An algorithm for calculating the lower confidence bounds
of C
PU
and C
PL
with application to low-drop-out
linear regulators
W.L. Pearn
a,*, Ming-Hung Shu
ba
Department of Industrial Engineering & Management, National Chiao Tung University, 1001 Ta Hsueh Road, Hsin Chu 30050 Taiwan, ROC
b
Department of Management Science, Chinese Military Academy, Taiwan, ROC Received 3 July 2002; received in revised form 13 August 2002
Abstract
In assessing the performance of normal stable manufacturing processes with one-sided specification limits, process capability indices CPUand CPLhave been widely used to measure the process capability. The purpose of this paper is to
develop an algorithm to compute the lower confidence bounds on CPUand CPLusing the UMVUEs of CPUand CPL. The
lower confidence bound presents a measure on the minimum capability of the process based on the sample data. We also provide tables for the engineers/practitioners to use in measuring their processes. A real-world example taken from a microelectronics device manufacturing process is investigated to illustrate the applicability of the algorithm. Imple-mentation of the existing statistical theory for capability assessment fills the gap between the theoretical development and the in-plant applications.
Ó 2003 Elsevier Science Ltd. All rights reserved.
1. Introduction
Process capability indices are used to measure the capability of a process to reproduce items within the specified tolerance preset by the product designers or customers. Several capability indices, Cp, CPU, CPL, Cpk,
and Cpm, have been developed for the manufacturing
industry (Kane [1], Chan et al. [2], Pearn et al. [3]). Those indices essentially compare the specification tol-erance range with the actual production toltol-erance range, which have been defined as
Cp¼ USL LSL 6r ; CPU¼ USL l 3r ; CPL¼ l LSL 3r ; Cpk¼ min USL l 3r ; l LSL 3r ; Cpm¼ USL LSL 6 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r2þ ðl T Þ2 q ;
where USL is the upper specification limit, LSL is the lower specification limit, l is the process mean, r is the process standard deviation (overall process variation), and T is the preset target value.
The indices Cp, Cpk, and Cpm are appropriate for
processes with two-sided specification limits. Some procedures have been developed based on these indices (Cheng [4], Pearn and Chen [5], Pearn and Chen [6]) for the practitioners to use in making decision on whether their processes meet the preset capability requirement. On the other hand, the indices CPUand CPLare designed
specifically for processes with one-sided specification limit. Pearn and Chen [7] considered the indices CPUand
CPL, and obtained their uniformly minimum-variance
www.elsevier.com/locate/microrel
*
Corresponding author. Tel.: 714261; fax: +886-35-722392.
E-mail address:roller@cc.nctu.edu.tw(W.L. Pearn).
0026-2714/03/$ - see front matterÓ 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0026-2714(02)00264-0
unbiased estimators (UMVUEs). Lin and Pearn [8] de-veloped some efficient SAS/Maple computer programs for calculating the critical values and the p-values using those UMVUEs in capability testing, particularly for normal processes.
Critical values are used for making decisions in ca-pability testing with designated type-I error a, the risk of misjudging an incapable process (H0: CPU6C) as a
ca-pable one (H1: CPU> C). The p-values are used for
making decisions in capability testing, which presents the actual risk of misjudging an incapable process (H0: CPU6C) as a capable one (H1: CPU> C). Thus, if
p <athen we reject the null hypothesis, and conclude that the process is capable with actual type-I error p (rather than a). Both approaches, the critical values and the p-values, do not convey any information regarding the minimal value (lower confidence bound) of the ac-tual process capability. The development of the lower confidence bound on the actual process capability is essential. The lower confidence bound not only gives us a clue on the minimal level of the actual performance of the process which is tightly related to the nonconform-ing units (product units fallout the specification limit USL or LSL), but is also useful in making decisions for capability testing. Table 1 displays some capability val-ues of CPU (or CPL) and the corresponding
noncon-forming unit (in ppm).
