[DESCRIPTION] |
[Title of Invention]
PLANT DIAGNOSIS DEVICE, DIAGNOSIS METHOD, AND DIAGNOSIS PROGRAM [Technical Field] j
5 [0001] j
The present invention relates to a plant diagnosis device, j
a diagnosis method, and a diagnosis program. |
[Background Art]
[0002] j
10 When an abnormal transition event or trouble occurs in j
j
a plant, a plant diagnosis device of the plant detects the j
occurrence of the abnormal event or trouble based on measured j
data from the plant. j
[0003]
15 Patent Document 1 discloses a diagnosis device that uses
ART (Adaptive Resonance Theory). The ART-based diagnosis j
device has the function of categorizing multidimensional data
in accordance with their similarity.
[0004]
20 According to the technology of Patent Document 1, measured j
I
data in the normal state are first classified by ART into a !
plurality of categories (normal categories). The current j
i
measured data are then categorized by ART. If the current I
measured data cannot be classified into any of the normal 25 categories, a new category is created. The creation of a new
2 |
"i
C
category means that the state of the plant has changed. Thus I
the occurrence of an abnormal event is determined by the creation j
of a new category. When the rate of occurrence of new categories
exceeds a threshold, a trouble is diagnosed. i
5 [Related Art Document] I
[Patent Document] [0005] j
[PTL 1]
Japanese Unexamined Patent Publication No. 2005-165375 I
10 [Summary of Invention]
[Problems to be solved by the invention]
[0006]
According to Patent Document 1, an abnormal event is
diagnosed on the assumption that "a new category is created
15 when the current measured data cannot be classified into any
of the normal categories, i.e., when the state of the plant [
has changed." (•
[0007]
However, the state of the plant changes not only upon 20 occurrence of an abnormal event but also through aging f
degradation and varying operating conditions. In such cases,
new categories are also created. It follows that the creation f
of a new category does not immediately mean the occurrence of I
an abnormal event in the plant. I
25 [0008] I
I
3 {
t
S
l
i
i
i
I
Here, the operating conditions refer to external
environmental factors not directly associated with equipment
characteristics, such as the environment conditions j
(atmospheric temperature, humidity, etc.) of the location where
5 the plant is installed, and quantities determined by the j
operator"s operations (e.g., amount of electric power generated
by the plant) . Regardless of the occurrence or non-occurrence
of abnormalities, the values of diverse measured data vary as
the operating conditions change.
10 [0009]
Although aging degradation and varying operating
conditions are actually not abnormalities, they are still
diagnosed as abnormalities according to the method of Patent
Document 1. That is, a normal state could be diagnosed as an
15 abnormal event causing the generation of a false alarm.
[0010] [
An object of the present invention is to reduce the rate s
of occurrence of false alarms by distinguishing aging f
degradation and varying operating conditions from j
1
20 abnormalities, all being factors causing the state of the plant i
to change. f
[Means for Solving the Problems] [0011] [
According to the present invention, there is provided t
25 a plant diagnosis device for detecting an abnormal event of I
4 r
c
a plant being diagnosed based on measurement signals coming
therefrom and for reporting the detected abnormal event, the
plant diagnosis device including: a state change detection unit
which classifies the measurement signals from the plant being
5 diagnosed into categories for storage and which determines that ;
the state of the plant has changed if any measurement signal
cannot be classified into any of the categories; an abnormal :
event determination unit which determines that an abnormal event
has occurred in the plant being diagnosed using a signal from
10 the state change detection unit as part of input to the abnormal [
event determination unit; an alarm generation means for j
reporting the output of the abnormal event determination unit to the outside, and a false alarm inhibition unit which detects events other than the abnormalities of the plant being diagnosed
15 so as to inhibit the output of the abnormal event determination [
i
unit.
[0012]
Preferably, the false alarm inhibition unit may include
an aging degradation detection unit which obtains a weight using
(
5
20 the measurement signals fromthe plant beingdiagnosedandwhich, I
I
if the amount of change in the weight has exceeded a predetermined J
f
maximum value, determines that the range of aging degradation j
i
is exceeded, the aging degradation detection unit further I
I
inhibiting the output of the abnormal event determination unit |
i
25 if the range of aging degradation is not exceeded. I
i
5 I
j
I
j
c
[0013] |
Preferably, the false alarm inhibition unit may include j
an operating condition change detection unit which classifies j
operating condition data about the plant being diagnosed into
5 categories for storage and which, when no new category is created
from the classification of the operating condition data, j
determines that no operating condition change has occurred,
the operating condition change detection unit further
inhibiting the output of the abnormal event determination unit
10 when the operating conditions change.
