a a procedure for ranking diagnostic test inputs … · k3 rn a 0 a procedure for ranking...
TRANSCRIPT
.I
\
REPORT) R-354 . h Y . 1967 -.-)
~ C O O R D I ~ ~ D ~ T E D SCIENCE LABORATORY, -.
rn w k3 rn 0 A PROCEDURE FOR a RANKING DIAGNOSTIC
TEST INPUTS CI
( A C C E S S I O N N U M B E R )
/UNIVERSITY OF ILLINOIS - URBANA, ILLINOIS
https://ntrs.nasa.gov/search.jsp?R=19670023989 2019-03-02T22:08:20+00:00Z
T h i s work was supported by the Jo in t Se rv ices E l e c t r o n i c s
Program (U.S. Army, U. S. Navy, and U.S. A i r Force) under
Con t rac t D A - 2 8 043 AMC 000(73(E).
Reproduction i n whole o r i n p a r t is permi t ted for any purpose of t h e
United S t a t e s Government.
D i s t r i b u t i o n of t h i s r e p o r t is unl imited. Q u a l i f i e d r e q u e s t e r s may
o b t a i n copies of t h i s r e p o r t from DDC.
I 1 1 I
8 1 E 1 1 I 1 8 1 8 I t 1 I
a
iii
ACKNOW LEDGMEN T
The author is indebted to Professor Gernot Metze whose suggestions,
comments, and cr i t ic isms helped greatly during the research work and the
writing of this thesis.
The author is also grateful to Ralph Marlett and R o s s Schneider for
their helpful discussions and to Edward Davidson for reading through the man-
uscript and making suggestions.
The. support extended by the National Aeronautical and Space Admin-
is tration is appreciated.
iv
TABLE O F CON.TENTS
Page
1. INTRODUCTION. . . . . . . . . . . . . . . . . . . 1
2. PACKAGE LEVEL DIAGNOSIS. . . . . . . . . . . . . 4
2. 1 Prime Implicant Diagnostic P rocedure . . . . . . 6 2. 2 Probability Weighting Procedure . . . . . . . . . 9 2. 3 Assumed Number of Outputs. . . . . . . . . . . 11 2.4 Example with Probabilistic Weights . . . . . . . . 15 2.5 Comparison. . . . . . . . . . . . . . . . . . . 23
3. PROBABILISTIC WEIGHT FORMULA. . . . . . . . . . 29
3. 1 Weight Formula Derivation . . . . . . . . . . . 3.2 Multi-Output Example . . . . . . . . . . . . . 29
34
4. FUTURE WORK AND CONCLUSIONS. . . . . . . . . . 43
4. 1 4.2 Future Research Areas . . . . . . . . . . . . 4. 3 C onclu s ion s . . . . . . . . . . . . . . . . .
Sequential Machine Extension . . . . . . . . . .
LIST O F REFERENCES. . . . . . . . . . . . . . . . .
43 46 48
49
APPENDIX. . . . . . . . . . . . . . . . . . . . . . . 50
1. IN TR ODU C TION
Today, digital computers a r e being built larger and with greater
complexity.
flights, tele-phone systems, and plant control which require highly reliable
operation.
needed to detect and locate the faulty component shoula be minimized.
Such a faulty component will be called the fault.
cate and detect faults in computers a r e needed to aid maintenance personnel
with maintenance of computers.
They a r e being used in applications connected with space
When a component in the machine becomes faulty, the time
Rapid procedures to lo-
Let a machine be defined as the logical circuit of a computer.
This thesis will be mainly concerned with combinational circuits. A ma-
chine which has a fault will give a s e t of outputs different f rom that given
by the same machine with no faults, fo r some se t of inputs.
a machine can be detected and specified by examining the outputs of the ma-
chine and by comparing them with the different output sets that a machine
with the different possible faults produces.
ing to the individual faults have been developed [l, 3, 4, 5, 61. The pur-
pose of these procedures has been to find the specific fault in as few tests
as possible and as quickly as possible.
many of these procedures, and that a r e also made in this thesis, are:
1) The machine will have a t most one fault; 2) the possible faults that can
occur in a machine a r e a logic element stuck at "1" or "0, I ' and lines be-
Most faults in
Several procedures f o r diagnos-
The assumptions that a r e made by
1
2
1 I 5 5 1 8 t 8 c e I 1 f
tween the logic elements opened; and 3) the fault is well behaved, i. e. , i t is
not intermittent.
Many computers today are constructed in small packages.
which fault has occurred is not as important as knowing which package con-
tains the fault; for when a package has a fault, the whole package is re-
placed.
ducing redundant tests and thereby for selecting a near optimal s e t of tests
for diagnosing a machine.
ison of a machine containing a fault, to a fault-free machine.
for a machine with a single line output but does not consider the additional
Knowing
Chang [2] has taken this into consideration in his procedure for re-
H i s procedure is based on a pass or fail compar-
It is useful
information that a multi-output machine provides.
defined as an output with two output symbols, "1" and "0"; in this study each
output symbol will be referred to as an output.
problem of selecting a near-optimal s e t of tests for diagnosing a multi-
output machine with the goal of finding only the package wherein the fault
lies.
that can be applied to a multi-output machine.
A single line output is
This study considers the
This procedure therefore removes redundant tes ts f rom a s e t of tests
One criterion for this near-optimal selection of tes ts has been to
find a procedure that can be applied to large combinational logic circuits.
OptimalI procedures have been developed for small combinational circuits
[ l , 31, but if these procedures a re applied to combinational circuits of rea-
sonable size, the computations and the bookkeeping involved would be so
voluminous that they would render the procedures impractical.
The method developed by this study can be applied to large combin-
ational logic circuits since i t examines only the package separation that each
possible additional test gives when applied together with previously selected
tests. The test that gives the greatest separation of packages i s chosen and
added to the l i s t of selected tests.
always chosen, a t least a near-optimal se t i s selected.
not examine the package separations of several combinations of tests applied
together which optimum procedures must do, and therefore the bookkeeping
is greatly reduced, making this test procedure more practical.
Although an optimal s e t of tests is not
This method does
In Chapter 2 various procedures for package diagnosis a r e discus-
sed and a procedure f o r choosing a near-optimal s e t of tests is developed.
Chapter 3 contains the development of a probabilistic formula to calculate
probabilistic weights which a r e used f o r selecting the tests to be applied
for diagnosis.
one stage of an adder i s also found in this chapter. In Chapter 4 a possible
extension of the procedure to sequential machines is given along with other
possible a reas of research.
An example of finding a se t of tests with this procedure fo r
4
1. IC 8 3 8 8
t 8 8 8 I 8 I 8 8 8 1 I
2. PACKAGE LEVEL DIAGNOSIS
In diagnosing failures of a machine, several procedures have been
A test developed for selecting the se t of tests that i s applied to a machine.
i s defined as a possible input to a machine to determine the output of the
machine.
chine. One main goal of computer diagnosis i s to minimize the number of
tests fo r complete diagnosis of amachine s o as to shorten the time for di-
agnosis and reduce program space for a stored program data processor.
Most test selection procedures have been developed on a fault level diagno-
s i s basis [ I , 3, 4, 5, 61.
faulty circuit until the specific fault i s known or i t is known that no fault
exists in the circuit.
The output then is compared with the output of the fault-free ma-
Fault level diagnosis i s defined as diagnosing a
The trend fo r machine construction today i s to build several logic
circuits in the same package f o r the reduction of size and the greater ease
in replacing components. When a fault occurs, the whole package i s re-
placed. Fo r diagnosis of such a circuit olily the package containing the.'
fault need be found, and not the specific fault.
a machine should require fewer tests since less information is needed.
agnosing a circuit until i t is known which package contains the fault, o r un-
til i t is known that the machine contains no faults, will be called diagnosis
to the package level.
Therefore diagnosis of such
Di-
Few procedures have been developed fo r selecting tests for diag-
5
J 0 I 8 8 8
8 B 8 8 1
8 8 8 8 I I
nosis to the package level which determines the faulty package but not an
individual fault.
tes ts for diagnosing a machine to the package level.
useful only if the machine essentially has just two outputs, namely, "1"
and "0.
give m o r e information than just a "1" o r a "0" for each input test. Util-
izing this extra information in selecting a sequence of tes ts improves the
procedure for removing redundant tests.
Chang [2] derived a procedure for removing redundant
His procedure is
However, many machines have more than two outputs and hence
In diagnosing to the package level, a very general procedure that
requi res only a knowledge of the truth table of each package would be de-
sirable.
tions of the packages, diagnose the combinational circuit to the package
level.
each package on its inputs would be needed.
mined, then all possible inputs fo r each package a r e needed to make su re
that the package is fault-free. With dependency of the output on i ts inputs
known, different sets of tests that would indicate whether a package has a
fault would be needed, and f r o m the interconnection of packages a diagnos-
tic procedure would then be developed to diagnose to the package level.
This procedure would then, f r o m the knowledge of the interconnec-
F o r such a procedure, ways to detect dependency of the output of
If no dependency can be deter-
On the other hand, if the circuit of each package is known then i t
is known immediately which faults can occur and what s e t of inputs will
give a cor rec t output for the package even if a fault is in the package and
II I 8 - 8 1 8 t t 8 1 8 8 8 8 8 8 t 1 1
what se t will detect the fault. The problem of separating packages is much
less complex and the diagnostic procedure for the separation can be started
a t once.
known.
For these reasons i t is assumed that the logic in each package is
2. 1 Prime Lmplicant Diagnostic Procedure
Before considering the probability approach, i t will be instructive
to examine an optima1,prQcedure.
th Let p. represent the i package and l e t the possible faults within
1
th i i i the i package be f l , f2 , . . . , fk. Although diagnosis to the package lev-
e l does not require the identification of the separate faults within each pack-
age, each fault must be l isted to insure that a specific fault f . in package i
has been separated f rom the faults of the k
i J
package, where all other faults th
in package i may have been separated from those of package k.
fault must be separated by at least one test, in a s e t of tests, f rom all
faults in the other packages in order to diagnose to the package level.