Montgomery [9] recommended some minimum quality requirements on CPUand CPL, shown in Table 2,
for specific process types which must run under some designated capability conditions. Therefore, it would be desirable to determine a bound which practitioners
would be expected to find the true value of the process capability no less than the bound value with certain level of confidence.
2. Estimations of CPUand CPL
A random sample is taken from a stable process. The conventional estimates of l and r are:
X ¼X n i¼1 Xi=n; S¼ ðn " 1Þ1X n i¼1 ðXi X Þ 2 #1=2 :
Thus, we may consider the natural estimators to esti-mate the indices CPUand CPLwhich can be defined as the
following: b C CPU¼ USL X 3S b C CPL¼ X LSL 3S :
Chou and Owen [10] showed that under normality assumption the estimator bCCPUand bCCPLare distributed as
ð3pffiffiffinÞ1Tn1ðdÞ, where Tn1ðdÞ is distributed as the
noncentral t distribution with n 1 degrees of freedom and noncentrality parameter d¼ 3p bCCffiffiffin PU and d¼
3p bCCffiffiffin PL, respectively. Both estimators are biased (Pearn
and Chen [7]). But, Pearn and Chen [7] showed that by adding the well-known correction factor
bn1¼ 2 n 1 1=2 C n 1 2 C n 2 2 1
to bCCPU and bCCPL, the unbiased estimators bn1CCbPU and
bn1CCbPL can be obtained which have been denoted
as eCCPU and eCCPL. Thus, we have Eð eCCPUÞ ¼ CPU and
Eð eCCPLÞ ¼ CPL. S ince bn1<1 for n > 2, then Varð eCCPUÞ <
Varð bCCPUÞ and Varð eCCPLÞ < Varð bCCPLÞ. Pearn and Chen
[7] further showed that eCCPU and eCCPL are the UMVUEs
of CPU and CPL, respectively. The probability density
function of eCCPU and eCCPL can be expressed as
fðxÞ ¼3 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n=ðn 1Þ p 2n=2 bn1 ffiffiffip p C½ðn 1Þ=2 Z 1 0 yðn2Þ=2exp ( 1 2 y " þ 3x ffiffiffiffiffiny p bn1 ffiffiffiffiffiffiffiffiffiffiffi n 1 p d 2#) dy:
3. Lower confidence bound on CPU(or CPL)
Chou et al. [11] established the lower confidence bounds on CPUand CPLbased on the natural estimator
b C
CPU and bCCPL. We note that those two estimators are
Table 1
Various CPUvalues and the corresponding nonconformities
CPU ppm 0.5 66 807 0.7 17 864 0.9 3467 1.1 484 1.3 48.10 1.5 3.40 1.7 0.1698 1.9 0.0060 2.1 0.0001488 2.3 0.0000026 Table 2
Some minimum capability requirements on CPUand CPL
CPU(or CPL) Process types
1.25 Existing processes
1.45 New processes, or existing processes on safety, strength, or critical parameters 1.6 New processes on safety, strength, or critical
Table 3
Lower confidence bounds CU of CPU for eCCPU¼ 0:7ð0:1Þ1:8, n ¼ 5ð5Þ200, c ¼ 0:95 (Panel A) and eCCPU¼ 1:9ð0:1Þ3:0, n ¼ 5ð5Þ200,
