Total Quality Management & Business Excellence

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ctqm20

Process quality indices: new metrics for process

quality capability with zero-loss baseline

Xinjia Xiong & Yongchang Feng

To cite this article: Xinjia Xiong & Yongchang Feng (2021): Process quality indices: new metrics
for process quality capability with zero-loss baseline, Total Quality Management & Business
Excellence, DOI: 10.1080/14783363.2021.1911634

To link to this article: https://doi.org/10.1080/14783363.2021.1911634

© 2021 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group

Published online: 22 Apr 2021.

Submit your article to this journal

Article views: 1506

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ctqm20
Total Quality Management, 2021
https://doi.org/10.1080/14783363.2021.1911634

Process quality indices: new metrics for process quality capability

with zero-loss baseline
a* b
Xinjia Xiong and Yongchang Feng
a
Shanghai Gengyin Intelligent Technology Development Center, Shanghai, People’s Republic of
China; bAcademy & Life Care, TÜV Rheinland (Germany), Cologne, Germany

The traditional process capability indices Cp/Cpk for evaluating process capability used
in the current industry are based on the probability model of a process distribution,
which makes it difficult for many processes whose actual output with non-normal
distribution to apply these indices. In this paper, new process quality indices (PQI)
Qp/Qpk are proposed for measuring the quality capability of the process. The new
PQI are derived from Taguchi’s quality loss function model and do not need to
estimate the probability distribution of the process output. The traditional process
capability indices Cp/Cpk are zero-defect indices, it focus on meeting the process
specification requirements, while the new PQI Qp/Qpk are zero-loss indices, it focus
on approaching the process target value. The construction of the new PQI is logical,
simple, and practical. The PQI approach is more suitable for quality management in
the Era of big data.
Keywords: process capability; process quality indices Qp/Qpk; zero-loss; quality loss
function; zero-defects

1. Background
A more important goal of process quality management is to control the process in stable
and capable. Shewhart (1931) created the control charts approach for improving and moni-
toring the process stability, and the process capability indices Cp and Cpk are used for
measuring the process capability. But when the process capability indices are used, we
should estimate the probability distribution of the process. When the process is non-
normal distributed, it is more difficult to find the probability model for the process distri-
bution. We must use the statistical software to work for it. However, even the same
sampling data sets were analysed by different statistical software or different analysis strat-
egies, most often the analysis results are very different. In many cases, even using statisti-
cal software we also cannot find a suitable probability distribution model for the reality
process. Just because of this, on manufacturing site quality engineers often encounter
such problems, they will be in trouble and helpless. Moreover, the purpose of the
process capability indices are only to evaluate the process defects level (i.e. defect rate),
but the problem is that even if the product conforms to the specification, as long as the
value of a characteristic deviates from the target value, there will be quality loss. So, it
seems that the process capability should be considered from the perspective of quality
loss. If we can create simple and easy-to-use quality indices based on quality loss, it
will be more benefits for the statistical quality control in manufacturing site. From the

*Corresponding author. Email: [email protected]

© 2021 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License
(http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium,
provided the original work is properly cited, and is not altered, transformed, or built upon in any way.
2 X. Xiong and Y. Feng

concept of process capability and Taguchi’s quality loss function model, new process
quality evaluation indices are proposed in this paper. The application condition of these
indices is very simple, and no need for obtaining the process probability distribution.
The proposed new process quality indices (PQI) can completely replace the traditional
process capability indices Cp, Cpk and so on.

2. Review of process capability indices (PCI)

The concept of process capability was most likely developed by American statisticians.
Feigenbaum (1951) and Juran (1951) used 6σ as a metric of process capability. Juran
(1962) created a stronger link between process variability and customer specifications by com-
paring 6σ to the tolerance width as a method of determining the need for process improvement
activities. Finally, Juran and Gryna (1980) proposed a Capability Ratio (CR), which provided
the first metric that directly compared process variability to customer specifications.

6s
CR = (1)
USL − LSL

where USL is the Upper Specification Limit, LSL is the Lower Specification Limit, σ is the
process standard deviation.
No one knows who is the first one to create these PCI which used in current industry
widely. It is only known that they are originated from Japan in the 1980s. The first pub-
lished article to describe the PCI was Sullivan (1984). Sullivan (1985) described the
indices of Cp and Cpk which he had observed in use at Japanese manufacturing facilities.