Thus each
This
leads to a method for diagnosis with multiple outputs which is somewhat an-
alogous to the pr ime implicant problem for logic design,
test to a combinational circuit, different outputs will be given f o r the differ-
ent faults, depending upon the fault.
After applying a
Note that a fault simply changes the
fault-free machine to a different machine with different outputs.
these different outputs the following table (cf. figure 2) can be formed:
F r o m
6
7
f12
p2 i f22
Down the s
00 lo 11 01 00
01 00 01 10 00
ie of the table the different possible tests a r e lis-3d. Across
the top of the table a r e listed all possible pa i r s of faults, such that the pair
of faults is not f rom the same package. A t will be placed in the t m’
i k J n m J n fi./f? table position i f test t
words, if a machine with fault f . produces a different output f r o m a ma-
separates fault f . f rom fault f , or in other
i J
i n J
chine with fault 8. When a machine with fault f . produces the same out-
put a s a machine with fault ? , then the table position t
blank.
fi/k is left n . m’ J n
An example will clarify this procedure. Let a combinational c i r -
and p with two possible faults each. Let the 1’ P2, 3 cuit have packages p
five tests have four possible outputs f o r the various faults as shown in fig-
u r e 1. t l +2 t3 t4 +5
[f? I o 0 11 00 01 00
[ f13 I o 1 00 01 10 11
I
FR-1267
F i g u r e 1
8
8 8
The fault test pattern table is s,,own in figure 2.
fll/ff f l l / fZ fl’/f? f $ f 2 f21/ f$ f 2 / f , 2 f & f ? f2‘/f,3 f,Z/f? f;/f; f ,2/f? f;/f;
FR-1266
F i g u r e 2
Consider the pa i r s of possible faults, i. e. fi/F, as elements of a
be consider-ed a s a se t of these pa i r s and contain ele-
.
J ‘ 4 set. Then le t test t
ment f./? i€ and only if a plus sign is in the t
Then test t will be a pr ime implicant if and only i f any test t S r
t 3 t implies that t = t .
m i k J 9 m J q
f i / f entry of the table.
such that
Thus if a test is not a pr ime implicant i t is a r - S r s .
proper subset of a pr ime implicant test, while prime implicants a r e proper
subsets of none of the other tests.
plus sign in a row of t whenever there is a plus sign in the row t
thermore, since t is not a proper subset of t
plicant.
plicant tests according to this new definition of a pr ime implicant.
In the example, t 1 2 C t since there is a
Fur - 2 l’
o r t i t is a pr ime im- 3’ t4’ 5’ 2
The f i r s t step in the diagnostic procedure is to find the pr ime im-
9
The remaining step is the same as a step in the p r ime implicant
problem for logic design which is that of finding a minimum cover of the
fault pairs. A minimum cover is the minimum number of pr ime impli-
cants whose union will contain all elements fi/fk where i # k. In the ex- J 9
ample, t and t or t and t a r e minimum covers. Thus one concludes
that tes ts t
age level.
2 5 3 5
and either tes t t or t will diagnose the circuits to the pack- 5 2 3
This procedure is an optimum procedure for combinational c i r -
cuits with multiple outputs since a minimum number of tests can be found
by this minimum cover. However, one serious drawback in using this pro-
cedure for diagnosis is the vast amount of bookkeeping necessary to find a
s e t of tes ts t h a t will cover all faults. As an example, suppose that a
combinational circuit has five modules with ten faults in each module;
then the size of the fault pattern a r r a y similar to figure 2 will be the num-
ber of tests by 1000.
pr ime implicants, and finding the minimum cover woul8 be very t ime con-
suming.
Formulating this fault pattern table, finding the
2. 2 Probability Weighting Procedure
A more practical procedure is the following which o rde r s the pos-
sible tes ts by a probability procedure and finds a near-optimal s e t of di-
agnostic tests. For this procedure the tes ts and the outputs of the ma-
10
I I 1 I 1 8 8 1)
1
8 8 8 8 I I
~
chines with different faults a r e listed in tables a s in figure 1.
i of the tables represent the possible faults that cam occur. In the f . , t en-
J k
t ry is listed the output that the machine with fault f . would give if test t
were applied.
The columns
i
J k
Similar to Chang's method for selecting tes ts [2], a weight for each
test i s given f o r the purpose of selecting the best tes t to be applied next to
the machine for diagnosis.
set, of tests i s obtained by this approach while not requiring the exhaustive
tr ials of all possible test combinations. This procedure weights the possi-
ble tests that can be applied to the machine according to probability.
culating' thel pohahilitcjr it i's assunied.that?:ebc;h &utpawfor ever$ test i's equally
likely and that there a r e enough outputs available so that a hypothetical
test could completely diagnose the circuit.
ity that a hypothetical test will diagnose the circuit after the test being
weighted has been applied.
applied to the machine.
the tests a r e applied again to the machine to see how much more they will
separate the remaining packages in conjunction with the test already select-
ed. The probability that another hypothetical €est will separate all remain-
ing packages not separated by the tes t selected and the test being weighted
taken together, i s the new weight. A second test with the highest weight i s
added to the selected test set, and the process i s continued.
At least a near-optimal set , if not an optimal
In cal-
The weight is then the probabil-
The test with the highest weight is chosen and
Some separation of packages is obtained. Next,
Each time a
.
11
new weight is found f o r the tests not selected,
that a hypothetical test wil l separate the remaining packages not separated
by the test being weighted and the selected tests taken together.
packages have been separated the procedure is terminated, and the selected
test se t will'diagnose the machine to the package level.
This weight is the probability
After all
2. 3 Assumed Number of Outputs
To find the weight of each test, i t i s necessary to see how many pack-
ages would still need to be separated if the test to be weighted were appfied
i , a fault f . in a ma- J
af ter previously selected tests. If for a given tes t t m i
Q- J chine gives a different output from fault 8, then f . i s separated from fault
fk Of course the interest is not in separating faults but packages, and so
if all faults of a package have been separated f r o m all other faults in the
other packages, then this package has been separated.
i s important to determine whether the machine is fault-free, then the fault-
f ree machine can be considered as a fault of a fictitious package denoted by
f whose outputs for the tests are the same as the machine's truth table.
This will assure that the fault-free machine is separated from all other
packages with faults. 1
4'
Note also that if i t
0 0
In calculating the weights, i t i s necessary to determine how many
possible outputs need be assumed.
that a hypothetical test wi l l separate all remaining packages not separated
The weight of a test is the probability
I 1 8 I 8 8 8 t 8 8 8 I 8 8 8 1 8 I I
12
by the tests already selected and-the tes t being weighted taken together. A
snffjiient number of outputs needs to be assumed for this hypothetical t es t
in order for the probability not to be zero. The number of assumed outputs
may then be more o r l e s s than the number of actual outputs of the machine.
The number assumed is taken as the maximum number of packages that
have not been separated af ter applying the tests selected for diagnosis
along with any one of the tes ts being weighted.
assumption that each output for every fault in the next applied test is equally
likely to occur. Finding the probability is then simply a process of count-
ing the number of ways that outputs can occur, such that the remaining un-
diagnosed circuit will be diagnosed to the package level.
divided by the total number of possible combinations of outputs to give
what will be called the probabilistic weight, denoted W
of diagnosis the probabilities a r e the weights, and i t has been decided that
a weight of zero is not to be given to any tes t since this would eliminate a
comparison of all tes ts weighted zero. Therefore by using the maximum
number of packages left to be separated as the assumed number of outputs,
none of the weights will be zero.
The probability is under the
This number is
In this procedure P'
Let the probabilities of the various tes ts order the tes ts and call
this the test order assuming a number of outputs. This tes t ordering is
not necessary to choose the test with the highest weight, but i t has been
introduced to help explain some properties of this procedure and to make
8 I 8 1 8 8 8 8 8 t 8 P 8 8 I I 8 1 1
13
comparisons.
designated 1 and will be the test to be applied next for diagnosis.
The tes t with the highest probabilistic weight will have order
Assuming more outputs than the machine actually has, is equiva-
lent to putting the outputs of several tes ts together so that the machine ap-
pears to have more outputs. F o r example, i f only two outputs are possible
then four possible combinations of outputs, i. e., 2 can be obtained by put- 2
ting the outputs of two different tests together. If the number of outputs as-
sumed is equal to the machine's actual number of outputs ra ised to a power
r , then the weight is also the probability of diagnosing the machine with r
tests. . Of course the assumption of all outputs having equal probability for
all faults is the same.
Assuming more outputs may rank: one tes t higher than another,
whereas for fewer assumed outputs the tests may be ranked equally. An ex-
ample of this is seen in figure 3.
Both weights are equal, assuming two outputs; however, assuming
three outputs gives tes t t a higher weight. FOT test t and with three out- 2 2
puts, different combinations of two outputs among the three faults in pack-
age p2, with the third output for fault f2, can skparate the packages, where-
a s with two outputs all three of the faults in package @ had to have the same
output.
1
2
There is less than half the number of combinations for two outputs
among two faults than among three faults, and so with three assumed out-
puts, test t has the higher weight. 2
14
0.125 Assuming 2 outputs 0.296 Assuming 3 outputs wp { z::::
(Wp = Probabilistic weight 1 FR- 1269
F i g u r e 3
There a r e cases where assuming more outputs will switch the or -
der of tes ts that are close together in a test set, but usually the tes t order
will remain the same.
er total faults o r fewer packages left to separate.
possible outputs will increase all the probabilistic weights.
tes ts will not change orders , however, since a fewer number of faults o r a
fewer number of packages will still have m o r e combinations of outputs that
will separate them.