c¼ 0:95 (Panel B) n eCCPU 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 Panel A 5 0.304 0.364 0.423 0.481 0.538 0.594 0.650 0.705 0.760 0.815 0.870 0.924 10 0.414 0.486 0.557 0.627 0.697 0.766 0.835 0.943 0.972 1.040 1.108 1.175 15 0.465 0.542 0.618 0.693 0.768 0.843 0.917 0.991 1.065 1.139 1.213 1.286 20 0.495 0.575 0.654 0.733 0.811 0.889 0.966 1.044 1.121 1.198 1.275 1.352 25 0.516 0.598 0.679 0.760 0.840 0.920 1.000 1.080 1.159 1.239 1.318 1.397 30 0.531 0.614 0.697 0.780 0.862 0.944 1.025 1.107 1.188 1.269 1.350 1.431 35 0.543 0.628 0.712 0.795 0.879 0.962 1.045 1.127 1.210 1.293 1.375 1.457 40 0.553 0.638 0.723 0.808 0.892 0.977 1.061 1.144 1.228 1.312 1.395 1.478 45 0.561 0.647 0.733 0.819 0.904 0.989 1.074 1.158 1.243 1.327 1.412 1.496 50 0.568 0.655 0.741 0.828 0.914 0.999 1.085 1.170 1.256 1.341 1.426 1.511 55 0.574 0.661 0.748 0.835 0.922 1.008 1.094 1.181 1.267 1.352 1.438 1.524 60 0.579 0.667 0.755 0.842 0.929 1.016 1.103 1.190 1.276 1.363 1.449 1.535 65 0.583 0.672 0.760 0.848 0.936 1.023 1.110 1.197 1.285 1.371 1.458 1.545 70 0.588 0.677 0.765 0.853 0.941 1.029 1.117 1.205 1.292 1.379 1.467 1.554 75 0.591 0.681 0.770 0.858 0.947 1.035 1.123 1.211 1.299 1.387 1.474 1.562 80 0.595 0.684 0.774 0.862 0.951 1.040 1.128 1.217 1.305 1.393 1.481 1.569 85 0.598 0.688 0.777 0.866 0.956 1.044 1.133 1.223 1.311 1.399 1.488 1.577 90 0.600 0.691 0.781 0.870 0.959 1.049 1.138 1.228 1.316 1.405 1.494 1.583 95 0.603 0.693 0.784 0.873 0.963 1.053 1.142 1.233 1.321 1.411 1.500 1.589 100 0.605 0.696 0.786 0.877 0.966 1.056 1.146 1.238 1.326 1.416 1.505 1.595 105 0.608 0.699 0.789 0.879 0.970 1.060 1.150 1.242 1.330 1.420 1.510 1.601 110 0.610 0.701 0.792 0.882 0.972 1.063 1.153 1.246 1.334 1.424 1.515 1.606 115 0.612 0.703 0.794 0.855 0.975 1.066 1.156 1.249 1.338 1.428 1.519 1.610 120 0.613 0.705 0.796 0.887 0.978 1.068 1.159 1.253 1.341 1.432 1.523 1.614 125 0.615 0.707 0.798 0.889 0.980 1.071 1.162 1.256 1.344 1.435 1.527 1.618 130 0.617 0.709 0.800 0.891 0.982 1.073 1.164 1.259 1.347 1.438 1.530 1.622 135 0.618 0.710 0.802 0.893 0.984 1.075 1.166 1.261 1.350 1.441 1.533 1.625 140 0.620 0.712 0.804 0.895 0.986 1.077 1.169 1.264 1.352 1.444 1.536 1.628 145 0.621 0.713 0.805 0.897 0.988 1.079 1.171 1.266 1.355 1.446 1.539 1.631 150 0.622 0.715 0.807 0.899 0.990 1.081 1.173 1.268 1.357 1.449 1.541 1.634 155 0.624 0.716 0.808 0.900 0.992 1.083 1.175 1.271 1.359 1.451 1.543 1.636 160 0.625 0.717 0.810 0.902 0.994 1.085 1.177 1.273 1.361 1.453 1.546 1.639 165 0.626 0.719 0.811 0.903 0.995 1.087 1.178 1.275 1.363 1.456 1.548 1.641 170 0.627 0.720 0.812 0.905 0.997 1.088 1.180 1.277 1.365 1.458 1.550 1.643 175 0.628 0.721 0.814 0.906 0.998 1.090 1.182 