[0014] l
Preferably, the operating condition data as input to the j
operating condition change detection unit may include external
environmental factors not directly associated with the
15 characteristics of equipment making up the plant being
diagnosed. j
[0015]
According to the present invention, there is also provided
a plant diagnosis device for detecting an abnormal event of [
20 a plant being diagnosed based on measurement signals coming I
therefrom and for reporting the detected abnormal event, the I
plant diagnosis device including: a state change detection unit {
which classifies the measurement signals from the plant being |
diagnosed into categories for storage and which determines that I
I
|
25 the state of the plant has changed if any measurement signal I 6 I
f
i
f
i
c
cannot be classified into any of the categories; an aging j
degradation detection unit which obtains a weight using the j
measurement signals from the plant being diagnosed and which, j
if the amount of change in the weight has exceeded a predetermined
5 maximum value, determines that the range of aging degradation i
is exceeded; an operating condition change detection unit which
classifies operating condition data about the plant being
diagnosed into categories for storage and which, when no new
category is created from the classification of the operating j
10 condition data, determines that no operating condition change j
has occurred; an abnormal event determination unit which ?
determines that an abnormal event has occurred in the plant j
being diagnosed if the state change detection unit determines I
that a state change is detected, if the aging degradation
15 detection unit determines that the range of aging degradation j
is exceeded, and if the operating condition change detection j
unit determines that the operating conditions have not changed, j
and an alarm generation means for reporting the output of the j
abnormal event determination unit to the outside. |
20 [0016] t
!
According to the present invention, there is also provided I
I
a plant diagnosis device including: a measurement signal I
database which stores measurement signals from a plant being |
f
diagnosed; a processing data extraction means for extracting 25 from the measurement signal database a diagnosis signal used I
7 |
f
|
c
to diagnose the state of the plant being diagnosed; a reference
signal database which stores the diagnose signals; a
classification means for classifying data stored in the
reference signal database into categories; a classification
i
5 result database which stores the categories as normal j
categories; a diagnosis means for diagnosing the state of the plant being diagnosed as belonging to one of the states of a
normal event, of an abnormal event, of operating condition change, (
and of aging degradation, using the latest diagnosis signal
10 extracted by the processing data extraction means as well as
information about the normal categories stored in the
classification result database; a diagnosis result database
which stores results of diagnosis by the diagnosis means, and |
an image display unit to which information stored in the t
15 diagnosis result database is output; wherein the diagnosis means >
includes an abnormal event determination unit, a state change
detection unit, an aging degradation detection unit, and an j
operating condition change detection unit; wherein the state i
change detection unit has the function of determining that a i
20 state change has occurred if the latest diagnosis signal I
extracted by the processing data extraction unit does not belong j
I
to any of the normal categories stored in the classification 5
I I
result database; wherein the aging degradation detection unit {
has the function of obtaining relations between the width of I
[
25 time change and the amount of change in a weight in the normal I
8 j
I
i
i
I
I
i
state, the function further determining that the range of aging j
degradation is exceeded if the amount of change in the weight I
I
in the normal state has exceeded a predetermined maximum value j
i
while diagnosis is being carried out; wherein the operating j
i
j
5 condition change detection unit has the function of determining j
j
that no operating condition change has occurred if no new j
category is created when operating condition data formed by
external environment factors not directly associated with
equipment characteristics are classified into categories, and
10 wherein the abnormal event determination unit determines that
an abnormal event has occurred if the state change detection •
unit determines that a state change has occurred, if the aging
i
degradation detection unit determines that the range of aging [
i
degradation is exceeded, and if the operating condition change f
15 detection unit determines that no operating condition change
has occurred. j
f
[0017] I
i
I
According to the present invention, there is also provided [
a plant diagnosis method for detecting an abnormal event of {
20 a plant being diagnosed based on measurement signals coming !
i
i
therefrom and for reporting the detected abnormal event, the }
S
plant diagnosis method including: classifying the measurement {
I
signals from the plant being diagnosed into categories for f
storage; determining that the state of the plant has changed {
f
25 if any measurement signal cannot be classified into any of the I
I
9 i
1
I
I
I
c
categories; obtaining upon such state change a weight using
the measurement signals from the plant being diagnosed;
determining that the range of aging degradation is exceeded ]
if the amount of change in the weight has exceeded a predetermined
5 maximum value; classifying operation condition data about the j
plant being diagnosed into categories for storage, and i
determining that an abnormal event has occurred in the plant being diagnosed if no new category is created from the ;
classification of the operating condition data, the determined
10 abnormal event being reported to the outside.
[0018] j
According to the present invention, there is also provided
a plant diagnosis program for detecting an abnormal event of I
a plant being diagnosed based on measurement signals coming
15 therefrom and for reporting the detected abnormal event, the
plant diagnosis program including: a state change detection
step which classifies the measurement signals from the plant ?
being diagnosed into categories for storage and which determines I
that the state of the plant has changed if any measurement signal j
20 cannot be classified into any of the categories; an abnormal j
event determination step which determines that an abnormal event [
I
has occurred in the plant being diagnosed using a signal from j
I
the state change detection step as part of input to the abnormal t
event determination step; an alarm generation steps which I
i
25 reports the output of the abnormal event determination step !