Tests that a r e weighted higher will often leave few-
Increasing the number of
Usually the
2. 4 Example with Probabilistic Weights
h order to make the procedure clearer , consider the diagnosis of
the following example that Chang used in his paper [2].
only two outputs, either 1 or 0; however, the multi-output probability
method is also applicable to this case since a single line output is a triv-
ia l mu1 ti- output.
His example has
The se t of tests that Chang used a r e shown in figure 4,
The test patterns can be written differently to see more clearly
which packages still need to be separated after the f i r s t test has been ap-
plied if the f i r s t test i s t. fo r i = 1, . . . , 8. This is done in figure 5(a) and 1
(b)
In figure 5(b) i t i s seen that t has a "0" in each package. The 1
1 3 1' f2 ' test t does give some information for diagnosis since i t separates f
1 5 1 and f f rom the other faults, and hence the probabilistic weight should not
be zero. An assumption of at least five outputs is required in order that
the probability of the remaining packages being separated by a hypothetical
test is not zero after applying test t Let these five outputs be called a, 1'
b, c, d, and e. A test pattern that will completely separate the packages
that have output "0" in test t must have the same output for all faults in
the package, and this output must not occur in m y of the other packages.
After referring to figure 5(b) one can see that one such possible output
1
pattern that would separate these packages is:
i 6
f l l 1 0 0 1 0 0 0 0
p1 { f ? I 0 0 0 1 0 0 1 1
I
Chang's Weight = 24 30 25 30 35 29 29 26
Chang's Ordering of Tests = e 2 7 2 1 4 4 6
Probabilist ic Weight = 0.0029 0.0177 0.0275 0.0472 0.4261 0.0531 0.0118 0.0059 x 10.'
Probabi li t y Ordering = 8 of Tests
5 4 3 1 2 6 7
FR- 1270
Figure 4
i i
+ l + 2 +3 t4 t5 t 6 t 7 t 8
p3 { :;
FR-1272
F i g u r e 5a
'1 +2 t 3 +4 t5 ' 6 t7 t 8
0 0 0 0
0 0 0 0 r ' i gu re 3b
FR-1273
I 1 8 I I 8 1 t I 1 I 1
8 8 1 I 1 1
e
.
I
I ,
1 0 LU
FR-1271
There a r e 51 ways for five outputs to completely separate the packages.
so there are 5 , ways for five outputs to occurcisince there a r e nine faults.
Therefore the probability of completely separating these remaining faults
9 is 51/51 . also separate the faults f
Al-
9
The next tes t to separate all packages af ter applying tes t t mus t
There are 5(4)(3) ways for five out-
1 5 , and f 1 3
1' f2 1' 3
puts to separate these faults out of a total 5
3 probability for completely diagnosing these faults ii 6 0 / 5 . possible test patterns. The
No dependency exists between the event of separating the 1 ' s in
figure 5(a) and the event of separating the 0 ' s in figure-5(b); and so the
probabilistic weight W P
two independent events.
is simply the product of the probabilities of these
The independence results f rom the assumption
I 1 I I 8 8 1 I 8
I 1 1 8 1 I 8 I 1
e
19
that every output is equally likely for every fault.
F o r tes t t the probabilistic weight is calculated: 1
9 W = (5:/5 ) ( 6 0 / 5 ~ ) = 7200/512 = 0. 0029 X l o m 2 P
By similar procedures the other probabilistic weights a r e
found and recorded in figure 4 along with the tes t order.
Test t has the highest weight and is therefore chosen as the tes t 5
to apply f i r s t in diagnosis. After applying tes t t the remaining test pat- 5
9
terns and packages to be separated a r e shown in figure 6. Figure 6 could
be split up again as figure 4 was in figure 5(a) and (b); however, this is
not necessary to calculate the probabilities. Since there a r e a t most
three packages left to be separated f rom each other, the maximum num-
ber of outputs needed is equal to o r l e s s than three.
separate packages p
Test t does not 1
and p completely, and- thus a t l eas t ;three out- 3’ P4’ 5
puts need to be assumed for calculating the weights.
Let u s choose t a s an example f o r calculating the probability.
and t have been applied to the circui t it can be seen that 5 6
6
After tes ts t
1 1 2 2 1 1 2 2 3 package p with faults f and f Z and package p with faults f and f need
to be separated. With three outputs there a r e 18 ways that these packages
can be separated. Since there a re four faults, the probability of separat-
4 2 1 ing p1 f r o m p is 18/3 . The probability of separating p with fault f
2 2 3
3 2 f r o m p with fault f is 2/3 under the same conditions. It is noted that
package p is completely separated, and so the probability of separating 5
20
Wp = 0.0049 0.0195 0.0073 0.0293 0.0439 0.0146 0.0012
Test Ordering by wp
3 5 2 1 4 7
Chong's Ordering 5 2 5 2 1 4 7 FR- 1274
Figure 6
2i
1 0 0 1 0 0
0 1 0 0 1 0
= 0.0313 0.5000 0.0156 0.0625 0.0625 0.0625
5 Test Ordering by wp
1 6
Chang's Ordering 5 1 6
2
4
2 2
2 3
FR-1275
Figure 7
I I I I I 1 1 I 8 1 B 1 1 8 I B 8 I I
22
4 i t is 1. 0 while the probability of separating p Multi-
plying these independent probabilities together will give the probabilistic
weight of the test.
f rom p 3 4 is 24/3
w = (~/3~)(2/3)(24/3~)(1. 0) = 0.0438 P
Other probabilistic weights a r e recorded in figure 6.
Since t has the highest weight, i t is ayplied and produces the test 6
patterns as shown in figure 7.
arated.
a r e recorded in figure 7.
3 weight, the only remaining packages to be separated a r e p with fault f l
and p with fault f
patterns a r e shown in figure 8.
Package p is omitted since i t has been sep- 5
By the same procedure the probabilistic weights a r e calculated and
After applying test t which has the highest 2
3 4
4 1' All other packages have been separated, and the test
W p = 0.500 0.500 0.500 0.500 0.500
Test Orders 1 1 1 1 1 FR- 1276
Figur 'e 8
2 3
All probabilistic weights a r e the same as the weight of test t2 and
with the same number of assumed outputs. Therefore there is no further
separation obtained by applying any of the remaining tests which indicates
that faults f and f a r e indistinguishable. The diagnosis process then is 3 4 1 1
2' terminated with the application of tests t 5, t6s and t
2. 5 Comparison
It is instructive to compare the s imilar i t ies and differences be-
tween Chang's ordering of the tests and the probabilistic ordering.
time the tests were weighted in this example, the tes t with the highest
weight by Chang's procedure also bad the highest probabilistic weight.
Therefore the test selected to be addednext to the diagnostic s e t of tests
was the same by both procedures.
separated there must be a t least three tests f o r separating all the packages,
and if one examines all other possible combinations of three tes ts , he will
s ee that tests t
both procedures the best possible tests were selected,
Each
Since there a r e five packages to be
and t a r e the best combination for diagnosis. In 5' t6' 2
After ordering the tests in f igu re 4 i t i s noted that the grobabilis-
tic weights ordered test t as second while Chang's procedure ordered 6
test t a s second and test t as fourth. By examining these two tes ts one 2 6
sees that tes t t does not separate any of the packages from the rest . Ev- 2
e r y package after applying tes t t has a zero for at l eas t one fault, and i f 2
24
this test were selected, a minimum of three additional tes ts would be r e -
quired to separate all packages, (Since two of these faults in different
packages have identical test signatures, i. e . , the same outputs for all
tes ts , they a r e indistinguishable; however, this is assumed not known until
the end of the diagnosis.
tes ts would be required. ) On the other hand, tes t t separates package
and pl.
If they could be separated, a t l eas t three m o r e
6
completely f r o m packages p There remain four packages to p5 4
be diagnosed that have 0 ' s recorded under tes t t
have 1 ' s recorded.
completely diagnosed with just two remaining tes ts i f tes t t
t
and three packages that 6
Thus there is a possibility that the circuit could be
is used. (Tes t 6
has been ignored for this comparison. ) 5
For this specific example with test t omitted, the same additional 5
tests a r e required whether tes t t
indicates the reason for the difference in ordering.
slightly more the splitting of the remaining faults to be diagnosed into two
equal pa r t s while the probabilistic weight simply selects the tes t with the
highest probability of complete diagnosis with a certain number of additional
tests.
into equal separations is the test with the highest probabilistic weight as
was seen in the selection of the best tests by both methods in Chang's ex-
ample.
is chosen or tes t t 2 6' but the comparison
Chang's method weights
Of course in many cases the tes t which spli ts the remaining faul ts
Exarnining a select set of test patterns will show m o r e clearly
25
some differences of ordering tests between the probabilistic weights and
Chang's weights. This s e t will also indicate one weakness of both proced-
ures.
Consider the set of tests and their weights a s shown in figure 9.
1' t6' Two se ts of tests will completely separate all faults; either tests t
2 , tg, t4, and t t and t or tests t
and t of the first s e t highest while the probability u r e weights tests t
It can be seen that Chang's proced- 5' 7' 8
1' t6' 8
weighting procedure weights tests t and t of the second se t highest. No- 2 5
tice that every test in the f i r s t set of tests except one has the same number
of 0 ' s and 1 's ' while every test in the second se t separates one package
each. Chang's procedure weights tests t t o r t highest since the 0 ' s
and 1's a r e divided more evenly in them. The probability weighting pro-
1' 6' 8
cedure weights tests t and t 2 5 highest since there a r e more possible com-
binations with five outputs that would separate the remaining four packages
than there a r e combinations which would separate the five packages. After
applying either test t
separate the remaining packages with two tests together, whereas a mini-
mum of three tests is requiredfor complete diagnosis after test t t o r
t
or t 2 5 there a r e output patterns that would completely
1' 6'
Admittedly fo r this example, two tests that would separate the remain- 8'
ing packages after applying t o r t would have the highest weight of the 2 5
tes ts listed and would be chosen f i r s t by both methods; however, this is not
always the case for se t s of packages and tests. The probability weighting
I I I I I U 1 D
1 D I 1 1 I 1 1 I n
e
26
Ch ang's Weight 29 27 2 0 20 27 29 28 29
Chang's Test Order 1 5 7 7 5 1 4 1
Wp Test Order 5 1 3 3 1 5 5 5
wp = 0.0059 0.0430 0.0111 0.0111 0.0430 0.0059 0.0059 0.0059
FR- 1277
F i g u r e 9
27
procedure chooses the test that a t l eas t rnakes a shorter dLagnosis possible.