1.279 1.367 1.460 1.552 1.645 180 0.629 0.722 0.815 0.908 1.000 1.092 1.184 1.280 1.369 1.462 1.554 1.648 185 0.630 0.723 0.816 0.909 1.001 1.093 1.185 1.282 1.371 1.464 1.556 1.650 190 0.631 0.724 0.817 0.910 1.002 1.094 1.187 1.284 1.373 1.465 1.558 1.652 195 0.632 0.725 0.818 0.911 1.004 1.096 1.188 1.286 1.374 1.467 1.560 1.653 200 0.632 0.726 0.819 0.912 1.005 1.097 1.189 1.287 1.376 1.469 1.562 1.655 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 Panel B 5 0.978 1.032 1.086 1.140 1.194 1.247 1.301 1.355 1.408 1.462 1.515 1.568 10 1.243 1.310 1.378 1.445 1.512 1.580 1.647 1.714 1.781 1.848 1.915 1.982 15 1.359 1.433 1.510 1.579 1.652 1.725 1.798 1.871 1.944 2.017 2.090 2.163 20 1.429 1.506 1.582 1.659 1.736 1.812 1.889 1.965 2.041 2.118 2.194 2.271 25 1.477 1.556 1.635 1.714 1.793 1.872 1.950 2.029 2.108 2.187 2.266 2.344 30 1.512 1.593 1.673 1.754 1.835 1.916 1.996 2.077 2.157 2.238 2.318 2.399 35 1.539 1.622 1.704 1.786 1.868 1.950 2.032 2.114 2.196 2.278 2.359 2.441 40 1.562 1.645 1.728 1.811 1.895 1.978 2.061 2.144 2.227 2.310 2.393 2.476 45 1.580 1.664 1.749 1.833 1.917 1.980 2.085 2.169 2.252 2.336 2.420 2.504 50 1.596 1.681 1.766 1.851 1.935 1.999 2.105 2.190 2.274 2.359 2.443 2.528 (continued on next page)
biased. In the following, we use eCCPU and eCCPL, the
UMVUEs of CPU and CPL, to obtain the lower
confi-dence bound on CPU and CPL.
Let USL¼ X þ k1S and LSL¼ X k2S, so k1¼
3 eCCPU=bn1¼ 3 bCCPU and k2¼ 3 eCCPL=bn1¼ 3 bCCPL. A
100c% lower confidence bound CU for CPU satisfies
PrðCPUP CUÞ ¼ c. It can be written as:
Pr USL l 3r P CU ¼ Pr Xþ k1S l 3r P CU ¼ Pr Z 3 ffiffiffi n p CU S=r P k1 ffiffiffin p ¼ Pr Z 3 ffiffiffi n p CU S=r P 3 eCCPU bn1 ffiffiffi n p ! ¼ PrðTn1ðd1Þ P t1Þ ¼ c;
and PrðTn1ðd1Þ 6 t1Þ ¼ 1 c. Similarly, a 100c% lower
confidence bound CLfor CPLsatisfies PrðCPLP CLÞ ¼ c.
It can be shown as PrðTn1ðd2Þ 6 t2Þ ¼ c, where Z is
distributed as Nð0; 1Þ, Tn1ðdÞ is the noncentral t
distri-bution with n 1 degrees of freedom and noncentrality parameter d, t1¼ k1 ffiffiffin p , t2¼ k2 ffiffiffin p , d1¼ 3 ffiffiffin p CU, and d2¼ 3 ffiffiffin p
CL. Thus, to obtain the lower confidence
bound (LCB), we may proceed as follows: Procedure of obtaining the LCB
(a) Determine the c level of confidence (normally set to 0.95) for the lower confidence bound.
(b) Calculate the value of estimator, eCCPU(or eCCPL), from
the sample data.
(c) Check the appropriate table listed in Table 3(a) and (b) and find the corresponding CU(or CL) based on
e C
CPU(or eCCPL) and n.
(d) Conclude that the true value of the process capabil-ity is no less than the CU(or CL) with 100c% level of
confidence.