USL − LSL
Cp = (2)
6s

min (USL − m, m − LSL)

C pk = (3)
3s

where μ is the mean value of the process.

Cp is the reciprocal of Juran and Gryna’s Capability Ratio (CR), so at the least, the
origin of PCI may ultimately be attributed to Juran and Gryna.
The application of PCI has been limited to measuring the conforming rate or the defect
rate of a process. However, if the process is not stable normal distributed, then these
indices cannot be directly converted to the process conforming rate or defect rate. Many
studies to the process capability have fallen into the deep and complex probability statistics
field, the purpose is only to estimate the suitable probability distribution for the process
output.
Since the Six-Sigma approach originated from MOTOROLA in 1987, the Six-Sigma
level has been talked about with great interest. In order to make the Six-Sigma level be
understood intuitively, the Six-Sigma approach always be preached that reaching the
six-sigma level means that the number of Defects Per Million Opportunities (DPMO) is
only 3.4. In the Six-Sigma approach, based on the underlying assumptions (1) the
process output is normal distributed, (2) generally, the process location has ±1.5σ drift,
a theoretical correspondence between sigma level and DPMO was linked, then the use
of PCI is more and more widely. People always want to measure their process sigma
level through PCI. In order to make the PCI estimate the conforming rate or defect rate
 Total Quality Management & Business Excellence 3

for the non-normal distributed process, generalised Cp/Cpk for non-normal distributed
process was proposed by Clements (1989).

USL − LSL
Cp = (4)
P0.99865 − P0.00135


USL − P0.5 P0.5 − LSL
C pk = min , (5)
P0.99865 − P0.5 P0.5 − P0.00135

where Pa is 100α percentile.

However, for the non-normal distributed process, in order to get the percentile for cal-
culating the PCI, we need to use software to transform the original process distribution into
normal distribution or find the appropriate probability distribution model for the original
process. The problem is that as Edwards Deming (1986) pointed out in his book out of
the crisis, ‘in practice, exactly stable processes never exist. Real processes are never
entirely free of perturbations’. But on the one hand, many statisticians who study PCI
are still addicted to use various statistical methods estimating the probability distribution
of the process. On the other hand, the engineers on the manufacturing site are generally
confused and helpless in analysing the actual process data. Even with the help of software,
using different probability estimation strategies, the same process data will get signifi-
cantly different capability indices estimation.
In fact, for the actual manufacturing process, the manufacturer essentially concerns the
quality loss caused by the manufacturing process. We can try to find a suitable approach to
evaluate the process from the view of quality loss for determining the need for process con-
tinuous improvement activities. The ideal goal of quality management should be to pursue
zero-loss based on target value, not merely zero-defects based on tolerance specification.
The PCI, which are limited to pursue zero-defects rather than zero-loss, are now increas-
ingly questioned in practical application.

3. Proposal of PQI
Return to the definition of quality terms in ISO 9000: 2015 standard: quality is the
degree to which a set of inherent characteristics of an object fulfils requirements. It is
better that the quality capability cannot be evaluated merely by the proportion of con-
formance products (i.e. within the specification tolerance), but by the degree of devi-
ation from the target value. The quality loss function model proposed by Genichi
Taguchi (2005) provides the basis for us to evaluate the quality capability in this
way, in other words, the quality capability can be measured by the loss caused by devi-
ation from the target value. Based on this fundamental, if there is a simple index can be
constructed to measure the quality level, then it is a good indicator for quality
management.
In fact, in the study of PCI, it has been considered that the PCI should be associated
with the target value, so the PCI Cpm and Cpmk were proposed.
Hsiang and Taguchi (1985) and Chan et al. (1988) respectively put forward PCI Cpm
considering process target value. Although their calculation equation looks the same,
Hsiang and Taguchi derived Cpm based on Taguchi’s quality loss function, while Chan
et al. derived this index based on the assumption of normal distribution. The Cpmk index
4 X. Xiong and Y. Feng

is proposed by Pearl et al. (1992) for harmonising Cpm and Cpk.

USL − LSL
C pm =  (6)
6 s2 + (m − m)2

min (USL − m, m − LSL)

C pmk =  (7)
3 s2 + (m − m)2

where m is the target value of the process.

