Sankhyā : The Indian Journal of Statistics

1994, Volume 56, Series B, Pt. 3, pp. 468 - 478

GERT ANALYSIS OF DODGE’S CSP-1 CONTINUOUS

SAMPLING PLAN∗
By GAURI SHANKAR and B. N. MOHAPATRA
Pt. Ravishankar Shukla University
SUMMARY. Graphical Evaluation and Review Technique (GERT), originally introduced
for systems design and analysis, has been applied here to model and analyse the dynamics
of the Dodge’s CSP-1 plan. Procedures and tables has been provided to find a unique com-
bination of (i, f ) that will achieve the AOQL requirement and also, to optimize the average
amount of inspection function, E(I), when the process level p = pw is known.

1. Introduction

The concept of continuous sampling plan (CSP-1) was introduced by Dodge

(1943) as a sampling inspection plan for a product consisting of individual units
manufactured in quantity by an essentially continuous process. The detailed
procedure and tables for construction and selection of CSP-1 plans have been
given by Stephens (1981). Ghosh (1988) and Govindaraju (1989). Whitehouse
(1973; 401 - 403) and also Ohta and Kase (1984) and Chakraborty and Rathie
(1989) modelled and analysed the Dodge’s CSP-1 continuous sampling plan
through GERT approach. However, the limitation in the GERT network mod-
eling of the CSP-1 plan in the above studies is two-fold. Firstly, the GERT
network representation is limited to the separate modeling; one during detailing
state (100% inspection) and, the other during sampling inspection state. Hence
they fail to model the dynamics of the plan during one inspection cycle (i.e.
100% inspection to sampling inspection). Secondly, no attention has been paid
to the selection of plan parameters in the light of new interpretation of the ex-
isting system.

The purpose of the present investigation is to model and analyse the dy-
namics of CSP-1 plan during one inspection cycle through GERT approach.
The advantage of GERT analysis in the present context is two fold. First, this
procedure gives the visual picture of the dynamics of the inspection system and

Paper received. February 1991; revised February 1993.

AMS (1985) subject classification. 62N10.
Key words and phrases. AOQL, average amount of inspection, GERT analysis, continuous
sampling plan.
∗ This research is financed by DST, New Delhi (Grant No. SP/YS/P01/87).
 gert analysis of sampling plan 469

second, it offers thorough characterization of the plan. Various performance

characteristics of interest to quality control engineers and plan engineers have
been derived and illustrated numerically. Finally, procedures and tables have
been provided for the selection of plans in the context of new developments.
This paper presents tables to find a unique combination of plan parameters
(i, f ) that will achieve AOQL requirement and maximizes the average number
of units inspected by CSP-1 plan during one inspection cycle E(I), at p = pw .
The E(I) function defined for one inspection cycle is maximum at p = pw when
the system remains in the detailing state for a long time. Thus, p = pw is the
worst incoming quality to be considered by the plan. Therefore, like AOQL the
specification of pw also alarms the state of corrective action to be taken by the
producer.

2. Operating procedure of the CSP-1 plan

The operating procedure of the CSP-1 plan as stated by Dodge is

as follows :

(a) At the outset , inspect 100% of the units consecutively as produced and
continue such inspection until i units in succession are found clear of de-
fects.
(b) When i units in succession are found clear of defects, discontinue 100%
inspection, and inspect only a fraction f of the units, selecting individual
units one at a time from the flow of product, in such a manner as to ensure
an unbiased sample.
(c) If a sample unit is found defective, revert immediately to a 100% inspection
of succeeding units and continue until again i units in succession are clear
of defects, as in step (a).
(d) Correct or replace with good units, all defective units found.

Thus, the CSP-1 plans are characterised by two parameters i and f .

3. Brief review of GERT

GERT was initiated by Pritsker and Happ (1966), Pritsker and Whitehouse
(1966) and Whitehouse and Pritsker (1969) as a procedure for the analysis of
stochastic networks having the following features:

(1) Each network consists of logical nodes (or events) and directed branches
(or activities).
 470 gauri shankar and b.n. mohapatra

(2) A branch has a probability that the activity associated with it will be
performed.
(3) Other parameters describe the activities represented by the branches. In
this paper, however, reference will be made to a sample size parameter
only.

The sample size n associated with a branch is

characterised by the moment
generating function (mgf) of the form Mn (θ) = exp (nθ)f (n), where f (n)
n
denotes the density function of n and θ is any real variable. The probability φ
that the branch is realised is multiplied by the mgf to yield the W -function such
that
W (θ) = φMn (θ) . . . (3.1)
The W -function is used to obtain the information on the relationship which
exists between the nodes.