4. An algorithm for calculating the lower confidence bound Based on the procedure described above, a Matlab algorithm, called LCB, is developed for computing the
Table 3 (continued) n CCePU 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 55 1.610 1.695 1.781 1.866 1.952 2.017 2.123 2.208 2.293 2.379 2.464 2.549 60 1.621 1.707 1.794 1.880 1.966 2.031 2.137 2.223 2.309 2.395 2.481 2.567 65 1.632 1.718 1.805 1.891 1.978 2.044 2.150 2.237 2.323 2.410 2.496 2.582 70 1.641 1.728 1.815 1.902 1.988 2.055 2.162 2.249 2.335 2.422 2.509 2.596 75 1.649 1.737 1.824 1.911 1.998 2.065 2.173 2.260 2.347 2.434 2.521 2.608 80 1.657 1.745 1.833 1.921 2.008 2.075 2.183 2.271 2.358 2.446 2.533 2.621 85 1.665 1.753 1.841 1.929 2.017 2.085 2.193 2.281 2.369 2.457 2.545 2.633 90 1.672 1.761 1.849 1.938 2.026 2.094 2.203 2.291 2.379 2.468 2.556 2.644 95 1.679 1.768 1.857 1.946 2.034 2.103 2.212 2.301 2.389 2.478 2.567 2.656 100 1.685 1.774 1.864 1.953 2.042 2.111 2.221 2.310 2.399 2.488 2.577 2.666 105 1.691 1.780 1.870 1.960 2.050 2.119 2.229 2.319 2.408 2.498 2.587 2.677 110 1.696 1.786 1.877 1.967 2.057 2.127 2.237 2.327 2.417 2.507 2.597 2.687 115 1.701 1.792 1.882 1.973 2.064 2.134 2.245 2.335 2.425 2.516 2.606 2.696 120 1.706 1.797 1.888 1.979 2.070 2.140 2.252 2.343 2.433 2.524 2.615 2.706 125 1.710 1.801 1.893 1.984 2.076 2.147 2.258 2.350 2.441 2.532 2.623 2.714 130 1.714 1.805 1.897 1.989 2.081 2.152 2.265 2.357 2.448 2.540 2.631 2.723 135 1.717 1.809 1.902 1.994 2.086 2.158 2.270 2.363 2.455 2.547 2.639 2.731 140 1.721 1.813 1.906 1.998 2.091 2.163 2.276 2.369 2.461 2.554 2.646 2.738 145 1.724 1.816 1.909 2.002 2.095 2.167 2.281 2.374 2.467 2.560 2.653 2.746 150 1.727 1.819 1.912 2.006 2.099 2.172 2.285 2.379 2.472 2.566 2.659 2.752 155 1.729 1.822 1.916 2.009 2.102 2.175 2.290 2.384 2.477 2.571 2.665 2.758 160 1.732 1.825 1.918 2.012 2.106 2.179 2.294 2.388 2.482 2.576 2.670 2.764 165 1.734 1.827 1.921 2.015 2.109 2.182 2.297 2.392 2.486 2.581 2.675 2.770 170 1.737 1.830 1.924 2.018 2.112 2.185 2.300 2.395 2.490 2.585 2.680 2.775 175 1.739 1.832 1.926 2.020 2.114 2.188 2.304 2.399 2.493 2.589 2.684 2.779 180 1.741 1.834 1.928 2.023 2.117 2.191 2.306 2.402 2.497 2.592 2.688 2.783 185 1.743 1.837 1.931 2.025 2.119 2.194 2.309 2.405 2.500 2.596 2.691 2.787 190 1.745 1.839 1.933 2.027 2.122 2.196 2.312 2.407 2.503 2.599 2.695 2.791 195 1.747 1.841 1.935 2.029 2.124 2.198 2.314 2.410 2.505 2.602 2.698 2.794 200 1.749 1.843 1.937 2.031 2.126 2.200 2.316 2.412 2.508 2.604 2.701 2.797
lower confidence bounds, CU(or CL), on CPU (or CPL).