Although the PCI Cpm and Cpmk are based on the target value to evaluate the process
capability, compared with Cp and Cpk, it seems that the initiators of these indices are sat-
isfied merely with the modification of Cp and Cpk relative to the target value. Maybe for this
reason, Cpm and Cpmk have not been widely used in the actual industry practice.
The quality definition and the quality loss function model proposed by Dr. Genichi
Taguchi provide a good perspective for us to evaluate the quality capability of the
process. Dr. Taguchi gives the definition of quality from the view of loss: quality is the
loss that a product imparts to society after the product is shipped. In Taguchi’s quality
engineering handbook (2005), he defined Tolerance as a deviation from target, denoted
by △. Tolerance is determined that the loss caused by the manufacturer and the one
caused by the customer are balanced.
There are three types of quality characteristics:

1. Nominal-the-best characteristic
2. Smaller-the-better characteristic
3. Larger-the-better characteristic.

For the Nominal-the-best characteristic, the expression of Taguchi’s quality loss func-
tion is as follows:

L(xi ) = k(xi − m)2 (8)

L(xi ) is the corresponding quality loss when a product characteristic value is xi , k is a scale
constant.

Figure 1. Taguchi’s quality loss function.

 Total Quality Management & Business Excellence 5

In Figure 1, A0 is the quality loss when a product does not function in the marketplace
(i.e. functional limit), and A is the quality loss when a product does not meet the delivery
standard (i.e. specification limit).

A0 = kD20 (9)

A = kD2 (10)

where D0 is the deviation of the functional limit from the target value m, △ is the toler-
ance, that is, the deviation of the specification limit from the target value m.
When the characteristic value is lower or higher than the target value m, the quality of
the Nominal-the-best characteristic will become worse. The specification standard is m
± △, and its tolerance △ is calculated as follows:


A
D= D0 (11)
A0

For the Smaller-the-better characteristic, the smaller the characteristic value the better. It
has no negative number, and the target value is m = 0, △ is the maximum allowable limit.
The expression of its quality loss function is as follows:

L(xi ) = kx2i (12)

Equation (11) is still can be used to calculate △ for the Smaller-the-better character-
istic. Equations (9) and (10) also apply to the Smaller-the-better characteristic.
For the Larger-the-better characteristic, the larger the characteristic value the better, △
is the minimum allowable limit. The quality loss function is different from the Nominal-
the-best characteristic and the Smaller-the-better characteristic. Its expression is as
follows:

1
L(xi ) = k (13)
x2i

Then △ is calculated as follows:


A0
D= D0 (14)
A

Now we can use the average quality loss to evaluate the product’s quality level.
According to the Equation (10), the scale constant k can be calculated for the Nominal-
the-best characteristic and the Smaller-the-better characteristic.

A
k= (15)
D2
6 X. Xiong and Y. Feng

For the Nominal-the-best characteristic, the average quality loss is calculated as follows:

1 N
A 1 N
A
L= k(xi − m)2 = 2 · (xi − m)2 = 2 [s2 + (
x − m)2 ] (16)
N i=1 D N i=1 D

where the tolerance D = USL − m or D = m − LSL, N is the number of products for cal-
culating the average quality loss, xi is the characteristic value of each product, and

N

x = xi is the average, used to evaluate the process mean (i.e. process location).
i=1
1 N
s2 = (xi −
x)2 is the variance, used to evaluate the process variation (i.e. process
N i=1
spread).
For the Smaller-the-better characteristic, the average quality loss is calculated as
follows:

1 N
A 1 N
A
L= kx2i = 2 · x2i = 2 (s2 +
x2 ) (17)
N i=1 D N i=1 D

where the maximum allowable limit D = USL,
x and s2 are same as in Equation (16).
For the Larger-the-better characteristic, according to Equation (13), the scale constant
k can be calculated:

k = AD2 (18)

While calculating the average quality loss of the Larger-the-better characteristic, convert
the original data to the reciprocal is the first step:
1
Let x′i =
xi
The average quality loss is calculated as follows:

1 N
1 1 N
L= k 2 = AD2 · xi ′2 = AD2 (s′2 + x′2 ) (19)
N i=1 xi N i=1


N 1 N
(x′ − x′ ) .
2
where the minimum allowable limit D = LSL, and x′ = x′i , s′2 =
i=1 N i=1 i
In order to create a stronger link between a PQI and the average quality loss, first we
can construct a calculation equation like Equation (6) for the Nominal-the-best character-
istic, and call it as the quality index Qpk:

D
Q pk =  (20)
3 s + (
x − m)2
2

where m is the target value, △ is the characteristic’s tolerance, D = USL − m

or D = m − LSL, N is the numbers of products for calculating the average quality loss,
 Total Quality Management & Business Excellence 7