If eight outputs a r e assumed in the probability weighting procedure,
i t tarns oirt that tes+ + has a higher probabilistic
order switches, indicating that with an additional
there exist more combinations that will separate
&l weight than t Thus the 2'
3 three tes ts (or 2 outputs)
the remaining packages
for tes t t than for test t By assuming three
ordering and the probabilistic ordering a r e the
However, this approach ignores the possibility
This points out one of the reasons for
sible number of assumed outputs that will st i l l
1 2' additional tests, Chang's
2' same for tes ts t and t
of diagnosis with fewer tests.
choosing the smallest pos-
give all tes ts a weight other
1
than zero.
correspond more closely to the minimum number of tests that will diagnose
the circuit.
outputs is to make the computations of the weights easier.
By choosing the smallest number of outputs, the probabilities
Another reason for choosing the smallest possible number of
Neither Chang's procedure nor the probability weighting proced-
u r e always chooses the optimum test sequence for all cases. This can be
seen if the only tes ts available for diagnosing the packages in figure 9 are
and t Chang's procedure wouldgick tes t t f i r s t and 5' 1 t es t s t l , t2$ t3, t4?
then the other four tests.
a r e sufficient to diagnose the circuit to the package level.
1, t2? t6, t7, and t 8' i f the only available tes ts are tests t
ility weighting procedure would pick test t
Test t is not needed since the other four tes ts 1
In the same way,
then the probab-
f i r s t and then the other four 2
28
i a
tests.
age level since the other f o u r tests a r e necessary and separate the packages
Test t in this case contributes nothing to the diagnosis to the pack- 2
themselves.
As was pointed out in the introduction, optimality was traded for
the process of choosing the next best test without looking a t all possible
combinations of tests. A glance at the two cases above yields the optimum
I I 8 I B I 1 I
se t of tests because of the simplicity of this problem and because the eye
can see all possible combinations of tests. But of course in la rger com-
binational logic circuits, considering all possible combinations of tests is
an enormous task.
I I
1 1 1 1
3 . PROBABILISTIC WEIGHT FORMULA
3 . 1 Weight Formula Derivation
Calculating the probabilities can be a fairly tedious job i f done by . hand. In order to facilitate the matter, the following i terative formula has
been developed to permit the probabilities to be calculated by computer.
The following recursive formula calculates the number of possible ways
of assigning NT possible outputs to N packages.
Define :
NX (I1, 12, . . . , I N ) = 1
and
NX (I1, 12, . . ., Ik, 0, 0 , . . ., 0 ) =
F R - 1 3 0 3
T = NT - N - I1 - I2 - ... - I k t k t 1
NT = Total number of outputs assumed
N = Number of packages to be separated
th = Number of faults that are in the k t 1 package
N F k t 1
Each I . represents the number of outputs that appear i n the
j package, for j = 1, 2, . . . , k t 1.
J th
NX (I1, 12, . . . , Ik, 0, . . . , 0) i s the number of ways that T t N - (k t 1 )
outputs will separate all remaining packages indexed greater than k, with
29 I
th I. outputs in the j package for j = 1, . . . , k. J
30
The number of possible combinations of NT outputs that will com-
pletely separate all packages is given by NX(0, 0, . . . , 0). The probabilis-
t ic weight is given by:
NX(0, 0, . ... 0) . w = P N T ~ ~ ~
9
(the total number of faults). where N F T = F N F k k= 1
In order to verify that this formula calculates the probabilistic
weight as desired, consider the individual, terms. The maximum number
of outputs T that may appear in a package and sti l l leave a&sufficient number
of outputs to cover the remaining packages is the total number of assumed
outputs NT less the number of outputs assigned already, and less a t l eas t
one output for each of the remaining packages. Therefore
- (N - ( k t l ) ). The minimum number required is - k T = NT - I - Iz - 7 . - 1
1. The t e rm
is the number of ways that outputs of the total number of outputs NT
less the previously used outputs can be selected. These I outputs are k t 1
then applied to the faults of package k+l.
k t l The number of ways that these \tl outputs can be assigned to N F
I I I I I I 1 I I I I I. I I I 1 I I 1
31
faults is given by:
- 1 NFkt 1
J k t l
('kt1 - Jk+l ' ') p (-1)
= 1 J k t 1
19 X2' x3, X4$ e.. 9 x I' To verify this, consider I outputs labeled x
N F There are (I) ways that these outputs can be arranged where N F is the
number of faults in this package. But we seek only those arrangements
that have used all I outputs; we do not want to include those combinations
with fewer than I outputs. There a r e (:)(I - l)NF combinations that have
NF (I - 1) outputs that have been included in the number (I) , and so we must
N F subtract (;)(I - l)NF from (I) . Next we must consider how many com-
binations of (I - 2) we have so far. The t e rm (I)NF has (;)(I - 2)NF too
(:> many. To see how many of these have already been subtracted in the
(I - l)NF te rm consider the matrix of outputs:
x x 2 3 4
X ... I X
I x - x3 x4 ... x 1
I X x4 *' "
x - x1 2
1 X x x
2 3 4 X ...
The dashes indicate the output missing in the listing of the (1-1) combina-
tions.
'I I I I I I I 1 1 I I I I I 1 I I I I
32
Notice that there are ways to choose any two outputs omitted
and that there a r e rows of the matr ix that can have the same remaining
outputs with one more output omitted. Fo r example, the f i r s t two rows
3' x4' omitting x in the second row and x in the first row have outputs x 1 2
Since all other outputs have an x., deleted for 3 L j I I, all . * * y xl' J
other rows with one more output deleted will have either x o r x 1 2 o r both.
Since a row of the a r r ay represents one of the te rms of (I - 1) outputs,
the number of combinations of (I - 2)NF has been subtracted twice, once too
many, and therefore ( i) (I - 2)NF must be added to (I)NF - ( i)(I-l)NF.
Consider now the number of (I - k) outputs. These combinations
also must be removed f rom the total number of combinations since only I
output combinations are wanted. There are (:)(:)(I - k)NF combinations
in the (i) (I)NF term; (;)(:)(I - k)NF combinations in the
t e rm that have been subtracted and (;) (L ) (I-k)NF combinations in the
(;)(I - 2)NF t e rm that have been added. The validity of this l a s t number
can be demonstrated by the array:
8 I I I 1 I 1 I
3 X 4 X ... I X
I X x4 . * * - x -
2
2 x 3
X ... I X
I X x4 . . * x - - 1
I x - x - ... x 1 3
x1 x2 x3 x4 * . - -
There are (:)ways to choose the k outputs to be deleted. Of
these k outputs there are (;)ways to choose two that have already been de-
leted f rom the rows of the array. Hence there a r e (:)(:)(I - kJNK com-
binations that have k outputs deleted in the t e rm (;)(I - 2 l N ~ 4 By a simi- . - ( )(:)(I - k)NF com- term has included N F m lar argument, the (L) (I - m)
N F binations of k outputs. Therefore the coefficient of (-;)(I - k)
t e r m forms a binomial distribution.
for each
Summing these up to see how m a n y
t imes (;)(I - k)NF need be added o r subtracted indicates that just one
t e r m is needed since the final term of the binomial coefficient is 1 and
k= 0
Tc+ 1 Thus the t e rm in (2) gives the number of combinations of
specific outputs where each of these outputs is used a t least once. The
C I I I 1 1 8 8 1)
I I 8 1 I 1 1 f 1 I
34
product of these two t e rms (1) and (2) gives the total number of allowed com-
binations that can occur with outputs.
The te rm NX(I1, 12, ..* I I 0, ..., 0) ( 3 ) k' k t l '
gives the number of combinations that the packages indexed greater than
possible outputs. By multiplying Ik t 1 k t l have with NT - I1 - I2 - ... -
i t is c lear that k t l ' ( 3 ) by (1) and (2) and summing the possible values of I
NX(Il, 12, . . . , Ik, 0, 0 , . . . , 0) is found and gives the total number of
combinations that will separate completely the packages indexed greater
than k with I outputs in the f i r s t package, I
and so forth up to J, outputs in the k
outputs in the second package, 1 2 th
package.
After NX(0, 0, . . . , 0 ) is calculated, the probabilistic weight is
given by :
NX(0, 0, . . . , 0) NFT w =
P NT
NFT The t e r m NT i s the total number of ways that NT outputs can
be assigned to the total number of faults, NFT.
A program that calculates the probabilistic weight for up to three
packages is included in the Appendix.
3. 2 Multi-Output Example
The example selected is a single stage of a paral le l adder com-
posed of two half adders and one "OR" gate. The circuit is shown in figure
1 1c I c I 1 8 1 8 8 8 I E E I 1 f I 1
35
10.
The faults and tes t inputs were simulated on the simulation portion
of Seshu's Sequential Analyzer [ 6 ] , and the dictionary of possible machine
fai lures is shown in figure 11.
between packages were included arbi t rar i ly in one of the packages. The
fault R1 to CZ OPEN was included in the package H2 while faults C1 to C
Some of the possible faults that could occur
OPEN and C2 to C OPEN were included in package C3. The fault-free ma-
chine called the good machine was considered as a separate package to as-
sure that all faults would be separated f rom i t for complete diagnosis.