Four auxiliary functions for evaluating CU(or CL), the
normal distribution function, the Gamma function, polynomial roots function, and the noncentral t distri-bution function, are required. In order to accelerate the computation, Levinson [12] provided a good initial guess of d (below the real d) for numerical iteration. This equation is used to solve for d if t1 is given.
t1 dRn1þ Zc ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi R2 n1þ ð1 R2n1Þðd 2 Z2 cÞ q R2 n1 Z2cð1 R 2 n1Þ ; where Rn1¼ ½2=ðn 1Þ 1=2Cðn=2Þ=C½ðn 1Þ=2 and Zc¼
100c% of the standard normal distribution. Algorithm for the LCB
Step 1. Input the sample data (X1; X2; . . . ; Xn), USL (or
LSL), and c.
Step 2. Calculate X , S, bn1, and eCCPU (or eCCPL).
Step 3. Compute a good initial guess for d.
Step 4. Find the lower confidence bound CU(or CL) on
CPU (or CPL), through numerical iterations.
Step 5. Output the conclusive message, ‘‘The true value of the process capability CPU (or CPL) is no less
than the CU (or CL) with 100c% level of
confi-dence’’.
We implement the algorithm, and develop a Matlab program to compute the lower confidence bounds (see Appendix A). Table 3(a) and (b) tabulate the lower confidence bounds CU (or CL) for eCCPU (or eCCPLÞ ¼
0:7ð0:1Þ3:0, for n ¼ 5ð5Þ200, and c ¼ 0:95. Table 4 compares the lower confidence bounds obtained by Chou et al. [11] with one obtained by our approach. It is noted that with the same confidence level a, our ap-proach provides better bound (particularly for small n), which is closer to the real eCCPU (or eCCPL), and therefore
should be recommended for real-world applications.
5. An application example of low-drop-out linear regula-tors
The positive linear series pass voltage regulators are tailored for low-drop-out applications where low qui-escent power is considered important. These regulators include the reverse voltage sensing that prevents current in the reverse direction. The regulators are fabricated with the BiCMOStechnology, which is ideally suited for the low input-to-output differential applications. These products are specifically designed to provide well-regu-lated supply for low IC applications such as high-speed bus termination, low current logic supply, and VGA cards.
The products investigated here are low-drop-out 3A linear regulators with a low dropout voltage and short-circuit protection. These regulators are in three-lead packages and five-lead packages, as depicted in Fig. 1. The three-lead packages have preset outputs at 3.3 V or 5.0 V. The output voltage is regulated to 1.5% at room temperature. The five-lead packages regulate the output
Table 4
Comparisons of the LCB CUbetween the ChouÕs and our approach (New), for eCCPU¼ 0:7ð0:1Þ1:5, n ¼ 10ð10Þ50, c ¼ 0:95
n CCePU 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 10 New 0.41 0.49 0.56 0.63 0.70 0.77 0.84 0.93 0.97 Chou 0.37 0.44 0.50 0.57 0.63 0.70 0.76 0.82 0.88 20 New 0.50 0.58 0.65 0.73 0.81 0.89 0.97 1.04 1.12 Chou 0.47 0.55 0.63 0.70 0.78 0.85 0.93 1.00 1.08 30 New 0.53 0.61 0.70 0.78 0.86 0.94 1.03 1.11 1.19 Chou 0.52 0.60 0.68 0.76 0.84 0.92 1.00 1.08 1.16 40 New 0.55 0.64 0.72 0.81 0.89 0.98 1.06 1.14 1.23 Chou 0.54 0.63 0.71 0.79 0.87 0.96 1.04 1.12 1.20 50 New 0.57 0.66 0.74 0.83 0.91 1.00 1.09 1.17 1.26 Chou 0.56 0.64 0.73 0.81 0.90 0.98 1.07 1.15 1.24
voltage programmed by an external resistor ratio. Short-circuit current is internally limited. The device responds a sustained overcurrent condition by turning off after a tON time delay. The device then stay off for a period,
tOFF, that is 32 times the tONdelay. The device then
be-gins pulsing on and off at the tON=ðtONþ tOFFÞ duty cycle
of 3%. This drastically reduces the power dissipation during the short-circuit, which means that the heat sinks need only accommodating the normal operation.