N

x = xi is the average, used to evaluate process mean (i.e. process location),
i=1
1 N
s2 = (xi −
x)2 is the variance, used to evaluate process variation (i.e. process
N i=1
spread).
Putting Equation (20) substitute into Equation (16), and marking the subscript of pk for
the L in Equation (16), the following relation is obtained:

A A
L pk = · [s2 + (
x − m)2 ] = 2 (21)
D2 9Q pk

It shows that the quality index can be very simply linked with the average quality loss.
Even if the initial loss A is not known, we can evaluate the product’s relative quality
level through the quality index as well.
Let’s look at another situation. If there is no deviation from process mean to target
value in the manufactured products, that is to say
x − m = 0, then Equation (20)
becomes the following equation:

D D
Qpk = √ = (22)
3 s2 + 02 3s

The quality index Qpk in Equation (20) have considered the process location and spread.
While in Equation (22) only process variation is considered, the PQI is calculated by
assumption with no deviation from process mean to the target value. For the sake of dis-
tinction, this quality index in Equation (22) is marked as Qp, then:

D
Qp = (23)
3s

We can see that Qp ≥ Q pk is constant.

For corresponding to Qp, we denote the corresponding average quality loss L as Lp:

A A
Lp = · s2 = 2 (24)
D2 9Q p

For the Smaller-the-better characteristic, the Equations to calculate its quality indices are
the same as that for the Nominal-the-best characteristic. Now the target value m = 0, so the
calculation Equations for their quality indices are as follows:

D
Q pk = √ (25)
3 s2 +
x2

D
Qp = (26)
3s

N 1 N
where the maximum allowable limitD = USL,
x = xi , s2 = (xi −
x)2 .
i=1 N i=1
8 X. Xiong and Y. Feng

It should be noted that the characteristic’s mean value of an actual process will be
surely greater than the target value 0 for the Small-the-better characteristic, so the actual
PQI truly cannot reach to this Qp value.
For the Smaller-the-better characteristic, the relationship between the quality indices
and the average quality loss are as follows:

A 1 N
A A
L pk = · x2i = 2 (s2 +
x2 ) = (27)
D2 N i=1 D 9Q pk 2

A 2 A
Lp = s = 2 (28)
D 2 9Q p

For the Larger-the-better characteristic, because its quality loss function is different
from Nominal-the-best characteristic and Smaller-the-better characteristic, its quality
indices Equations are also different. Their calculation Equations are respectively as
follows:

1
Q pk =  (29)
3D s′2 + x′2

1
Qp = (30)
3Ds′

1 N
1 N
x′i , s′2 = (x′ − x′ ) .
2
where the minimum allowable limit D = LSL, x′ =
N i=1 N i=1 i
It should also be noted that the mean value of the reciprocal of the actual process
characteristic values will be surely greater than the target value 0 for the Larger-the-
better characteristic, so the actual PQI truly cannot reach to this Qp value as well.
For the Larger-the-better characteristic, the relationship between the quality indices
and the average quality loss are as follows:

1 N
1 A
L pk = AD2 · = AD2 (s′2 + x′2 ) = 2 (31)
N i=1 x2i 9Q pk

A
Lp = AD2 s′2 = (32)
9Q2p

From the relationship between the quality indices and corresponded average quality
loss, we can see that although the average loss function of the three types of characteristics
have different forms, they all have same simple relationships between quality index and
 Total Quality Management & Business Excellence 9

corresponding average quality loss as follows:

A
L pk = (33)
9Q2pk

A
Lp = (34)
9Q2p

This uniform and concise relationship between the quality index and the process
average quality loss enables us to use the PQI to effectively evaluate the process capability
for all three types of characteristics and provide direction for continuous process
improvement.
We not only can estimate the PQI based on the data of all sampled subgroups, but also
can calculate the quality indices for a single subgroup. The formula is exactly the same. In
addition, the quality index can also be calculated for a single characteristic value, the
quality index of the ith characteristic value can be expressed by Q pki . The formulas for
three types of single characteristic value are in Table 1.
The PQI or the quality index of a single subgroup can also be calculated based on the
quality index of a single characteristic value.
The quality loss corresponding to a single characteristic value can be expressed as
follows:

A
L(xi ) = (35)
9Q2pki

Therefore, the average process quality loss can be calculated as follows:

1 N
A 1 N
1
L pk = L(xi ) = · (36)
N i=1 9 N i=1 Q2pki

Combined with formula (33) and (36), the following relationship can be obtained:

1 1N
1
= (37)
Q2pk N i=1 Q2pki


N
Q pk = (38)
N 1
Q 2
i=1 pki

Table 1. PQI calculation formulas for three types of single characteristic value.
Type of characteristic Nominal-the-best Smaller-the-better Larger-the-better
D D 1
Q pki Q pki = Q pki = Q pki =
3|xi − m| 3xi 3Dx′i
10 X. Xiong and Y. Feng

where N is the total data numbers of all sampled subgroups. When the N becomes the sub-
group size n, the quality index Qpk of each subgroup can be calculated by using formula
(38).
In the process improvement, it is easy to calculate the reduced average quality loss rate
by using the PQI Qpk before and after the process improvement, to evaluate the perform-
ance of process improvement. According to formula (33), the reduced average quality loss
rate is calculated as follows:

2
L(0) − L(1) L(1) Q pk(0)
DL% = =1− =1− (39)
L(0) L(0) Q pk(1)

where L(0) represents the quality loss before improvement, L(1) represents the quality loss
after improvement, Q pk(0) represents the PQI before improvement, and Q pk(1) represents the
PQI after improvement.
For example, before improvement, Q pk(0) = 1.0, and after improvement,
Q pk(1) = 1.33, then the average quality loss rate reduced by process improvement is:


Q pk(0) 2 1 2
DL% = 1 − =1− = 43.47%
Q pk(1) 1.33

4. Application of PQI
For convenience, the formulas for calculating the quality indices of three types of charac-
teristic are summarised in Table 2.
The scaling factor of 3 in the quality indices are derived from using 6σ as the metrics
for the process capability by Juran. The quality indices approach is also used to evaluate
process capability. We can modify the definition of sigma level which used in the Six-

Figure 2. Relationship between process average quality loss L pk and quality index Q pk .
Table 2. PQI calculation formulas for three types of characteristic.
Type of characteristic Qp Q pk s2
x △
N
 N
D D 1 1
Nominal-the-best Qp = Q pk =  s2 = (xi −
x)2
x = xi D = USL − m or D = m − LSL
3s 2 N N
3 s + (
x − m)2 i=1 i=1
N N
D D 1 1
Smaller-the-better Qp = Q pk = √ s2 = (xi −
x)2
x = xi D = USL
3s 3 s2 +
x2 N i=1
N i=1
N N
1 1 1 2 1 1
Larger-the-better Qp = Q pk =  s′2 = (x′ − x′ ) x′i = , x′ = x′ D = LSL
3Ds′ 3D s′2 + x′2 N i=1 i xi N i=1 i
Total Quality Management & Business Excellence
11
12 X. Xiong and Y. Feng

Table 3. Cross-reference table between sigma level and PQI Qpk.

Sigma Level PQI Qpk
1.0 0.33
2.0 0.67
3.0 1.00
4.0 1.33
5.0 1.67
6.0 2.00

Sigma approach by linking it to PQI, Table 3 shows the cross-reference between sigma
level and quality index Qpk. The sigma level’s redefinition implies that we strongly
suggest the industry to use the new PQI instead of the traditional PCI for process capability
evaluation.
Below Figure 2 is the relation curve between Qpk and Lpk with Q pk = 1and the corre-
sponding process average quality loss unit L pk = 1.
In Figure 2, Zone A is the area with a quality index Qpk less than 1, Zone B is the area
with a quality index Qpk between 1 and 1.33, and Zone C is the area with a quality index
Qpk between 1.33 and 2.0. It can be seen that when the quality index Qpk falls into the Zone
A, the decline gradient of quality loss is very steep, that is to say, after improvement the
reduced unit quality loss is very high, so the process needs to be improved when the
quality index Qpk falls into the Zone A. In the zone B, although the reduced unit quality
loss after improvement is not as high as that in the Zone A, the unit value reduced is
still considerable, so it is suggested if the quality index Qpk falls into the Zone B, the
improvement is still needed. In the Zone C, the reduced unit quality loss after improvement
is further reduced, but there is still space to generate benefits by reducing the quality loss.
For the special characteristic that have a great impact on product quality cost, it is rec-
ommended to improve it when the quality index Qpk is lower than 1.67. When the
quality index Qpk is up to 3.5, the reduced unit quality loss after the improvement is
very small, the process can be considered as in a near perfect state (Table 4).
In order to make the PQI direct the quality improvement effectively, it is suggested that
the quality indices Qp and Qpk can be jointly evaluated.
In statistical process control, first, using the Shewhart process control charts approach
to evaluate the process stability (i.e. whether the evaluated process is under statistical

Table 4. Process improvement guidelines based on PQI Qpk.