The tes t fault pattern in figure 12 l i s t s the various faults that cor-
respond to the dictionary of machine f a i l u r e s . in ;figure 11. The outputs are
shown in the entries of figure 12.
been recorded as well as the test order.
Under the tables the probabilities have
The figures f rom figure 12 to
f igure 17 show the complete diagnosis of the full adder stage to the package
level.
The full adder stage was completely diagnosed to the package level
with tes t inputs 010, 011, 101, 001, 000, and 100. T w o possible tests were
not required, and in three cases the best choice was arbi t rary between two
tests.
the same package since the locations of these faults were close to each
other in the physical circuit. Since these indistinguishable faults were in
the same package, no weight was given to separating them, and therefore
no useless search fo r a tes t that would separate them was made.
It can be noted that a few indistinguishable faults were included in
3 ‘7
DICTIONARY OF MACHINE FAILURES FOR LOGIC CIRCUIT CKTl
MACH NOS FAILURE MACH NOS FAILURE
1 GOOD MACHINE 2 A T O C 1 OPEN
3
5
7
9
11
13
15
17
19
21
23
25
27
29
LC
c1
LC
01
N1
R1
B
c 2
B
0 2
c 2
C
02
R2
TO C1 OPEN
OUTPUT 1
TO 01 OPEN
OUTPUT 1
T O R 1 OPEN
OUTPUT 1
TOCZ OPEN
OUTPUT 1
TO 02 OPEN
OUTPUT 1
TO C OPEN
OUTPU T 1
T O R 2 OPEN
OUTPUT 1
INPUT ORDER A B LC
OUTPUTORDER R2 C
F i g u r e 11
4
6
8
10
12
14
16
18
20
22
24
26
28
c 1
A
01
01
R1
R1
c 2
R1
02
c 1
C
N2
R 2
OUTPUT 0
T O 0 1 OPEN
OUTPUT 0
T O R 1 OPEN
OUTPUT 0
T O C 2 OPEN
OUTPUT 0
TOO2 OPEN
OUTPUT 0
T O C OPEN
OUTPUT 0
T O R 2 OPEN
OUTPUT 0
- 2 3 4 5 6 7 8 9 10 11 12 13
14 15 16 17 18 19 20 21 26 27 28 29
-
-
38
Test ( Inputs 1 000 001 010 011 100 101 110 111
00 01 10 11 10 01 01 11 00 10 10 01 01 01 11 11 00 10 10 0 : 10 10 01 01 01 0 1 11 11 01 01 11 11 00 10 10 01 00 01 10 11 00 00 10 10 10 01 01 11 00 00 10 10 00 01 10 11 10 10 01 01 10 01 01 11 10 10 01 01 10 01 01 11 00 10 10 01 10 11 01 01 00 00 10 10 00 01 10 11 10 10 01 01 10 11 01 01
00 00 00 01 00 0 0 00 10 00 10 00 10
10 01 01 01 10 01 10 10 10 01 01 01 00 10 01 10 00 01 00 00 01 10 10 01 10 10 11 10 10 01 00 00 01 10 10 11
10 01 01 0 1 01 01 01 11 10 01 10 11 01 01 01 01 00 01 01 11 10 01 01 01 00 01 01 01 10 11 01 11 10 01 11 11 10 11 01 11 00 01 01 01 10 11 11 11
00 10 10 01 10 00 01 lo 00 10 10 00 10 01 00 11 00 10 10 00 10 00 00 10 01 11 11 01 11 01 01 11
00 10 10 01 10 01 01 11
3.1812 2.5365 2.8161 2.6090 2.5365 0.2347 2.6090 0.617' x 10-1
8 4 1 2 4 7 2 6 FR- 1279
F i g u r e 12
I 1 I I 8 1 E 1 1 I I I 1 8 8 8 t I I
39
Test (In Package 8 Fault No.
2 H l [ i 3
8 11 12
15
H2 1 26 :! 27 29
-
-
M { 1- H1{ 5
01 1 100 101 110 000 00 00 00 00 00 00 00 00
00 00 00 10 00 10 10
00 00 00
00
--_
---
---
001 01 10 10 10 00 00 10 00
01 10 00 10 10 10 10
lo 10 10
lo ---
111 10 01 10 00 10 00 10 00
01 10 00 10 10 10 10
lo 10 10
lo
--
--
---
01 01 10 01 01 01 11 01
01 01 01 11 01 1 1 11
00 01 00
01
---
---
-
01 11 01 10 01 10 01 10
01 10 01 01 11 01 11
01 00 00
0 1
---
---
11 11 01 11 11 11 01 11
11 11 11 11 11 11 11
lo 11 10
11 ---
11 01 O! 01 10 10 01 10
01 10 01 01 11 01 11
01 00 00
01
11
01
01 01 01
01 01
--
--
--
_-
01
11
01
11
-- 01
01
--- 11
01
11
11 - - .
01
01
10 10 10
00 0 1
I c 3 { 2 5
10 10 10
lo 01
3.047
--
-
4
01 01 11
01 01
---
01 01 0 1
01 01
---
11 11 01
01 01
---
10 10 10
lo 01
Hl[ l”0 13 - I
j:, ~
1.0042
7
0.0477
4
1.082
1
0.1485
3
1.0821
1
0.0094
6 x10-f
FR- 1280 Figure 13
I I I 8 1 8 I E I D 6 1 8
Package 8 Fau I t No.
3 4 6 11
15 18 21 27 -
22 - 1
2 26 29
-
H 2{ 16
Hl[ 1;
13
WP
TO.
Test ( InDuts) 000 001 100 101 110 111
00 01 10 0 1 0 1 11 00 10 10 0 1 11 11 10 10 10 11 11 11
-------------------
00 00 10 01 0 1 11 00 00 00 01 10 11 00 00 00 01 10 11
.-----------__------- 00 10 10 0 1 10 11
10 10 10 01 01 11
10 10 10 01 01 11
10 10 10 11 01 01
00 10 10 0 1 0 1 01
0 1 0 1 01 0 1 0 1 0 1
0.1159 0.1545 0.5866 4.1062 0.0410 1.7598
5 4 3 1 6 2
.-------------------.
x 10-3
40
F R - 1 2 8 l
Figure 14
R
Package 8 Fault No.
Test (Inputs) 000 001 100 110 111
11 l1 I 10 01 11 10 00 10
11 11
00 10 10 01
10 10 10 01 10 10 01
------- ------- H 1 { 11
H 1 { 2
H 2 ( 2 6
H1 [I H2 { 16 --
H1 { ,'o
00 00 10 01 11 00 00 00 10 11 00 00 00 10 11
00 10 10 10 11
10 10 lo 01 11
10 10 10 01 11
00 lo 10 01 01
01 01 01 01 01
----------- --- -
__--------- ----
Wp 0.4877 8.779 3.9018 1.9509 0.4877
TO. 4 1 2 3 4
x 10-2
F R - 1282
Figure 15
8 8
000 30 00
00
00
10 10
10 10
00
1.2500
- - - -
_ _ -
- - - --
1
package 8 Fault No. 100
01 00
10
- - - -
10
10 10
10 10
10
0.0625
3
- -- -
- - - -
H 1 { 11 -_
Mb WP
TO.
Ii2 {
_ _ _ _ _ _ 10 01
1.00 1.00
1 1
H1 {Po
H2 { 14
WP
T.O.
Test ( I n p u t s )
1 10 11 10
01
----
01
01 01
01 01
01
0.0625
3
111 11 11
11
01
1 1 11
--
1 1 11
---- 0 1
0.250C
1 FR- 1283
F i g u r e 16
q 0.25
FR-1284 F i g u r e 17
43
4. FUTURE WORK AND CONCLUSIONS
4. 1 Sequential Machine Extension
The probability weighting procedure could be extended to sequen-
tial machines.
following tables :
A possible extension would involve forming and using the
(Entries consist of an output plus the next state the machine would go to.)
F i g u r e 18
. . .
. . . FR - 1285
It is assumed that there a r e k states and s packages with a vary-
ing number of faults in each package. A row of the table represents a fault
I 0 1 I 8 1 8 1 8 I 8 I 8 8 8 t 8 1 1
in a given package with i ts outputs f o r various test inputs.
represents a test input with the machine in a specific state.
the tes t tq
machine in state q.
bookkeeping since a fault sequential machine may not only have a different
output but also a different next state when compared with the fault-free ma-
chine.
test as well a s the next state that the machine with the fault would go to.
In applying the probability weighting procedure i t is assumed that an initial
state of the machine i s given or that the machine can be put into a given
state.
tion the packages and shorten the diagnostic process.
Each column
F o r example,
i s the test input t applied to the sequential machine with the m m
The states of the machine must be accounted fo r in the
Therefore each table position entry has the output for the applied
If the state that the machine goes to is observable i t will also parti-
If state i is the initial state then the subarray shown in figure 19
i s formed by all columns of tests with superscript i and all rows.
When this a r r ay is used, the probabilistic weights a r e found and
the f i r s t test to be applied is selected.
the a r ray , f rom which to find the next test, i s obtained by f i r s t examining
the next state that each fault would put the machine in.
plying the f i r s t test puts the machine in state h, then the portion of the f
row under column t
i fo r row f The tests of the new ar ray drop their superscripts. After the
a r ray has been completed i t contains the condition of the machine after the
After the f i r s t test has been applied,
i If fault f . after ap-
3 i j
h h 1’ * . * ’ n
t of figure 18 would be placed in the new a r ray
j’
45
8 8 8 8 I 1 8 1 1
p1 I p2 1
8 I 8 B 1
FR - 1286
Figure 19
secopd test has been applied, This a r r ay thus shows the outputs that the
sequential machine would give and i ts next state for any given fault after
applying the first tes t and any second test.
then calculated and a next test selected.
the machine is diagnosed.