10 j
V
to the outside, and a false alarm inhibition step which detects
events other than the abnormalities of the plant being diagnosed j
so as to inhibit the output of the abnormal event determination
unit. j
5 [0019] i
Preferably, the false alarm inhibition step may include !
an aging degradation detection step which obtains a weight using j
the measurement signals from the plant being diagnosed and which, if the amount of change in the weight has exceeded a predetermined <
10 maximum value, determines that the range of aging degradation l
is exceeded, the aging degradation detection step further inhibiting the output of the abnormal event determination step if the range of aging degradation is not exceeded. J
[0020] I
15 Preferably, the false alarm inhibition step may include
an operating condition change detection step which classifies I
r
I
operating condition data about the plant being diagnosed into I
categories for storage and which, when no new category is created from the classification of the operating condition data, J
20 determines that no operating condition change has occurred, i
|
I
the operating condition change detection step further |
r
inhibiting the output of the abnormal event determination step f
i
when the operating conditions change. i
[0021] I I
i
25 Preferably, the operating condition data as input to the |
i
1 1 f
f
if
\| I
i
Ioperating condition change detection step may include external
environmental factors not directly associated with the
characteristics of equipment making up the plant being
diagnosed.
5 [0022]
According to the present invention, there is also provided I
i
a plant diagnosis program for detecting an abnormal event of i
a plant being diagnosed based on measurement signals coming 1
therefrom and for reporting the detected abnormal event, the [
10 plant diagnosis program including: a state change detection I
step which classifies the measurement signals from the plant I
being diagnosed into categories for storage and which determines that the state of the plant has changed if any measurement signal cannot be classified into any of the categories; an aging
15 degradation detection step which obtains a weight using the ;
measurement signals from the plant being diagnosed and which, if the amount of change in the weight has exceeded a predetermined [
maximum value, determines that the range of aging degradation I
is exceeded; an operating condition change detection step which
20 classifies operating condition data about the plant being i
diagnosed into categories for storage and which, if no new [
category is created from the classification of the operating j
l
condition data, determines that no operating condition change has occurred; an abnormal event determination step which j
25 determines that an abnormal event has occurred in the plant I
12 j
I
I
c
being diagnosed if the state change detection step determines ]
i
that a state change is detected, if the aging degradation j
j
detection step determines that the range of aging degradation j
i
is exceeded, and if the operating condition change detection j
i
5 step determines that the operating conditions have not changed, j
i
and an alarm generation step which reports the output of the j
i
<
abnormal event determination step to the outside. j
i
[Effects of Invention] j
[0023] j
I
10 Bydistinguishingagingdegradationandvaryingoperating i
conditions from abnormalities, it is possible to reduce the rate of occurrence of false alarms. j
[0024] Also, allowing the operator to know which of the categories I
15 of aging degradation, of changes in operating conditions, and f
of abnormalities the cause of the changes in data trends belongs j
to contributes to working out a plant maintenance schedule. $
i
[Brief Description of Drawings] f
[0025] |:
20 [Fig. 1] I
Fig. 1 is a block diagram showing a diagnosis device i
I
according to the present invention. |
I
I
[Fig. 2A] j
l"
i
I
Fig. 2A is a flowchart showing what is performed in normal {
I
25 state learning mode. i
|
j
f
c
[Fig. 2B] !
Fig. 2B is a flowchart showing what is performed in
diagnosis mode. j
i
[Fig. 3A] j
5 Fig. 3A is a schematic view showing how the two modes !
I
are carried out in each sampling period. |
[Fig. 3B] |
Fig. 3B is a schematic view showing a case of diagnosis I
in which normal state learning mode is carried out for a
10 predetermined time period while diagnosis mode is carried out in each sampling period.
[Fig. 3C]
Fig. 3C is a schematic view showing an example in which I
the modes are carried out with the timing set by the operator.
15 [Fig. 4A]
Fig. 4A is a block diagram showing a data preprocessing
unit and an ART module.
[Fig. 4B]
Fig. 4B is a block diagram showing an F0 layer of the
20 ART module. i
[Fig. 4C] I
Fig. 4C is a block diagram showing an Fl layer of the s
|
ART module. I
j
[Fig. 5A] j
I
25 Fig. 5A is a graph showing results of classification into j
I
14 I
i
I
categories by the working example indicated in Figs. 4A through
4C. I
[Fig. 5B] |
Fig. 5B is a graph showing an example of changes in j
5 measurement signals over time upon occurrence of an abnormal ;
i event.
[Fig. 6A] I
Fig. 6A is a schematic view of a screen on which data ]
items for use in diagnosis are set. 10 [Fig. 6B] Fig. 6B" is a block diagram explaining a classification I
means 400 indicated in Fig. 1.
[Fig. 7A]
Fig. 7A is a schematic view of a screen showing some data 15 to be stored in a measurement signal database. [Fig. 7B] I
Fig. 7B is a schematic view of a screen showing some data I
I
to be stored in a reference signal database. [
i;
[Fig. 7C] 1
i
I
|
20 Fig. 7C is a schematic view of a screen showing some data j
I
to be stored in a classification result database. |
f
[Fig. 8A] I
|
Fig. 8A is a schematic view showing a structure of an t
|
abnormal event determination unit 510. f
25 [Fig. 8B] j
15 j
c
Fig. 8B is a schematic view showing how some data from
a diagnosis result database are displayed on the screen.