The quiescent current is an essential product char-acteristic, which has significant impact to product quality. For the quiescent current of a particular model of low-drop-out 3A linear regulators, the upper specifi-cation limit, USL, is set to 650 lA. Sample data of 80 observations are collected, which are displayed in Table 5. Fig. 2 displays the histogram of the 80 observations. Fig. 3 displays the normal probability plot of the sample data, and from this figure the sample data appears to be normal. Further, we perform Shapiro–Wilk test to check whether the sample data is normal. The statistic W , is found to be 0.9926, we can conclude that the sample data can be regarded as taken from normal process.
In order to obtain the lower confidence bound on CPU, we execute the Matlab program attached in
Ap-pendix A. The program reads the sample data file, and input of n¼ 80, USL ¼ 650, and c ¼ 0:95, then output
with the sample mean X ¼ 398:85, the sample standard deviation S¼ 61:65, the correction factor bn1¼ 0:9905,
the estimator eCCPU¼ 1:371, and the lower confidence
bound CU¼ 1:191. The corresponding execution output
is included in Appendix B. We therefore conclude that the true value of the process capability CPU is no less
than 1.191 with 95% level of confidence.
6. Conclusions
In assessing the performance of normal stable man-ufacturing processes with one-sided specification limits, process capability indices CPUand CPLhave been widely
used to measure the process capability. In this paper, we developed an algorithm to compute the lower confidence bounds on CPUand CPLusing the UMVUEs of CPUand
CPL. The lower confidence bound presents a measure on
the minimum capability of the process based on the sample data. We also provided tables for the engineers/ practitioners to use in measuring their processes. The lower bounds we provided are closer to the true CPUand
CPL index values than the existing ones. A real-world
example taken from a microelectronics device manu-facturing process is investigated to illustrate the appli-cability of the algorithm. Implementation of the existing statistical theory for capability assessment fills the gap between the theoretical development and the in-plant applications.
Table 5
The 80 sample observations
428.63 408.09 417.62 317.20 519.40 438.75 380.42 457.60 474.87 395.22 344.88 497.17 373.81 337.62 430.78 363.31 475.43 270.65 357.53 454.24 402.19 462.12 403.40 463.31 513.74 433.62 361.49 360.67 400.46 540.87 361.72 303.60 377.56 376.22 379.27 332.91 258.16 357.04 352.24 432.00 432.66 395.20 371.64 315.74 330.47 405.25 324.87 337.89 435.80 399.28 433.73 358.42 422.63 419.25 387.12 277.72 464.39 388.89 384.68 455.70 387.57 337.50 444.43 494.75 472.60 369.92 393.09 492.78 429.29 403.41 274.69 324.70 528.84 443.18 331.91 428.18 423.13 397.66 440.76 332.47
Fig. 2. Histogram of the 80 observations.
Appendix A
% -% Input the sample dataðX1; X2; . . . ; XnÞ, USL, and c
% -function lcb(n,usl,r) t¼ 0; b¼ 0; z¼ 0; e¼ 0; f¼ 0; g¼ 0; rt¼ zerosð2; 1Þ; x¼ 0; x1¼ 0; x2¼ 0; y¼ 0; y1¼ 0; delta ¼ 0; delta1 ¼ 0; x3 ¼ 0; data¼ zeros(n,1); ½datað1 : n; 1Þ ¼ textread(Ôlingen.datÕ,Ô%fÕ,n); % -% Compute X , S, bn1, and eCCPU. % -mdata¼ mean(data); stddata¼ std(data); b¼ sqrtð2=ðn 1ÞÞ gammaððn 1Þ=2Þ= gammaððn 2Þ=2ÞÞ;
ecpu¼ ðusl mdataÞ=ð3 b stddata); fprintf(ÔThe sample Mean is %g.nnÕ,mdata)
fprintf(ÔThe sample standard Deviation is %g.nnÕ,std-data)
fprintf(ÔThe Correction Factor is %g.nnÕ,b) fprintf(ÔThe UMVUE of Cpu is %g.nnÕ,ecpu)%% % -% Compute a good initial value of d.