The probabilistic weights a r e
This process is continued until
4. 2 Future Research Areas
Applying probabilistic weights to sequential machines will require,
in addition, a procedure for determining the next test to be applied when
the present possible tests do not give any additional diagnostic information,
but other tests in other states do,
the diagnostic process so that all packages a r e separated that can be, but
indistinguishable faults in different packages do not cause useless tests to
be inserted in the sequence of diagnostic tests.
l em is to provide the detailed cri teria fo r the extension of this probability
weighting method to sequential machines.
A criterion is needed for terminating
A possible research prob-
The probability weighting procedure could b e altered in two ways
that may improve the procedure fo r combinational machines.
ity would be to purposely weight some of the tests zero in order to shorten
the time for calculating the weights.
each time is to select the best test that would next be applied to the machine.
The order of the tests i s unnecessary especially f o r those tests which give
the least information.
zero would not prevent selecting the best test to be applied.
weight of zero to some tests could be done by assuming fewer outputs for
the hypothetical test so that this test cannot separate the remaining pack-
ages.
four outputs a r e assumed possible for the hypothetical test, then the probab-
One possibil-
The purpose for weighting the tests
Weighting the tests that give the leas t information
Giving the
For example, if five packages remain af ter applying a tes t and only
i l ist ic weight is zero. Other tests that reduce the number of packages left
to be diagnosed sufficiently would sti l l have non- zero probabilistic weights
which would provide the means f o r selecting the best test. Under this new
consideration a procedure would have to be developed for determining how
many outputs should be assumed for the next test.
puts may jeopardize the selection of the best tes t while too many assumed
outputs would not take the greatest advantage of the shortened procedure.
Too few assumed out-
Another variation of the probability weighting procedure to consider
is the calculation of the probabilistic weights based on two or three tests
taken a t a time in order to select the best two or three next tes ts for diag-
nosis.
s e t of diagnostic tests but would require more bookkeeping and computa-
tions.
This approach would increase the chances of achieving an optimal
F o r certain problems this may be desirable.
A further a r e a of research would be to examine m o r e closely diag-
nosis to the package level with only a truth table for each package.
sumptions are necessary or useful in forming such a general diagnostic
procedure ?
What as-
A multiple output optimum procedure that would diagnose to the
package level is another research problem,
knight's [l] might be considered.
A procedure similar to Bou-
What types of elements should be included in packages that would
lend themselves better to diagnosis to the package level? Since package lev-
I I I I 8 8 0 I 8 1 I I 8 0 1 1 I I I
, 48 i
e l diagnosis is possible, could this be extended to a computer block diagno- ~
s i s ? If so, what a r e the advantages? The disadvantages? These questions
suggest further a reas of research. I
4. 3 Conclusions
This investigation has produced an instructive, although not prac-
tical, optimum procedure fo r diagnosing multi-output machines to the pack-
age level somewhat analogous to the pr ime implicant problem in logic de-
sign. The investigation a l s o has produced a procedure f o r selecting a se t
of tests sequentially to diagnose multi-output machines to the package level,
thereby eliminating redundant tests.
select the optimum set, i t does select a near-optimal s e t for combinational
machines. The probability weighting procedure has the advantage that the
s e t of selected tests is found without exhaustive t r ia l combinations of tests
which other optimal procedures require.
used f o r diagnosis one at a time. After applying previously selected tests,
the test that has the largest probabilistic weight is chosen and added to the
set. The probabilistic weight is the probability that a hypothetical test
would completely separate all remaining unseparated packages after the sg-
lected tests and the test being weighted had been applied to the machine.
Since diagnosis is accomplished to the package level, fewer tests a r e usual-
l y required than would be fo r diagnosing to the fault level.
Although the procedure may not always
The tes t s a r e added to the se t
LIST O F REFERENCES
49
I 8 8 I I 8 I I
1. Bouknight, W. J . , "On the Generation of Diagnostic Tes t Procedures , Coordinated Science Laboratory Report 292, University of Illi- nois, May 1966.
2. Chang, H. Y., "An Algorithm for Selecting an Optimum Set of Diag- nostic Tests, I ' IEEE Transactions on Electronic Computers, Volume EC-14, Number 5, pp. 706-71'0, October 1965.
3. Howarter, D. R . , "The Selection of Failure Location Tests by Pa th Sensitizing Techniques, I ' Coordinated Science Laboratory Report 316, University of I l l inois , August 1966.
4. Seshu, S. and Freeman, D. N . , "The Diagnosis of Asynchronous Se- qsmtia2Switching Systems, I ' IRE Transactions on Electronic m u t e r & , Volume E C - 1 1 , Number 4, pp. 459-465, August 1962.
5. Seshu, S. , "On an Improved Diagnosis Program, Coordinated Science Laboratory Report 207, University of Illinois, May 1964.
6. Seshu, S . , "The Logic Organizer and Diagnosis P rograms , Coordi- nated Science Laboratory Report 226, University of Illinois, July 1964.
APPENDIX
The program SEPPROB is a FORTRAN program for finding the
probability of separating up to three packages.
MNP is the maximum number of outputs, CKN is the number of tests whose
probabilities a r e to be calculated, NM is the number of packages, and
th N F ( i ) , the number of faults in the i package.
The inputs a r e a s follows:
program sepprob
dimension
'nf(31,jj ( 3 ) , 4 16,16,16)
c o m n nf, jj, nx, mno, U, nm, mindx
C mno=nwdmumnumber of outputs. ckn=nmber of tes ts
read 1 , rum, ckn
1 format ( l i2 , l f3 .1)
ck.jd'.(2l
2 read 3, m,nf(l ),nf(2),nf(3)
3 format (4i2)
ck=ck+l.@
do 4 k=l,mn
4 33 (k)=a'
18h module ,2( 1 i2,l h, ), 1 i2,/, lx, 1 bhprobability = ,l f7.5 1 if (ck. eq. ckn) 1 g,2
19 end
C t o cdculate nx( i1,i2,g,j~f) from stan over j of nx( il ,i2, j,$)
subroutine nxtv(nfk)
dimension
W31,JJ (3),d6,16,16)
c o m n nf, jj, nx, m, 11, nm, mindx
call nindxm(maxi,nfk)
if(nfk.eq.la’) 56J3
53 maxj=mindx+nm-U--l
do 54 nn=l,maxi
imodcg
do 55 mm=l,nn
I I I I 8 8 I D 1 I B 1 I 1 1 I I I I
mmlo=mn-l
55 imodc=imOdc+(-1 ) ' 'mmlo'nbinomc(nn,mmlo) '(nn-nmiLo)"nfk
D t 1 8 8 8 I I 8 1 I 1 I 1 I 1 I 1 1
79
81
w
82
83
84
173
1 74
176
fm-m
binoslc=fnn/rr
nrrlo=nrr-l
do 176 k=l ,nrrlo,l
dl a - k
s~b2-m-k
subm=sub 1 /sub2
binom=binmc'stibm
nbinomc=binomc
DISTRIBUTION LIST A S OF APRIL 1,196'7 1
1
1
1
1
50
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
1
1
1
1
Dr. Edward H. R e i l l e y A s s t . Di rec tor (Research) Ofc. of Defense Res. h Engrg. Department of Defense Washington, 0 . C . 20301
O f f i c e of Deputy Direc tor (Research and Informat ion Rm. 3D1037) Department of Defense The Pentagon Washington, D . C . 20301
Direc tor Advanced Research Projects Agency Department of Defense Washington, D. C . 20301
Ulrector f a r M a t e r i a l s Sc iences Advanced Research ProJects Agency Department o f Defense Washington, D. C . 20301
Headquarters Defense Cornunica t ions Agency (333) The Pentagon Washington, D. C . 20305
Defenoe DOCYmentatiOn c e n t e r A L L = "ISIA Cameron S t a t i o n , Bldg. 5 Alexandr ia , V i r g i n i a 22314
D i r e c t o r Nat iona l S e c u r i t y Agency Arm: TDL F o r t George G. Heade, Maryland 20755
Weapons Syatems Evalua t ion Group Attn: Col. Danie l W . McElwee
Washington, 0. C . 20305
Nat iona l S e c u r i t y Agency Arm: RCJarnen Tippet
F o r t George G . Heade, Maryland