[Fig. 8C]
Fig. 8C is a schematic view of a typical screen
5 chronologically displaying results of diagnosis from the
diagnosis result database. j
[Fig. 9A] j
Fig. 9A is an explanatory view explaining how an aging j
degradation detection unit operates. j
10 [Fig. 9B] I
Fig. 9B is an explanatory view explaining criteria for
determining the range of aging degradation. [Fig. 10A] I
Fig. 10A is an explanatory view explaining how a first {
15 working example of an operating condition change detection unit operates. f
[Fig. 10B]
Fig. 10B is an explanatory view explaining how a second
working example of the operating condition change detection
20 unit operates. j
[
[Fig. IOC] |
i
Fig. IOC is an explanatory view explaining how a third j
I
i
!
working example of the operating condition change detection I
unit operates. i
25 [Fig. 11] j
16 j
i
I
i
c
Fig. 11 is a block diagram showing a gas turbine power
plant.
[Fig. 12]
I
Fig. 12 is an explanatory view explaining typical results j
5 of the workings of a gas turbine power plant to which the diagnosis device of the present invention is applied. j
[Fig. 13] l
Fig. 13 is a tabular view summarizing the operating
conditions of the components making up diagnosis means.
10 [Description of Embodiments]
[0026]
The present invention is explained below using the
accompanying drawings. !
[Embodiment] I
15 [0027] ;
Fig. 1 is a block diagram showing a diagnosis device 200 f
according to the present invention. In Fig. 1, the diagnosis
device 200 diagnoses the state of a plant 100.
[0028] |
i
I
20 As its arithmetic unit, the diagnosis device 200 has a j
j
processing data extraction means 300, a classification means j
i
I
400, a diagnosis means 500, and an alarm generation means 600. j
i
And as its databases, the diagnosis device 200 has a measurement I
I
i
signal database 230, a reference signal database 240, a {
i
25 classification result database 250, and a diagnosis result I
!
17 |
j
I
database 2 60. In Fig. 1, the term "database" is abbreviated
to DB. These databases are components of the diagnosis device
200, and the information recorded in each of the databases is
electric information that is usually called electronic files
5 (electronic data).
[0029] i
The diagnosis device 200 also has an external input j
interface 210 and an external output interface 220 as the
i
interfaces with the outside. Via the external input interface j
10 210, a measurement signal 1 including the measured values of j
diverse state quantities of the plant 100 is input to the {
diagnosis device 200, along with an external input signal 2 !
generated by operation of an external input device 910 including j
a keyboard 920 and a mouse 930 . Via the external output interface 15 220, image display information 14 is output from the diagnosis
device 200 to an image display device 940 (image display unit) I
i
inside an operation control room 900. j
[0030] i
In this example, the processing data extraction means
20 300, classification means 400, diagnosis means 500, alarm
generation means 600, measurement signal database 230,
reference signal database 240, classification result database 250, and diagnosis result database 260 are all incorporated I
in the diagnosis device 200. Alternatively, part of these {
25 components may be located outside the diagnosis device 200 and I
18 [
i
i
I
f
I
t
i
c
only data may be communicated therebetween. Whereas this
example has one plant targeted for diagnosis, a single diagnosis
device 200 may alternatively diagnose a plurality of plants.
[0031]
5 Explained below is how the diagnosis device 200 operates .
First, a measurement signal 3 input via the external input
interface 210 is stored into the measurement signal database j
230. j
[0032] j
10 The processing data extraction means 300 extracts a j
diagnosis signal 6 for use in diagnosis from a measurement signal 5 stored in the measurement signal database 230, and stores (
the extracted signal 6 into the reference signal database 240. The reference signal database 240 stores the measurement signals I
I
15 in effect over the period in which the state was determined t
by the operator to be normal. Also, the processing data f
extraction means 300 extracts data items based on the settings I
made by the operator. The data item extraction will be discussed later in detail by referring to Figs. 6A and 6B. 1
20 [0033] j
The classification means 400 classifies a reference f
I
signal 7 into a category. Classification results 8 are stored [
I
I;
into the classification result database 250. The detailed j
processing by the classification means 400 will be discussed I
f
I
25 later by referring to Figs. 4A through 4C. i
19 j
|
Ji
I.
c
[0034]
The diagnosis means 500 diagnoses the state of the plant
by processing the latest diagnosis signal 6 as well as
classification results 9 stored in the classification result
5 database 250. The diagnosis means 500 is made up of an abnormal
event determination unit 510, a state change detection unit :
520, an aging degradation detection unit 530, and an operating
condition change detection unit 540.
