai University, China We thank the editor Professor David

More information

Financial Econometrics

Financial Econometrics Financial Econometrics Nonlinear time series analysis Gerald P. Dwyer Trinity College, Dublin January 2016 Outline 1 Nonlinearity Does nonlinearity matter? Nonlinear models Tests for nonlinearity Forecasting

More information

Approximating the step change point of the process fraction nonconforming using genetic algorithm to optimize the likelihood function

Approximating the step change point of the process fraction nonconforming using genetic algorithm to optimize the likelihood function Journal of Industrial and Systems Engineering Vol. 7, No., pp 8-28 Autumn 204 Approximating the step change point of the process fraction nonconforming using genetic algorithm to optimize the likelihood

More information

Stochastic Inequalities for the Run Length of the EWMA Chart for Long-Memory Processes

Stochastic Inequalities for the Run Length of the EWMA Chart for Long-Memory Processes Stochastic Inequalities for the Run Length of the EWMA Chart for Long-Memory Processes Authors: Yarema Okhrin Department of Statistics, University of Augsburg, Germany (yarema.okhrin@wiwi.uni-augsburg.de)

More information

2.830J / 6.780J / ESD.63J Control of Manufacturing Processes (SMA 6303) Spring 2008

2.830J / 6.780J / ESD.63J Control of Manufacturing Processes (SMA 6303) Spring 2008 MIT OpenCourseWare http://ocw.mit.edu 2.830J / 6.780J / ESD.63J Control of Processes (SMA 6303) Spring 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

More information

Application and Use of Multivariate Control Charts In a BTA Deep Hole Drilling Process

Application and Use of Multivariate Control Charts In a BTA Deep Hole Drilling Process Application and Use of Multivariate Control Charts In a BTA Deep Hole Drilling Process Amor Messaoud, Winfied Theis, Claus Weihs, and Franz Hering Fachbereich Statistik, Universität Dortmund, 44221 Dortmund,

More information

A process capability index for discrete processes

A process capability index for discrete processes Journal of Statistical Computation and Simulation Vol. 75, No. 3, March 2005, 175 187 A process capability index for discrete processes MICHAEL PERAKIS and EVDOKIA XEKALAKI* Department of Statistics, Athens

More information

Robust control charts for time series data

Robust control charts for time series data Robust control charts for time series data Christophe Croux K.U. Leuven & Tilburg University Sarah Gelper Erasmus University Rotterdam Koen Mahieu K.U. Leuven Abstract This article presents a control chart

More information

Practical Statistics for the Analytical Scientist Table of Contents

Practical Statistics for the Analytical Scientist Table of Contents Practical Statistics for the Analytical Scientist Table of Contents Chapter 1 Introduction - Choosing the Correct Statistics 1.1 Introduction 1.2 Choosing the Right Statistical Procedures 1.2.1 Planning

More information

Two widely used approaches for monitoring and improving the quality of the output of a process are statistical process control

Two widely used approaches for monitoring and improving the quality of the output of a process are statistical process control Research Article (www.interscience.wiley.com) DOI:.2/qre.45 Published online 8 July 9 in Wiley InterScience CUSUM Charts for Detecting Special Causes in Integrated Process Control Marion R. Reynolds Jr

More information

MONITORING BIVARIATE PROCESSES WITH A SYNTHETIC CONTROL CHART BASED ON SAMPLE RANGES

MONITORING BIVARIATE PROCESSES WITH A SYNTHETIC CONTROL CHART BASED ON SAMPLE RANGES Blumenau-SC, 27 a 3 de Agosto de 217. MONITORING BIVARIATE PROCESSES WITH A SYNTHETIC CONTROL CHART BASED ON SAMPLE RANGES Marcela A. G. Machado São Paulo State University (UNESP) Departamento de Produção,

More information

Surveillance of Infectious Disease Data using Cumulative Sum Methods

Surveillance of Infectious Disease Data using Cumulative Sum Methods Surveillance of Infectious Disease Data using Cumulative Sum Methods 1 Michael Höhle 2 Leonhard Held 1 1 Institute of Social and Preventive Medicine University of Zurich 2 Department of Statistics University

More information

Section II: Assessing Chart Performance. (Jim Benneyan)

Section II: Assessing Chart Performance. (Jim Benneyan) Section II: Assessing Chart Performance (Jim Benneyan) 1 Learning Objectives Understand concepts of chart performance Two types of errors o Type 1: Call an in-control process out-of-control o Type 2: Call

More information

A Control Chart for Time Truncated Life Tests Using Exponentiated Half Logistic Distribution

A Control Chart for Time Truncated Life Tests Using Exponentiated Half Logistic Distribution Appl. Math. Inf. Sci. 12, No. 1, 125-131 (2018 125 Applied Mathematics & Information Sciences An International Journal http://dx.doi.org/10.18576/amis/120111 A Control Chart for Time Truncated Life Tests

More information

Research Article Robust Multivariate Control Charts to Detect Small Shifts in Mean

Research Article Robust Multivariate Control Charts to Detect Small Shifts in Mean Mathematical Problems in Engineering Volume 011, Article ID 93463, 19 pages doi:.1155/011/93463 Research Article Robust Multivariate Control Charts to Detect Small Shifts in Mean Habshah Midi 1, and Ashkan

More information

CHAPTER-II Control Charts for Fraction Nonconforming using m-of-m Runs Rules

CHAPTER-II Control Charts for Fraction Nonconforming using m-of-m Runs Rules CHAPTER-II Control Charts for Fraction Nonconforming using m-of-m Runs Rules. Introduction: The is widely used in industry to monitor the number of fraction nonconforming units. A nonconforming unit is

More information

Control Charts Based on Alternative Hypotheses

Control Charts Based on Alternative Hypotheses Control Charts Based on Alternative Hypotheses A. Di Bucchianico, M. Hušková (Prague), P. Klášterecky (Prague), W.R. van Zwet (Leiden) Dortmund, January 11, 2005 1/48 Goals of this talk introduce hypothesis

More information

The Shiryaev-Roberts Changepoint Detection Procedure in Retrospect - Theory and Practice

The Shiryaev-Roberts Changepoint Detection Procedure in Retrospect - Theory and Practice The Shiryaev-Roberts Changepoint Detection Procedure in Retrospect - Theory and Practice Department of Statistics The Hebrew University of Jerusalem Mount Scopus 91905 Jerusalem, Israel msmp@mscc.huji.ac.il

More information