C e n t r a l I n t e l l i g e n c e Agency Arm: OCRIDD P u b l i c a t i o n s Washington, D. C . 20505
Colonel Kee AFRSTE Hqr. USAF Room 1D-429. The Pentagon Washington, D. C . 20330
Colonel A. Swan Aerospace Hedica l Div is ion Brooks A i r Force Bare, Texas 78235
AllL3T-9663 Haxwell AFB, Alabama 36112
AFFTC (FTBPP-2) Technica l L i b r a r y Edwards AFB, C a l i f o r n i a 93523
Space Systems Divis ion A i r Force Systems Command Los Angeles A i r Force S t a t i o n h s A n g e l e ~ . C a l i f o r n i a 90045 Atfn: SSSD
Major Char les Waeopy Technica l Div is ion Deputy for Technology Space Systems Divis ion , AFSC los Angeles, C a l i f o r n i a 90045
Department of Defense
O f f i c e of Research 20755
SSD(SSTRT1Lt. S ta rbuck) AFWO Los Angeles, C a l i f o r n i a 90045
DeL. 116, OAR (LWAR) Air Force Uni t Pos t O f f i c e Los Angeles, C a l i f o r n i a 90045
Systems Engineer ing Group (RTD) Technica l Informat ion Reference Branch At tn : SEPIR D i r e c t o r a t e a f Engineering S tandards
Wright -Pa t te rson Am, Ohio 45433
ARL (ARIY) Wright -Par te r ron AFB, Ohio 45433
D r . H . V. Noble Air Force Avionics Labora tory Wright -Pa t te rson AFB, Ohio 45433
Hr. Peter Hurray A i r Force Avionics Labora tory Wright -Pa t te rson AFB, Ohio 45433
AfAL (AVTE1R.D. Larson) Wright -Pa t te rson AFB, Ohio 45433
Commanding Genera l Attn: STEWS-US-VT WhlCr Sands H i s s i l e Rangc, N ~ W Mexico 88002
6 Technica l Informat ion
RADC ( W L - 1 ) G r i f f i s s A F B , N e u York 13442 A t m : b c u r n e n c ~ L i b r a r y
Academy L i b r a r y (DFSLB) u. s. Ai? Force Academy Colorado S p r i n g s , Colorado 80912
LL. Col. Bernard S . Morgan Frank J . S e i l e r Research Labora tory u. s . A i r Force Academy Colorado S p r i n g s , Colorado 80912
APGC (PGBPS-12) Elg in AFB, F lor ida 32542
1
1
1
1
1
1
1
1
1
1
1
-.--.>4_.. n C I . " D _ I".."..".."-..0 "._ ___. Human Engineering L a b o r a t o r i e s Aberdeen Proving Ground, Haryland 21005
Direc tor U. S . Army Engineer Geodesy, I n t e l l i g e n c e
Research and Developoent Agency F o r t Belv ior , V i r g i n i a 22060
and napping
Commandant U. S. Army Command and General S t a f f Col lege Attn: S e c r e t a r y F o r t leavenworth, Kansas 66270
Dr . H. Rob1 Deputy Chief S c i e n t i s t U . 5. Amy Research O f f i c e (Uurham) Box M, Duke S t a t i o n Durham, North Caro l ina 27706
Commanding O f f i c e r U. S. Army Research Off ice (hrham) Attn: CRD-A4-IP (Richard 0. Ulsh) Box M, Duke S t a r i o n Durham, North Caro l ina 27706
L i b r a r i a n U. 5. Amy H i l i t a r y Academy Wesr P o i n t , New York 10996
The Walter Reed I n s t i t u t e o f Research Walter Reed Hediea l Center Washington, D. C . 20012
Conmanding O f f i c e r U. S. Army E l e c t r o n i c s RbD A c t i v i t y F o r t Huachuca, Arizona 85163
Cornanding O f f i c e r U. S. Army Engineer RbD Labora tory Atm: STINFO Branch F o r t Belvoi r , Vi rg in ia 22060
Commanding O f f i c e r U. S. Army E l e c t r o n i c s RbD A c t i v i t y 'dhite Sands H i s s i l e Range, New Hexico 88002
Dr. S . Benedict Levin, Direc tor I n s t i t u t e f o r Explora tory Research U. S . Amy E l e c t r o n i c s Cornand For t nonmouth, New J e r s e y 07703
Direc tor I n s t i t u t e f o r Explora tory Research U. S. Army E l e c t r o n i c s Cornand ALL": nr. Robert 0. Parker. Executive
F o r t Honmouth, New Jersey 07703
Conrnanding General U. S . Army E l e c t r o n i c s Command For t Hanmoufh, New J e r s e y 07703
ALtn: AMSEL-SC
S e c r e t a r y , JSTAC (AUSEL-XL-D)
RD-D RD-G RD-GF RL-MAT XL-D XL-E XL-C XL-S HL-D HL-CT-R HL-CT-P HL-CT-L HL-CT-0 ML-CT-I HL-CT-A NL-D NL-A NL-P N L - R NL-S KL-D KL-E KL-S KL-T VL-D WL-D
Chief of Naval Research Department of the Navy Washington. D. C . 20360 ALL.: Code 427
Chief o f Naval Research Department o f the Navy Washington, D. C . 20360 Arm: Code 437
Naval E l e c t r o n i c s Systems Command ELM 03 F a l l s Church, Vi rg in ia 22046
Naval Ship Systems Command SHIP 031 Washington, D. C . 20360
Naval Ship Systems Command SHIP 035 Washington, D . C . 20360
Naval Ordnance S y s t e m Command ORD 32 Washington, D. C . 20360
Naval A i r Systems Command A I R 03 Washington, 0. C . 20360
Commanding O f f i c e r O f f i c e o f Naval Research Branch Off ice Box 39, Navy No. 100 F.P.O. New York, New York 09510
1
1
1
1
1
I
1
1
1
2
1
2
5
1
1
1
1
1
1
1
1
1
1
1
1
1
AFETR Technica l Lib:ary (ED', HU-135) P a t r i c k AFB,Florids 32925
AFETR ( E R E - I ) STINFO O f f i c e r (For Library) P a t r i c k AFB, F l o r i d a 32925
Dr. L. n. H a l l i n g s v o r t h AFCRL (CRN) L. G . HBnscm F i e l d Bedford, Hsseachurer rs 01731
AFCRL ( C r W a R ) AFCRL Research L i b r a r y , Stop 29 L. G . Hanaeon F i e l d Bedford, Hassachuse t t s 01731
Colonel Robert E. Fonrana Department of E l e c t r i c a l Engineer ing A i r Force I n s t i t u t e of Technology Wright -Pa t te rson AFB, Ohio 45433
Colonel A. D. Blue RTD (RTTL) Bol l ing Air Force Base, 0. C . 20332
Dr. I. R. Hirman AFSC (SCT) Andreus AFB. Haryland 20331
Colonel J. D. Warthman AFSC (SCTR) Andrevs AFB, Haryland 20331
It. Col. J. 1. Reeves
Andrevs A F B , Maryland 20331
ESD (ESTI) 1. G. Manscorn F i e l d Bedford, Massachuse t t s 01731
AFSC (SCBB)
AEDC (ARO, INC) Attn: L i h r a r y l b c u m e n r s Arnold AFS, Tennessee 37389
European Off ice of Aerospace Research S h e l l Bui ld ing 47 Rue c a n t e r s t e e n Brusselo, Belgium
L r . Col. Robert B . Kal iach C h i e f , E l e c t r o n i c s Div is ion D i r e c t o r a t e of Engineer ing Sc iences Air Force O f f i c e of S c i e n t i f i c Research A r l i n g t o n , Vi rg in ia 22209
U. S. Army Research O f f i c e Attn: P h y s i c a l Sc iences Div is ion 3045 Columbia Pike Ar l ing ton , V i r g i n i a 22204
Research P lans O f f i c e U. S. Army Research O f f i c e 3045 Columbia P ike Ar l ing ton , Vi rg in ia 22204
Commanding Genera l U . S . Army M a t e r i e l Command Attn: UICRD-RS-DE-E Washington, D. C . 20315
Cornanding Genera l u. s. Army S t r a t e g i c Camrmnieafions Comnand Washington, D. C . 20315
Commanding O f f i c e r U. S . Amy H a t e r i a l s Research Agency Wa~ermwn Arsenal Watertown, HassachuSet t s 02172
Conmanding O f f i c e r u. S. Army B a l l i s t i c s Research Labara tory Attn: V . W . Richards Abcrdeen Proving Ground Aberdeen, Haryland 21005
Commandant U. S . Army A i r Defense School Attn: H i s s i l e Sc iences Div is ion , ChS Dept. P.O. Box 9390 F o r t Bliss, Texas 79916
Redstone S c i e n t i f i c Informat ion Center Attn: Chief , Ducurnent Sec t ion Rcdsrone Arsena l , Alabama 35809
Cornanding General Frankford Arsenal Attn: SMIJfA-1310 (Dr. Sidney Ross) P h i l a d e l p h i a , Pennsylvania 19137
u. s. Army Huni f ions Command ALtn: Technica l Informat ion Branch Pica t inney Arsena l Vver, New J e r s e y 07801
Cornanding O f f i c e r Harry Diamond Labora tor ies ALL^: D r . Berrhold Alma" (*HXDO-TI) COnnect icur Avenue and Van Ness S t r e e t , N.W Washington, D. C . 20438
Commanding O f f i c e r u. s . Army S e c u r i t y Agency Ar l ing ton Mall Arl ington , V i r g i n i a 22212
Connnanding O f f i c e r U. S . Army Limited War Labora tory ALL": Technica l D i r e c t o r Aberdeen Proving Ground Aberdeen, Hary lsnd 21005
I
B I b 1 8 1 1 I I 8 1 I I 1 I I I
i
1 Commanding O f f i c e r U t f l c e o t Naval Rerearch Branch O f f i c e 219 South Dearborn Street Chicago, I l l i n o i s 60604
Commanding O f f i c e r O f f i c e o f Naval Research Branch O f f i c e 1030 E a s t Green Street Pasadena, C a l i f o r n i a 91101
Commanding O f f i c e r Off ice o f Naval Research Branch O f f i c e 207 West 24th S t r e e t New York, New York 10011
Cornanding O f f i c e r O f f i c e o f Naval Research Branch Off ice 495 summer street Boston, Massachuse t t s 02210
Direc tor . Naval Research Labora tory Technica l Informat ion O f f i c e r Washington, D. C . 20390 A t m : Code 2000
Commander Naval A i r Development and MaCerLal Center J o h n s v i l l e . Pennry lvania 18974