[0035] I
10 The state change detection unit 520 determines whether the state of the plate 100 has changed. A state change ]
i
determination result 15 output from the state change detection j
unit 520 is "1" if the state is determined to have changed and is "0" if no state change has occurred. 15 [0036] The state change detection unit 52 0 compares the latest I
diagnosis signal 6 extracted by the processing data extraction
means 300 with the classification results 9 stored in the f
classification result database 250. If it is determined that
20 the latest diagnosis signal 6 belongs to a category included
in the classification results 9, the state change detection
unit 520 classifies the diagnosis signal 6 into that category. j
I
If the comparison between the latest diagnosis signal 6 extracted !
by the processing data extraction means 300 and the 25 classification results 9 reveals that the signal 9 does not I
20 S
i
c
belong to any of the categories included in the classification
results 9, a new category is created. The detailed processing
by the state change detection unit 520 will be discussed later
in detail by referring to Figs. 4A through 4C.
5 [0037] |
The aging degradation detection unit 530 determines
whether a change in the state of the plant falls within the ]
range of aging degradation. An aging degradation determination I
I
result 16 output from the aging degradation detection unit 530 j
10 is "1" if the change in the plant state is determined to have exceeded the range of aging degradation and is "0" if the change falls within the range of aging degradation. The detailed processing by the aging degradation detection unit 530 will j
t
be discussed later in detail by referring to Figs. 9A and 9B. i
15 [0038] j
The operating condition change detection unit 540 1
determines whether a change has occurred in the operating j
conditions of the plant. An operating condition change I
determination result 17 output from the operating condition 20 change detection unit 54 0 is "1" if no change is determined i
to have occurred in the operating conditions and is "0" if an j
operating condition change has occurred. The detailed processing by the operating condition change detection unit 540 will be discussed later in detail by referring to Figs. {
f
25 10A through IOC. |
I
21 }
I
I
!
i
i
c
[0039]
The abnormal event determination unit 510 determines
whether an abnormal event has occurred in the plant,"using the
state change determination result 15, aging degradation
5 determination result 16, and operating condition change
determination result 17. The detailed processing by the
abnormal event determination unit 510 will be discussed later
in detail by referring to Figs. 8A through 8C.
[0040]
10 A diagnosis result 10 output from the diagnosis means
500 includes the result of the determination by the abnormal event determination unit 510, state change determination result |
15, aging degradation determination result 16, and operating I
condition change determination result 17 . The diagnosis result j
15 10 is stored in the diagnosis result database 260.
[0041] The alarm generation means 600 determines whether or not j
to generate an alarm, using diagnosis results 11 stored in the diagnosis result database 2 60 and a measurement signal 4 with f
20 the latest timestamp stored in the measurement signal database [
230. j
[0042] I
i
!
The alarm generation means 600 has two criteria (conditions 1 and 2, given below) for determining alarm f
25 generation. Whether or not to generate an alarm is determined !
22 |
i
i
based on desired combinations of these criteria. The
combinations of the criteria triggering the alarm may include,
for example, a case where the two conditions 1 and 2 below are
both met, and a case where one of the two conditions 1 and 2
5 below is met.
[0043] |
i
Condition 1: The measurement signal 4 with the latest j
i
timestamp exceeds a predetermined range (threshold). I
[0044] I
10 Condition 2: The number of abnormalities determined by t
the abnormal event determination unit 510 over a predetermined time period exceeds a predetermined value (threshold). I
i
[0045] |
Incidentally, the thresholds of the conditions 1 and 2 {
15 are to be established by the operator.
[0046]
Upon determining that an alarm is to be generated, the I
I
alarm generation means 600 transmits an alarm signal 13 to the I
i
external output interface 220 . The alarm signal 13 is converted |
20 by the external output interface 220 into the image display I
I
information 14 that is displayed on the image display device
940.
[0047] 1
I
This working example uses the image display device 940 {
f
25 to report the alarm to the operator. Alternatively, an alarm !
23 I
{
I
i
I
sound may be first generated to let the operator know that an
alarm has been triggered. The image display device 940 may
be set to display the image display information 14 following
or simultaneously with the generation of the alarm sound.
5 [0048]
Also, diagnosis results 12 (results of determinations
by the abnormal event diagnosis unit 510, state change |
determination result 15, aging degradation determination j
result 16, and operating condition change determination result 10 17) stored in the diagnosis result database 260 may be displayed
on the image display device 94 0 via the external output interface
220. I
f
[0049] j
Furthermore, diagnosis device information 50 stored in {
15 the measurement signal database 230, reference signal database
240, classification result database 250, and diagnosis result I
database 2 60 can be displayed on the image display device 94 0.
These kinds of information may also be modified as needed using
the external input signal 2 generated by operation of the
20 external input device 910.
[0050]
The present invention is characterized in that the diagnosis means 500 includes the abnormal event diagnosis unit !
t
510, aging degradation detection unit 530, and operating j
25 condition change detection unit 540 and that these units serve l
i
24 |
f
c
to prevent the detection of aging degradation or of an operating
condition change from getting determined as an abnormal event.