Zone Qpk Improvement recommendation
A Qpk < 1.00 The decreasing gradient of quality loss after process improvement is
very steep, and the unit quality loss after improvement is very high,
so the process needs to be improved
B 1.00 ≤ Qpk < Although the unit quality loss reduced after improvement is not as high
1.33 as that in zone A, the reduced unit value is still considerable, so the
process should be improved
C 1.33 ≤ Qpk < There is still space for profit through reducing quality loss. For the
2.00 special characteristic that has a great impact on product quality cost,
it is recommended to improve it when the quality index Qpk is lower
than 1.67
Perfect Qpk ≥ 2.00 When the quality index Qpk is up to 3.5, the reduced unit quality loss
Zone after the improvement is very small, the process can be considered
as in a near perfect state
 Total Quality Management & Business Excellence 13

control). If the process is out of statistical control, then identify the possible assignable
causes, and take actions to eliminate or control it. Next, to evaluate the process capability
by using PQI, first to evaluate Qp, and then to evaluate Qpk. If Qp fails to meet the customer
requirements or the company requirements, improvement actions shall be taken to reduce
the process variation; if Qp meets the requirements, but Qpk fails to meet the customer
requirements or the company requirements, improvement actions shall be taken to
reduce the deviation of the process mean from target value. For process capability
improvement, the suggestion is first to reduce the process variation as much as possible
(i.e. to improve the process robustness), and then consider reducing the deviation of the
process mean from the target value.
In Table 5, the suggested criteria were listed for evaluating process quality capability
based on PQI. Each company can establish their appropriate quality indices evaluation cri-
teria according to the impact of different quality characteristics on product quality loss
(Figure 3).
It does not mean that the quality indices can only be used to evaluate the stable process.
In fact, it also can be used to evaluate the unstable process, the reason is that the evaluation
of quality indices has no other requirements for the process data except requires the
measured data can be accurate and represent the actual manufacturing process.
However, when evaluating the process, if the process is in unstable, it means that assign-
able cause maybe exists. Therefore, efforts should be made to identify the assignable
causes at first, and actions should be taken to eliminate the assignable causes or to
control it effectively (for example, to control such as tool wear this kind of assignable
cause by controlling the counts of tool used). It should be noted that for unstable
process, the calculated PQI are only used as the review for the previous process perform-
ance, it cannot be used to predict the quality capability for the future process (Figure 4).

5. Cases study for using PQI

Case 1: Nominal-the-best characteristic
The characteristic specification is 28.5 ± 0.3 mm, 25 sub-groups with 125 data col-
lected in the initial process capability study. Target value m = 28.5 mm, tolerance △ =
0.3 mm (Figure 5 and Table 6).

Table 5. Suggested evaluation criteria for PQI.

Quality Indices Suggested Evaluation Criteria
Qp ≥ 1.33
Process quality capability is sufficient, release process or continue to maintain
Q pk ≥ 1.33 and/or improve process
Qp ≥ 1.67
Process quality capability is insufficient. Actions shall be taken to reduce the
Q pk , 1.33 deviation of process mean from target value
1.33 ≤ Qp , 1.67
Process quality capability is insufficient. It is suggested that appropriate actions
Q pk , 1.33 could be taken to reduce the process variation, and then other actions should
be taken to reduce the deviation of process mean from target value
Qp , 1.33
Process quality capability is insufficient, actions shall be taken to reduce
Q pk , 1.33 process variation
14 X. Xiong and Y. Feng

Figure 3. The SPC evaluation process based on PQI.