1 Coddard Space F l i g h t Center "Inrziln,-l Aeroaeut ic r and Spare AdmlninfraCion ALLn: Library C3ITDL Green Be l t , Maryland 20771
1 The Johns Hopkins Univers i ty Applied Phys ics Labora tory
S i l v e r Spr ing , Maryland 20910 Attn: B o r i s W . Kuvshinoff
8621 Gecrg is Avenue
DOwment L i b r a r i a n
Hunt L i b r a r y Carnegie I n s t i t u t e o f Technology Schenley Park P i t t s b u r g h , Pennsylvania 15213
D r . Leo Young Stanford Research I n s t i t u t e Menlo Park. C a l i f o r n i a 94025
Mr. Henry L. Bachmann A s s i s t a n t Chief Engineer Wheeler Labora tor ies 122 C u t t e r m i l l Road Great Neck, New York 11021
School o f Engineering Sc iences Arizona S t a t e Univers i ty Tempe, Arizona 85281
Univers i ty o f C a l i f o r n i a a t Los Angeles Department o f Engineer ing t o s Angelea, C a l i f o r n i a 90024
C a l i f o r n i a I n s t i t u t e o f Technology Pasadena, C a l i f o r n i a 91109 Actn: Documents Library
u n i v e r s i t y Of C a l i f o r n i a S ~ n r a Barbara, C a l i f o r n i a 93106 Arm: Library
Carnegie I n s t i t u t e of Technology E l e c t r i c a l Engineering Department P i t t s b u r g h , Pennsylvania 15213
Univers i ty o f Michigan E l e c t r i c a l Engineering Deparmnent Ann Arbor, Michigan 48104
New York Univers i ty College o f Engineering New Yark, New York 10019
Syracuse u n i u e r s i t y Department o f E l e c t r i c a l Engineering Syracuse , New York 13210
Yale Univers l ry Engineer ing Deparhoent New Haven, Connecticut 06520
Airborne Ins t ruments Labora tory Deerpark, New York 11729
1 1 NASA Lewis Research Center Acm: Library 21000 Brookpark Road Cleveland, Ohia 44135
1 Nat iona l Science FoundatLon Attn: Dr. John R. Lehmann
1800 C S t r e e t , N.W. Washington, D. C. 20550
Divis ion o f Engineering
1 U. S . Atomic Energy ConniEsion Div is ion o f Technica l Information Extension P. 0. Box 62 Oak Ridge, Tennessee 37831
1 Lor Alsmos S c i e n t i f i c Labora tory Aftn: Reports L i b r a r y P. 0. Box 1663 Lo6 Alamos, New Mexico 87544
2 NASA S c i e n t i f i c & Technical Informat ion F a c i l i t y
Attn: Acquis i t ions Branch (SIAKIDL) P. 0 . BOX 33 College Pa rk , Maryland 20740
1 D i r e c t o r Research Laboratory o f E l e c t r o n i c s Massachusetts I n s t i t u t e o f Technology Cambridge, Massachuse t t s 02139
1
1
1
8
1
1
1
1
1
1
1
1
1
1
1
1
L i b r a r i a n U. S. Naval E l e c t r o n i c s Labora tory Sa" Diego. California 95152
Commanding o f f i c e r and Direc tor U. S. Naval Underwater Sound Labora tory For t Trumbull NEW London, Connect icu t 06840
L i b r a r i a n U. S . Navy PosL Graduate School Monrerey, C a l i f o r n i a 93940
Comnnder
Poin t Mug", C a l i f o r n i a 95468 u. s . ~ a v a i ~ i r nisslie Test center
D i r e c t o r U. S . Naval Observa tory Washington, D. C. 20390
1 Poly technic I n s f i t u f e o f Brooklyn 55 Johnson Street Brooklyn, New York 11201 Attn: Nr, Jerome Fox
Research Coordinator
1 Direc tor Columbia Radia t ion Labora tory Columbia U n i v e r r ~ t y 538 West 120th S t r e e t New York, New York 10027
1 Direc tor Coordinvced Science Labora tory Univers i ty o f I l l i n o i s Urbana, I l l i n o i s 61801
1 D i r e c t O r S tanford E l e c t r o n i c s Labora tor ies S tanford Unruer r i ty S tan fo rd , C a l i f o r n i a 94305
1 DirectOr E l e c t r o n i c s Research Laborarory Univers i ty o f C a l i f o r n i a Berkeley, C a l i f o r n i a 94720
1 Direc tor Eleccronic Sc iences Labora tory Univers i ty o f Southern C a l i f o r n i a Lor Angelel. C a l i f o r n i a 90007
1 Professor A . A. Dougal, Di rec tor Labora tor ies f o r E l e c t r o n i c s and Related
Sciences Research Univers i ty o f Texas Aus t in , Texas 78712
Division o f Engineer ing and Applied Physics 210 Pierce H a l l Howard Univers i ty Cambridge, Massachuse t t s 02138
1
1 Aerospace Corpora t ion P. 0 . Box 95085 Lo5 Angeles, C a l i f o r n i a 90045 Attn: Library A c q u i r i r ~ o n s Group
1 Professor Nicholas George Ca l i fo rn ia I n s t i t u t e of Technology Pasadena, C a l i f o r n i a 91109
1 Aeronautics L i b r a r y Crnduntc Aeronaut ica l Labora tor ies C a l i f o r n i a I i i s f i r u t e o f Technology 1201 East C s l i i o r n l a Boulevard Pasadena, C a l i f o r n i a 91109
1 Di rcc to r . USAF P r o j e c t RAND Via: Air Force Lia ison O f f i c e
1700 Main S t r e e t Sonta Monica, C a l i f o r n i a 90406 Attn: Library
The BAND c a r p o r a t i o n
Chief o f Naval Opera t ions OP-07 Washington, D. C. 20150
D i r e c t o r , U. S . Naval S e c u r i t y Group Affn: G43 3801 Nebraska Avenue Washington, D. C. 20016
Commanding O f f i c e r Naval Ordnance Labora tory White Oak, Maryland 21162
Camanding O f f i c e r Naval Ordnance Labora tory Corona, C a l i f o r n i a 91120
Commanding O f f i c e r Naval Ordnance Test S t a t i o n China Lake. C a l i f o r n i a 93555
Commanding O f f i c e r Navel Avionics F a c i l i t y I n d i a n a p o l i s , Ind iana 46218
Commanding O f f i c e r Naval Tra in ing Device Center Orlando, F l o r i d a 32813
U. 5. Naval Weapons Labora tory Dahlgren, V i r g i n i a 22448
Wesponr Systems Tes t Div is ion Naval A i r Test Center PatuxLent R ive r . Maryland 20670 Acm: Library
Head, Technica l Div is ion U. S. Naval Counter I n t e l l i g e n c e S u m o r t Center
2
Bendix P a c i f i c Div is ion 11600 Sherman Way North Hollyuood, C a l i f o r n i a 91605
General E l e c t r i c Company Research Labora tor ies Schenecrady, New York 12301
Lockheed A i r c r a f t Corpora t ion P. 0. Box 504 Sunnyvale, C a l i f o r n i a 94088
Raytheon Company Bedford, Massachuse t t s 01730 Attn: L i b r a r i a n
Or. G . J. Uurphy The Technologica l I n s t i t u t e Northwestern Univers i ty Evanrton, I l l i n o i s 60201
k. John C . Hancock, Di rec tor E l e c t r o n i c Systems Research Labora tory Purdue Univers i ty L a t a y e t t e , Ind iana 47907
Direc tor Microwave Labora tory S tanford Univers i ty S t a n f o r d , C a l i f o r n i a 94305
1
1 1
F a i m o n r Bui ld ing 4420 North F a i r f a x Drive A r l i n g t o n , Vi rg in ia 22203
MT. Char les F. Ynst S p e c i a l A r r t . t o t he Direc tor o f Research Nat iona l Aeronaut ics and Space Adminis t ra t ion Washzngton, D. C . 20546
D r . H. Har r i son , Code RRE C h i e f , E lec t rophys ics Branch Nat iona l Aeronaut ics and Space Adminxstration Washington, D. C . 20546
1 E m i l Schafer , Head E l e c t r o n i c s P r o p e r t i e s I n f o Center Hughes A i r c r a f t Company Culver C i t y , C a l i f o r n i a 90230
1
1
Secu n t y Classification
DOCUMENT CONTROL DATA - R 8, D 1 (Secur i ty c l a s s i f i c a t i o n of t i t l e , body of abs trar t and indexing annotation m u s t be entered when the overall report is r J a a s ; t i e d )
2a. R E P O R T 5 E C U R I T Y C L A 551 F I c A T I ON . O R I G I N A T I N G A C T I V I T Y (Corporate author) U n i v e r s i t y of I l l i n o i s UncJassif i ed
Urbana, I l l i n o i s 61801 i Coor d ina t ed S c ience Labor a t or y Zb. G R O U P
I 1 . R E P O R T T I T L E
A PROCEDURE FOR RAHKING DIAGNOSTIC TEST INPUTS
4 . D E S C R I P T I V E N O T E S (ripe ot report and. inc los ive d a t e s )
5 . A U T H O R ( S ) {First name, middle initial, l a s t name) 1 POWELL, The0
,. R E P O R T D A T E
May, 1967
DA 28 043 AMC 00073(E) 2 0 0 14 5 0 1B 3 1F
?.e. C O N T R A C T OR G R A N T N O .
b. P R O J E C T N O .
C .
d.
10. D l S T R l 8 U T l O N S T A T E M E N T
I
\ 7 6 . NO. O F R E F S 7a. T O T A L NO. O F P A G E S t
55 6 9a. O R I G I N A T O R ' S R E P O R T N U M B K R ( S )
t
R-354 I
96. O T H E R R E P O R T NO(S1 (Any other numbers that may be a s s i g n e d th is report)
4
D i s t r i b u t i o n o f t h i s r e p o r t is un l imi t ed I
1 1 . S U P P L E M E N T ~ R Y N O T E S 12. S P O N S O R I N G M I L I T A R Y A C T I V I T Y
Jo i n t Se rv ices E lec tr on i c s Program t h r u U. S. Army E l e c t r o n i c s Command F t Monmouth, N e w J e r s e y 07703
13. A 8 S T R A C T
S/N 01 01 -807-681 1 Securitv Classification