[0051]
That is, although the state change detection unit 52 0 |
5 aggressively detects plant abnormalities, such abnormalities include cases of aging degradation or of operating condition j
changes which should not be considered abnormalities in the j
first place. These cases of aging degradation or of operating j
condition changes are detected as such by the abnormal event >
10 determination unit 510 in a comprehensive determination process so as to prevent false alarms. In that sense, the aging j
degradation detection unit 530 and operating condition change I
detection unit 540 may be said to constitute a false alarm I
{
inhibition unit as opposed to the state change detection unit I
15 520. j
[0052] |
£
The state change detection unit 52 0 generates a new f
f
category if the current measured data cannot be classified into any normal category, i.e., if the state of the plant has changed. I
20 [0053] j
Since the state of the plant changes not only upon occurrence [
I
i
of an abnormal event but also due to aging degradation or a f
I
change in the operating conditions, new categories are bound !
I
to be generated when the state change detection unit 520 is f
I
I
25 in operation. The operation conditions here refer to the 1
25 j
I
|
I
I
I
c
environment conditions (atmospheric temperature, humidity,
etc.) of the location where the plant is installed and quantities
such as the amount of electric power generated by the plant.
Regardless of the occurrence or non-occurrence of abnormalities,
5 changes in the operating conditions entail changes in the values
of various measured data.
[0054] j
Although aging degradation and operating condition j
changes are not abnormalities, they are diagnosed as such by
10 methods for determining abnormalities based solely on the !.
generation of new categories. This may cause the generation of false alarms. [
[0055] [
In this working example, the diagnosis means 500 includes
15 the abnormal event determination unit 510, aging degradation >
detection unit 530, and operating condition change detection
unit 540 so as to distinguish aging degradation and operating
l
condition changes from abnormalities (all these are causes of f
state changes) , whereby the rate of occurrence of false alarms
20 is reduced. j
f
[0056] j
f
If a diagnosis device not equipped with the abnormal event f
determination unit 510, aging degradation detection unit 530 I
and operating condition change detection unit 54 0 of this I
25 invention were used, distinguishing aging degradation and j
i
26 [
I
I
l
operating condition changes from abnormalities would require
the use of operating data collected over a long period of time.
That is, when normal state learning mode is carried out, the
data for evaluating all operating conditions and aging "
5 degradation are needed. This entails a long time period from
the start of plant operation until the start of diagnosis of ]
the plant. By contrast, using the diagnosis device of this !
invention makes it possible to distinguish aging degradation and operating condition changes from abnormalities starting 10 from a stage where numerous operations have yet to be accumulated i
for diagnosis. This can shorten the time required for I
i
introducing the diagnosis device. f
[0057] }
Figs. 2A and 2B are flowcharts showing the basic workings f
15 of the diagnosis device 2 00 indicated in Fig. 1. Fig. 2A shows I
normal state learning mode and Fig. 2B indicates diagnosis mode . I
[0058]
I
The explanation that follows will refer to the components {
I
shown in Fig. 1 as well. f
20 [0059] j
} •
I*
The diagnosis device 200 has two basic operation modes: I
1
normal state learning mode in which the data in the normal state f
I
are classified into categories based on the information stored I
I
in the reference signal database 240, and diagnosis mode in I
f
|
25 which the state of the plant 100 is diagnosed. |
I i
27 S
I
£
f.
I
I I
l
c
[0060]
In Fig. 2A, normal state learning mode is carried out
by executing steps S1000 and S1010, in that order.
[0061]
5 First in step S1000, the processing data extraction means
300 is operated to extract the diagnosis signal 6 from the
measurement signal 5 in the measurement signal database 230.
The diagnosis signal 6 is stored into the reference signal j
database 240. The data stored in the reference signal database
10 240 are data in effect over the time period in which the operating j
state of the plant 100 was determined to be normal by the operator. j
The data items stored in the reference signal database 240 will >
be discussed later by referring to Figs. 6A and 6B.
[0062] j.
15 Next in step S1010, the classification means 400 is
operated to classify the reference signals 7 stored in the reference signal database 240. The classification results 8 !
are stored into the classification result database 250. [0063] j
20 In normal state learning mode of Fig. 2A, as described above, the processing is performed using the processing data f
extraction means 300, measurement signal database 230, j
reference signal database 240, classification means 400, and I
i
classification result database 250 shown on the left side inside t
!
25 the diagnosis device 200 of Fig. 1. j
28 f
i
I
I
j
c
[0064]
Diagnosis mode indicated in Fig. 2B is carried out by
executing steps SHOO, S1110 and S1120, in that order.
[0065]
5 First in step SHOO, the measurement signal 1 coming from
the plant 100 is input to the diagnosis device 200 via the external
input interface 210. The measurement signal 3 is stored into j
the measurement signal database 230. j
[0066] j
10 Next, the processing data extractionmeans 300 is operated f
i
i
to extract the measurement signal 5 from the measurement signal f
database 230. The measurement signal 6 with the latest I
timestamp is transmitted to the diagnosis means 500. I
[0067] }
15 In step S1110, the diagnosis means 500 is operated. Of j
the arithmetic units making up the diagnosis means 500, the l
state change detection unit 520, aging degradation detection I
unit 530, operating condition change detection unit 540, and |
i l
f.