4. GERT analysis of the plan

The possible states of the CSP-1 inspection system described in section (2)
can be defined as follows:

S0 : Initial state of the plan.

S1 (k) : State in which k(= 1, 2..., i) preceding units are found clear of defects
during 100% inspection.
SP : Initial state of partial inspection.
S2 : State in which a unit is not inspected (i.e. passed) during sampling in-
spection.
SPA : State in which a unit is found free of defects during partial (sampling)
inspection.
SPR : State in which a unit is found defective during partial inspection.
SA : State in which current unit is accepted.
SR : State in which current unit is rejected.

The above states enable us to construct GERT network representation of the

inspection system as shown in Fig. (1) and (2). Suppose that the process is in
statistical control, so that the probability of any incoming unit being defective
is (p) and the probability of any unit being non-defective is q = 1 − p. First
of all, we will show that the probability of acceptance and rejection of a unit
 gert analysis of sampling plan 471

during sampling inspection [see Fig. (1)] is same as that of its acceptance and
rejection during 100% inspection. Now, by applying Mason’s (1953) rule in
the representation in Fig. (1), the W -functions from the initial node S0 to the
terminal nodes SPA and SPR are respectively found as

f q i+1 [1 − (1 − f )eθ ]eθ + f (1 − f )q i+1 e2θ

W1A (θ) = − . . . (4.1)
1 − [(1 − q i ) + (1 − f )eθ ] + (1 − f )(1 − q i )eθ

and
f pq i+1 eθ [1 − (1 − f )eθ ] + f pq i+1 (1 − f )e2θ
W1R (θ) = − . . . (4.2)
1 − [(1 − q i ) + (1 − f )eθ ] + (1 − f )(1 − q i )eθ

From the W -functions defined above, we obtain the probability that a unit is
accepted and rejected respectively by sampling procedure as

[W1A (θ)]θ=0 = q

and
[W1R (θ)]θ=0 = p
Also, average number of units considered during a period of sampling inspection
(v1 ) is
d d
v1 = q[ M1A (θ)]θ=0 + p[ M1R (θ)]θ=0 = 1/f
dθ dθ
where M1A (θ) = W1A (θ)/W1A (0) and M1R (θ) = W1R (θ)/W1R (0).
472 gauri shankar and b.n. mohapatra

Furthermore, Dodge (1943) has shown that average number of units inspected
during sampling inspection is 1/p. Therefore, average number of units passed
during sampling inspection is

v = (1/f ).(1/p) = 1/f p

Keeping the above fact in mind, the acceptance and rejection sequence of the
CSP-1 inspection system during one inspection cycle can be represented by Fig.
(2). Consequently, the W -function from the initial node S0 to the terminal
nodes SA and SR are respectively given as

f qeθ + (1 − f )(qeθ )i
WA (θ) = − . . . (4.3)
1 − (1 − f )peθ [{1 − (qeθ )i }/(1 − qeθ )]

f peθ
WR (θ) = − . . . (4.4)
1 − (1 − f )peθ [{1− (qeθ )i }/(1 − qeθ )]
Therefore,

PA = [WA (θ)]θ=0 = [f q + (1 − f )q i ]/[f + (1 − f )q i ] . . . (4.5)

and
PR = [WR (θ)]θ=0 = f p/[f + (1 − f )q i ] . . . (4.6)
where PA and PR stands for probability of acceptance and rejection (of a unit)
by CSP-1 plan respectively. These results coincide with Perry (1973) derived
for skip-lot sampling plan (SkSP-2).
Since PA fraction of accepted units are defective with probability p and
(1 − PA ) fraction are non-defective with probability q = 1 − p. Again, since all
 gert analysis of sampling plan 473

defective units are replaced by good ones, therefore, average outgoing fraction
defective (Average Outgoing Quality, AOQ) is defined as

AOQ = [pPA − qPR ]/(PA + PR )

. . . (4.7)
= [(1 − f )pq i ]/[f + (1 − f )q i ]
474 gauri shankar and b.n. mohapatra

Proceeding in the same way as above, if defective units found are removed but
not replaced, then, average outgoing fraction defective is given as

AOQ
= [pPA − qPR ]/PA = [(1 − f )pq i ]/[f q + (1 − f )q i ] . . . (4.8)

These results agree with Dodge (1943).