Conclusion: The process is in stable and its quality capability meets the requirement
(Q pk = 1.39 . 1.33), the process can be released. The PCI of this process are Cp = 1.96,
C pk = 1.61.
Case 2: Nominal-the-best characteristic
This case is introduced before Case 1 for the same characteristic, it is failed to pass the
previous initial process study (Figure 6 and Table 7).
Conclusion: The process is in unstable and its quality capability does not meet the
requirement (Q pk = 1.06 , 1.33). The process needs to be improved. First, we need to
find the assignable causes which result in process out of control, then re-evaluate the
process quality capability.
The assignable cause is the positioning reference drifted by 0.1 mm after the machine
was adjusted during subgroups No. 2 to No. 5. Because the process is in unstable, only the
process performance indices are calculated: Pp = 1.48, P pk = 1.15.
Case 3: Smaller-the-better characteristic
The runout specification of a product is USL = 0.2 mm. 25 sub-groups of 125 data are
collected in the initial process capability study. △ = USL = 0.2 mm, and the target value is
m = 0 (Figure 7 and Table 8).

Figure 4. Example of control charts which shows Qpk for each subgroup.
 Total Quality Management & Business Excellence 15

Figure 5. Control charts of Case 1.

Table 6. Case 1 study results summary.

1 
N
(
x − m)2 = 0.00284

x = xi = 28.5533
N i=1
1 N
s2 = (xi −
x)2 = 0.00235 s = 0.0484
N i=1
D D
Qp = = 2.06 Q pk =  = 1.39
3s 3 s2 + (
x − m)2

Figure 6. Control charts of Case 2.

16 X. Xiong and Y. Feng

Table 7. Case 2 study results summary.

1 
N
(
x − m)2 = 0.00426

x = xi = 28.5653
N i=1

1 N
s2 = (xi −
x)2 = 0.00459 s = 0.0678
N i=1
D D
Qp = = 1.48 Q pk =  = 1.06
3s 3 s + (
x − m)2
2

Figure 7. Control charts of Case 3.

Table 8. Case 3 study results summary.

1 
N
1 N

x = xi = 0.08456 s2 = (xi −
x)2 = 0.00041.
N i=1
N i=1
s = 0.020248.
x2 = 0.00715.
D D
Qp = = 3.29. Q pk = √ = 0.77.
3s 3 s2 +
x2

Table 9. Case 4 study results summary.

1 N
1 N
x′ i = 0.026 (x′ − x′ ) = 0.000045
2
x′ = s′2 =
N i=1 N i=1 i
s′ = 0.006708 x′2 = 0.000676
1 1
Qp = = 4.97 Q pk =  = 1.24
3Ds′ 3D s′2 + x′2
 Total Quality Management & Business Excellence 17

Figure 8. Control charts of Case 4.

Conclusion: The process quality capability is insufficient (Q pk = 0.77 , 1.33) and

actions should be taken to reduce the deviation of the process mean from the target value.
The PCI is C pk = 2.02.
Using traditional PCI to evaluate the process, it is in stable and capable. But obviously,
using PQI to evaluate the process makes more sense.
Case 4: Larger-the-better characteristic
The product drawing force specification is LSL = 10N. 25 sub-groups of 125 data were
collected in the initial process capability study (Figure 8 and Table 9).
Conclusion: Process quality capability is insufficient (Q pk = 1.24 , 1.33), actions
should be taken to put the process mean value away from the lower specification limit.
The PCI is C pk = 1.21. The evaluation results of the two indexes are similar.

6. Summary
In fundamentally, the PQI focus not on the defect rate of the process, but on the average
quality loss of the process. In other words, the ideal goal of the process quality is zero-loss
based on the target value, not merely on the zero-defects based on the specification limit.
Using PQI to evaluate process quality capability is a great change of quality management
philosophy. For producer, the ideal goal of quality management should be zero-loss.
Compared with traditional PCI, new PQI have the following four significant
advantages:

1. The PCI need to be evaluated based on the probability distribution of the process,
which is very troublesome for the case of non-normal distributed process. The PQI
have no requirements for process distribution, and do not need to make any assump-
tions or make great efforts to find the probability model of a process. As Deming
and Wheeler & Neave (2015) pointed out, in practice, exactly stable processes
never exist. Real processes are never entirely free of perturbations. The PQI do
18 X. Xiong and Y. Feng