abnormal event determination unit 510 are operated, in that |
f
20 order. The diagnosis result 10 output from the diagnosis means f
I
500 is transmitted to and stored into the diagnosis result f
i
|
database 2 60. The diagnosis results 12 output from the j
f
i
diagnosis result database 260 are converted to the image display I
f
I
information 14 by the external output interface 220 before being I
25 output to the image display device 94 0. I
f
i
29 f
!
i
i
|
j
f
c
[0068]
In step S1120, the alarm generation means 600 is operated
to determine whether or not to generate an alarm. If an alarm
is to be generated, the alarm signal 13 output from the alarm
5 generation means 600 is converted by the external output
interface 220 into the image display information 14 before being
output to the image display device 940. In this manner, the j
alarm is reported to the operator of the plant 100. j
[0069] j
10 In diagnosis mode of Fig. 2B, as described above, the I
processing is performed using the units shown on the right side jj
inside the diagnosis means 500 (state change detection unit j
520, aging degradation detection unit 530, operating condition I
change detection unit 540, and abnormal event determination j
15 unit 510) , as well as the external output interface 220, image j
display device 940, and alarm generation means 600 of the
diagnosis device 200 in Fig. 1. Needless to say, as the f
precondition for the above processing, the external input I
interface 210, measurement signal database 230, and processing 20 data extraction means 300 are assumed to be already in use. f
[0070] I
?
Figs. 3A through 3C are explanatory views explaining the j:
i j
timings with which the flowchart of normal state learning mode I
i
(Fig. 2A) and that of diagnosis mode (Fig. 2B) are carried out I
t |
25 by the diagnosis device 200. |
30 j
c
[0071]
The diagnosis device 200 acquires the measurement signal
1 from the plant 100 in each sampling period.
[0072]
5 Fig. 3A shows how normal state learning mode and diagnosis
mode may both be carried out in each sampling period for
diagnosis. j
[0073] Also, Fig. 3B shows how normal state learning mode may I
10 be carried out in each predetermined period and how diagnosis mode alone may be performed in each sampling period for f
diagnosis. I
[0074] j
Furthermore, Fig. 3C shows how the operator may act to
15 establish a learning period and a diagnosis period and how normal
state learning mode and diagnosis mode may be carried out with
such timings.
[0075]
In each of the methods above, diagnosis mode is carried
20 out in each sampling period so that the state of the plant may
be diagnosed online. |
[0076] I
In the paragraphs that follow, the detailed processing I
of normal state learning mode and that of diagnosis mode will f
I
I $
25 be explained, in that order. As the precondition for the f
31 j
^5
explanation, the process quantities input in each mode are
assumed to be already classified into categories. Such
classification is carried out by the classification means 400
in normal state learning mode and by the state change detection
5 unit 520 in diagnosis mode. Thus prior to the explanation of j
!
the operations in each mode, the concept of category j
classification methods will be explained below in reference i
i 1
to the block diagrams of Figs. 4A through 4C as a common ground j
for understanding.
10 [0077] j
The ensuing paragraphs will discuss a case in which ART I
(Adaptive Resonance Theory) is applied to the classification I
means 400 and state change detection unit 520. Alternatively, j
other clustering techniques including vector quantification iX
15 may be applied instead. j
[0078] l
As shown in Fig. 4A, the classification means 400 and I
|
state change detection unit 52 0 are made up of a data j
tI preprocessing unit 710 and an ART module 720. Furthermore,
20 the ART module 720 includes an F0 layer 721, an Fl layer 722,
an F2 layer 723, a memory 724, and an orienting subsystem 725,
all interconnected with one another. The F0 layer 721 and Fl
layer 722 may be configured as shown in Figs. 4B and 4C, for
example.
25 [0079]
32
1
{
f t
i
c
In Fig. 4A, the data preprocessing unit 710 first converts
operating data to input data for the ART module 72 0.
Specifically, the mathematical expressions (1) and (2) given
below are carried out. The steps involved (process) are
5 explained below.
[0080] j
First, a maximum value and a minimum value are calculated
for each measurement item. Here, the method of normalization j
is explained using a process quantity xi of the plant as an f
10 example. J
[0081] j
It is assumed that there are as many as N data items of the quantity xi and that an n-th measurement value is xi (n) . i
If the maximum value and minimum value of the N data items are I
15 represented by Max_i and Min_i, respectively, then normalized
data Nxi(n) is expressed by the mathematical expression (1) I
i
below. j
[0082] |
[Math. 1] 20 Nxi (n)=a+(l-a)x(xi (n)-Min_i) / (Max_i-Min_i) ... (1) [0083] I
|
I
In the expression above, a stands for a constant (0<a<0.5). I
The mathematical expression (1) normalizes the data to a range [
f
of [a, 1-a]. {
25 [0084] !
I
-,., i
33 I
|
5
c
Next, the complement number of the normalized data is
calculated and added to the input data. The complement number
CNxi(n) of the normalized data Nxi(n) is calculated using the
following mathematical expression (2):
5 [0085]
[Math. 2]
CNxi(n)=l-Nxi(n) ... (2) i
[0086] {
The data preprocessing unit 710 performs the mathematical 10 expressions (1) and (2) above on a plurality of input data items j
to obtain data including the normalized data Nxi(n) and the
complement number CNxi(n) of the normalized data, and inputs I
the data thus obtained to the ART module 720 as the input data I
Ii(n). The above steps are included in the process performed I
15 by the data preprocessing unit 710 for converting the operating »
data to be input to the ART module 720.