For further characterization of the plan, the average number of units in-
spected by CSP-1 plan during one inspection cycle can be defined as follows:
d d
E(I) = PA [ dθ MA (θ)]θ=0 + PR [ dθ MR (θ)]θ=0
. . . (4.9)
= [1 − {(f q + (1 − f )q }]/[f p + (1 − f )pq i ]
i

where MA (θ) = WA (θ)/WA (0) and MR (θ) = WR (θ)/WR (0).

 gert analysis of sampling plan 475

Here, it may be observed that when f = 0 the resulting plan becomes 100%
inspection only. Therefore, putting f = 0 in (4.9), the average amount of
inspection, E(I), comes out to be

E(I) = (1 − q i )/pq i = u

Knowing u and v, the average fraction of units inspected (F ) is

F = (u + f v)/(u + v) = f /[f + (1 − f )q i ]

5. Selection of CSP-1 plans

Dodge (1943) remarked that there are several combinations of plan parame-
ters i and f that will ensure the same AOQL over all possible values of incoming
quality p. Stephens (1981) provided tables for selection of (i, f ) for consumer
protection based on LQL with 0.10 risk. Ghosh (1988) developed a procedure
to find a unique (i, f ) that will achieve AOQL requirement and minimizes F
(average fraction inspected) when the process average p is known. Govindaraju
(1989) provided tables for the selection of a CSP-1 plan for a given set of con-
ditions (AQL, AOQL) and (LQL, AOQL).

This paper provides procedures and tables to find a unique combination

of (i, f ) that will achieve AOQL requirement and also, maximizes the average
amount of inspection, E(I), when the process level p = pw is known. This later
criterion also alarms the state of corrective action to the producer.
Example. Suppose a CSP - 1 plan is required having AOQL = 0.05 and
which maximizes E(I) at p = pw = 0.12. Table (1) yields a CSP-1 plan with
i = 48 and f = 0.0123. Now, for this plan AOQ and E(I) curves have been
drawn in Fig. (3) and (4). It is seen from the figure that this plan gives the
required AOQL on one hand and also maximizes E(I) at p = pw .

6. Construction of tables

The expression for the average amount of inspection, E(I), is written

E(I) = [1 − f q − (1 − f )q i ]/[f p + (1 − f )pq i ]

Now, differentiating E(I) with respect to p and equating to zero, one gets

f = [ipq i−1 − q i (1 − q i )]/[(1 − q i )2 − ip2 q i−1 ] . . . (6.1)

If pm is the incoming quality at which AOQL occurs, then the following results
due to Dodge (1943)
i+1 i+1
f = qm /[i(AOQL) + qm ] . . . (6.2)
476 gauri shankar and b.n. mohapatra

where qm = 1 − pm and pm = [1 + i (AOQL) ] / (i + 1).

Thus, from equations (6.1) and (6.2), on simplification, we have

pq i−1 [i2 (AOQL) + iqm
i+1
(1 + p)] − (1 − q i )[qm
i+1
+ i(AOQL)q i ] = 0 . . . (6.3)
For given value of p = pw and AOQL, equation (6.3) can be solved for i by
numerical methods and then the value of f can be found from equation (6.2).
Table (1) is constructed for some selected values of AOQL and proces level pw .
TABLE 1. VALUES OF i AND f OF CSP-1 PLAN FOR
GIVEN AOQL AND PROCESS LEVEL Pw
AOQL = 0.01 AOQL = 0.02
pw i f pw i f
0.0200 459 0.0008 0.040 226 0.0008
0.0210 392 0.0018 0.042 193 0.0019
0.0220 338 0.0036 0.044 166 0.0037
0.0230 293 0.0065 0.046 144 0.0068
0.0240 254 0.0101 0.048 125 0.0113
0.0250 221 0.0176 0.050 109 0.0180
0.0260 191 0.0270 0.052 94 0.0273
0.0270 165 0.0404 0.054 82 0.0405
0.0280 140 0.0600 0.056 70 0.0589
0.0290 114 0.0923 0.058 58 0.0865
0.0292 108 0.1022 0.060 44 0.1410
0.0296 91 0.1373 – – –
AOQL = 0.03 AOQL = 0.04
pw i f pw i f
0.055 198 0.0001 0.08 109 0.0009
0.060 149 0.0009 0.09 75 0.0054
0.065 115 0.0031 0.10 53 0.0187
0.070 90 0.0083 0.11 37 0.0485
0.075 72 0.0183 0.12 25 0.1103
0.080 56 0.0385 0.13 8 0.4194
0.085 44 0.0653 0.14 6 0.5166
0.090 32 0.1201 0.15 5 0.5498
AOQL = 0.05 AOQL = 0.06
pw i f pw i f
0.11 63 0.0042 0.11 94 0.0002
0.12 48 0.0123 0.12 70 0.0010
0.13 36 0.0285 0.13 55 0.0035
0.14 28 0.0571 0.14 43 0.0091
0.15 20 0.1043 0.15 34 0.0196
0.16 14 0.1825 0.16 27 0.0368
0.17 10 0.2987 0.17 22 0.0627
0.18 7 0.3877 0.18 17 0.1001
0.19 6 0.4371 0.19 14 0.1503
0.20 5 0.4693 0.20 11 0.2120
0.21 5 0.4930 0.21 9 0.2746
0.22 5 0.5116 0.22 7 0.3265
0.23 4 0.5270 0.23 6 0.3657
0.24 4 0.5401 0.24 6 0.3964
0.25 4 0.5515 0.25 5 0.4208
 gert analysis of sampling plan 477