% -t¼ 3 ðecoy=bÞ sqrtðnÞ; R¼ sqrtð2=ðn 1ÞÞ gammaðn=2Þ= gammaððn 1Þ=2Þ; z¼ norminv(r,0,1); e¼ R^2 z^2 ð1 R^2Þ; f¼ 2 R ðt R^2 z^2 ð1 R^2Þ tÞ; g¼ ðt R^2 z^2 ð1 R^2Þ tÞ^ 2 z^2 R^2þ z^4 ð1 R^2Þ; pol¼ ½e f g ; rt¼ roots(pol); x¼ nctinvðr; n 1; rtð2ÞÞ; x1¼ nctinvðr; n 1; rtð2Þ þ 0:01Þ; x2¼ ðx1 xÞ=9:5; y¼ absðx sqrtðnÞ 3 ecpu=bÞ; y1¼ y=x2; delta1¼ rtð2Þ þ y1 0:001; % -% Evaluate the lower confidence bound cU(or cL) on
CPU(or CPL), by numerical iterations.
delta¼ delta1 þ 0:001 : 0:001 : delta1 þ 0:5; for i¼ 1 : 1 : 500
x3¼ nctinvðr; n 1; deltaðiÞÞ;
if(absðx3 sqrtðnÞ 3 ecpu=bÞÞ <¼ 0:01 cu¼ deltaðiÞ=ð3 sqrtðnÞÞ;
fprintf(ÔThe Lower Confidence Bound is %g.nnÕ,cu)
break end end
% -% Output the conclusive message, ‘‘The true value of % the process capability CPU(or CPL) is no less
% than cU (or cL) with 100c% level of confidence’’
% -fprintf(ÔThe true value of the process capability Cpu is no less than %gÕ, cu)
fprintf(Ôwith %gÕ,r) fprintf(Ôlevel of confidenceÕ) % -% The End % -Appendix B Input: >>lcb(80,650,0.95) Output:
The sample Mean is 398.85.
The Sample Standard Deviation is 61.6503. The Correction Factor is 0.990471. The UMVUE of Cpu is 1.37099. The Lower Confidence Bound is 1.19104.
The true value of the process capability Cpu is no less than 1.19104 with 0.95 level of confidence
References
[1] Kane VE. Process capability indices. J Qual Technol 1986;18(1):41–52.
[2] Chan LK, Cheng SW, Spiring FA. A new measure of process capability: Cpm. J Qual Technol 1988;20(3):162–75.
[3] Pearn WL, Kotz S, Johnson NL. Distributional and inferential properties of process capability indices. J Qual Technol 1992;24(4):216–31.
[4] Cheng SW. Practical implementation of the process capa-bility indices. Qual Eng 1994;7(2):239–59.
[5] Pearn WL, Chen KS. A practical implementation of the process capability index CPK. Qual Eng 1997;9(4):721–37.
[6] Pearn WL, Chen KS. Making decisions in assessing process capability index Cpk. Qual Reliab Eng Int 1999;
15:321–6.
-[7] Pearn WL, Chen KS. One-sided capability indices CPUand
CPL: decision making with sample information. Int J Qual
Reliab Manag 2002;19(3):221–45.
[8] Lin PC, Pearn WL. Testing process capability for one-sided specification limit with application to the voltage level translator. Microelectron Reliab 2002;42(12):1975–83. [9] Montgomery DC. Introduction to statistical quality
con-trol. 4th ed New York, NY: John Wiley & Sons, Inc; 2001.
[10] Chou YM, Owen DB. On the distributions of the estimated process capability indices. Commun Stat: Theory Meth 1989;18(12):4549–60.
[11] Chou YM, Owen DB, Borrego SA. Lower confidence limits on process capability indices. J Qual Technol 1990;22(3):223–9.
[12] Levinson WA. Exact confidence limits for process capa-bilities. Qual Eng 1997;9(3):521–8.