not need to consider the probability model of the process at all. If the samples which
collected for the process capability analysis can well represent the process, the eval-
uated result of the process quality capability will be unique. Neither the PQI nor the
Shewhart control charts approach require probability model, used them in combi-
nation, can ensure the process to be improved and controlled in the most economi-
cal, simple and effective way, to bring the process truly maintained in an expected
stable and capable state.
2. The traditional PCI based on the delivery specification limits only concern the
product conforming rate or defect rate, the deviation from the target value of the
product within the delivery specification limits is not evaluated. The new PQI
based on Taguchi’s quality loss function can include all products within the func-
tional limits into the evaluation, and the evaluation baseline is the deviation of each
characteristic value from the target value. In other words, the evaluation scope of
the new PQI not only cover the non-conforming products which out of the delivery
specification limits, but also include the products which within the delivery speci-
fication limits. In the actual production process, even if there is a case that the pro-
ducts which out of the functional limits are shipped to the customer, it will be a rare
accident. So, for evaluating the process quality capability, it is perfectly possible
only to consider the case that all product characteristic values are within functional
limits.
3. The new PQI used combined with Shewhart control charts approach will make the
statistical process control completely bypass the attention to the probability distri-
bution of the process, this will greatly simplify the Statistical Process Control
(SPC), which is more suitable for engineers and even operators to use SPC on
the manufacturing site.
4. No matter how many numbers of the data have, the new PQI can be evaluated. An
individual characteristic value can be evaluated, one subgroup can be evaluated, a
big data process with full automation inspection can also be evaluated. This makes
the new PQI approach very suitable for quality management in the Era of big data.

Obviously, it is do necessary to use new PQI instead of traditional PCI. PQI are the true
metrics for evaluating the process quality level. In addition to the assumption based on the
Taguchi’s quality loss function, the new PQI do not care about the actual probability dis-
tribution of the process at all. The standard and easy understood calculation formulas for
three types of quality characteristics have been given. In the process improvement, by com-
paring the PQI before and after the improvement, we can easily evaluate the improvement
performance and calculate the average loss rate reduced by the process improvement. The
Zone which the PQI located can indicate the continuous improvement potential.
Although the proposed new PQI based on Taguchi’s quality loss function are very suit-
able for evaluating the capability of any distributed process, the following topics are still
worthy of further study.

1. How to evaluate the confidence level of the estimated PQI and determine its con-
fidence interval from the perspective of statistics?
2. How to evaluate the influence of measurement process capability on the PQI
evaluation?
3. Similar with Rolled Throughput Yield (RTY), how to use PQI to evaluate the
overall process capability of the entire manufacturing process?
4. Other extension topics related to PQI need to be further explored.
 Total Quality Management & Business Excellence 19

Disclosure statement
No potential conflict of interest was reported by the author(s).

ORCID
Xinjia Xiong http://orcid.org/0000-0002-7186-6376
Yongchang Feng http://orcid.org/0000-0002-7696-6683

References
Chan, L. K., Cheng, S. W., & Spiring, F. A. (1988). A new measure of process capability: Cpm.
Journal of Modeling and Simulation, 11, 1–6.
Clements, J. A. (1989). Process capability calculations for non-normal distributions. Quality
Progress, 22(9), 95–100.
Edwards Deming, W. (1986). Out of the crisis. MIT. Center for Advanced Educational Services,
334–335.
Feigenbaum, A. V. (1951). Quality control. McGraw-Hill. 383–389.
Hsiang, T. C., and Taguchi, G. (1985). A tutorial on quality control and assurance – the Taguchi
methods. Unpublished presentation given at the Annual Meeting of the American Statistical
Association, LasVegas, NV.
Juran, J. M. (1951). Quality Control handbook. McGraw-Hill. 253–279, 404-409.
Juran, J. M. (1962). Quality Control Handbook (2nd ed). McGraw-Hill. 11–27.
Juran, J. M., & Gryna, F. M. (1980). Quality planning and analysis (2nd ed). McGraw-Hill. 283–295.
Pearn, W. L., Kotz, S., & Johnson, N. L. (1992). Distributional and inferential properties of process
capability indices. Journal of Quality Technology, 24(4), 216–231. https://doi.org/10.1080/
00224065.1992.11979403
Shewhart, W. A. (1931). Economic control of quality of manufactured products. D. Van Nostrand.
249–272.
Sullivan, L. P. (1984). Reducing variability: A new approach to quality. Quality Progress, 17(7),
15–21.
Sullivan, L. P. (1985). Letters. Quality Progress, 18(4), 7–8.
Taguchi, G., Chowdhury, S., & Wu, Y. (2005). Taguchi’s quality engineering handbook. John Wiley
& Sons. Inc, 16–22, 174–175, 180–191.
Wheeler, D. J., & Neave, H. R. (2015). Shewhart and the Probability Approach: The difference is
much greater than how we compute the limits. Quality Digest Daily, November 2.