TABLE 1. Continued

AOQL = 0.07 AOQL = 0.08

pw i f pw i f
0.15 48 0.0031 0.15 63 0.0003
0.16 39 0.0074 0.16 51 0.0011
0.17 32 0.0148 0.17 42 0.0029
0.18 26 0.0265 0.18 35 0.0063
0.19 22 0.0434 0.19 29 0.0119
0.20 18 0.0666 0.20 25 0.0205
0.21 15 0.0968 0.21 21 0.0326
0.22 12 0.1337 0.22 18 0.0488
0.23 10 0.1754 0.23 15 0.0693
0.24 9 0.2184 0.24 13 0.0943
0.25 8 0.2586 0.25 11 0.1229
AOQL = 0.09 AOQL = 0.10
pw i f pw i f
0.18 45 0.0012 0.18 56 0.0002
0.19 38 0.0028 0.19 47 0.0005
0.20 32 0.0056 0.20 40 0.0012
0.21 27 0.0101 0.21 34 0.0027
0.22 23 0.0167 0.22 29 0.0051
0.23 20 0.0260 0.23 25 0.0088
0.24 18 0.0380 0.24 22 0.0143
0.25 15 0.0531 0.25 19 0.0215
0.26 13 0.0713 0.26 17 0.0309
0.27 12 0.0924 0.27 15 0.0426
0.28 10 0.1158 0.28 13 0.0566
0.29 9 0.1409 0.29 12 0.0729
0.30 8 0.1661 0.30 10 0.0910

Acknowledgement. The authors are thankful to the referees for their valuable
comments.

References
Chakraborty, S. and Rathie, U. K. (1989) Analysis of inspection error through GERT.
Sankhyā, 51, Ser. B, Pt. I, 125-133.
Dodge, F. (1943). A sampling plan for continuous production. Ann. Math. Statist. 14, 3,
264-279.
Ghosh, D. T. (1988). A continuous sampling plan that minimizes the amount of inspection.
Sankhyā, 50, Ser. B, Pt. 3, 412-427.
Govindaraju, K. (1989). Procedures and tables for the selection of CSP - 1 plans. Jour.
Quality Technology, 21, 1, 46-50.
Ohta, H. and Kase, S. (1984). GERT analysis of economic design of Dodge’s CSP - 1
continuous sampling plan under inspection error. Frontiers in Quality Control, 2, Ed.
by Lenz et al., 247-260. Physica-Verlag.
Mason, S. J. (1953). Some properties of signal flowgraphs. Proc. IRE, 41, 9, 1144-1156.
Perry, R. L. (1973) Skip-lot sampling plans. Jour. Quality Technology, 5, 3, 123-130.
Pritsker, A. A. B. and Happ, W.W. (1966) GERT; Graphical Evaluation and Review
Technique, Pt-I, Fundamentals. Jour. Ind. Engg., 17, 267-274.
Pritsker, A.A.B. and Whitehouse, G. E. (1966). GERT; Pt. II - Probabilistic and
industrial engineering applications. Jour. Ind. Engg.. 17, 293-301.
478 gauri shankar and b.n. mohapatra

Stephens, K. S. (1981). CSP - 1 for consumer protection. Jour. Quality Technology, 13,
4, 249-253.
Whitehouse, G. E. and Pritsker, A. A. B. (1969) GERT; Pt. III - Further statistical
results: Counters, renewal times and correlations. AIIE Trans., 1, 45-50.
Whitehouse, G. E. (1973). Systems Analysis and Design using Network Techniques, Pren-
tice Hall, N. J.

School of Studies in Statistics

Pt. Ravisankar Shukla University
Raipur, M.P.
India.