[0087] |
The ART module 720 classifies the input data Ii(n) into I
a plurality of categories. For that purpose, the ART module
20 720 includes the F0 layer 721, Fl layer 722, F2 layer 723, memory
724, and orienting subsystem 725, all being interconnected. The Fl layer 722 and F2 layer 723 are coupled via weights. A I
weight represents the prototype of a category into which input j
data are classified. The prototype denotes the representative I
I.
25 value of the category in question. (
I
34 I
|
[0088]
The algorithm of the ART module 720 is explained next.
[0089]
The algorithm to be performed when input data are input
5 to the ART module 720 is outlined below in the form of processes
1 through 5.
[0090]
Process 1: The F0 layer 721 of which the detailedprocessing
is shown in Fig. 4B normalizes input vectors to remove noise. j
10 [0091] j
Process 2 : With a comparison made between the input data input to the Fl layer 722 and weights, a promising category j
I
candidate is oriented through the detailed processing shown j
in Fig. 4C. j
15 [0092] |
Process 3: The validity of the category oriented by the j
I
orienting subsystem 725 is evaluated with regard to a parameter j
f
I
p . If the oriented category is determined to be valid, the I
I
input data are classified into the category in question, and 20 the process advances to process 4. If the category is not |
determined to be valid, that category is reset and another j
i
promising category candidate is oriented from among the other categories (i.e., process 2 is repeated). Where the value of I
j
the parameter p is made larger, the classification of the f
25 categories is made finer; where the value of the parameter p f
35 [
j
(
c
is made smaller, the classification of the categories is made
coarser. The parameter p is called the vigilance parameter.
[0093]
Process 4: If all existing categories are reset in the
5 process 2 above, the input data are determined to belong to
a new category. In this case, a new weight is generated to
represent the prototype of the new category. j
[0094] I
Process 5: When the input data are classified into a j
10 category J, a weight WJ(new) corresponding to the category J !
is updated using a past weight WJ(old) and input data p (or j
data derived from the input data) in the following mathematical l
expression (3): |
I
[0095] |
15 [Math. 3] 1
WJ(new)=Kw • p+(l-Kw) -WJ(old) ... (3) {
[0096] |
In the above expression, Kw stands for a learning rate
parameter (0<Kw<l), i.e., a value that determines the rate at
20 which the input vector is reflected in the new weight. }
[0097] }•
f
The data classification algorithm of the ART module 720 j
i
is characterized by the process 4 above. [0098] j
i |%
25 In the process 4, if input data are input which have a j
36 j
|
i
f
pattern different from any of the patterns recorded (stored)
in the classificationresult database 250 in Fig. 1, anewpattern
may be recorded without modifying the recorded patterns. Thus
it is possible to record new patterns while the patterns learned
5 in the past are kept recorded.
[0099]
As described, when predetermined operating data are given
as input data, the ART module 72 0 learns a predetermined pattern
of the data. Thus when new input data are input to the ART module j
j
10 720 following completion of the learning, the algorithm above j
may be used to determine which of the past patterns is close to the new input data. If the input data have a pattern not j
t
i
experienced in the past, the input data are classified into j
i
a new category. [
15 [0100] I
The processes 1 through 5 described above are carried out specifically using the following steps: J
[0101] j
Fig. 4B is a block diagram showing the structure of the |
20 F0 layer 721. The F0 layer 721 includes processing function blocks 71, 72, 73 and 74. The processing function blocks 71, j
72, 73 and 74 perform the mathematical expressions (4), (5), (6) and (7) given below, respectively, to obtain a normalized input vector. I
25 [0102] f
37 j
I
I
I
I
I
The series of the processing by the FO layer 721 in Fig.
4B involves renormalizing the input data Ii each time they are
fed into the processing function block 71 so as to create a
normalized input vector Ui that is output from the processing
5 function block 74 to the Fl layer 721 and orienting subsystem
725. In the mathematical expressions (4), (5), (6) and (7)
given below, "i" represents the number of data items and "0"
denotes the Fl layer. [
[0103] |
10 First of all, from the input data Ii, WiO is calculated j
using the mathematical expression (4) below where "a" stands j
for a constant. [
[0104] i
[Math. 4] |
15 Wi0=Ii+aUi0 ... (4) f
[0105] Next, XiO, obtained by normalizing WiO in the mathematical expression (4), is calculated using the mathematical expression f
(5) below where W0 denotes a normalized WiO. 20 [0106] [Math. 5] f
Xi0=Wi0/W0 ... (5) [
i
[0107] (
|
ThenViO, obtained by ridding XiO of noise, is calculated f
i
25 using the mathematical expression (6) below, where 8 denotes I
38 I
i
a constant for noise removal. Because a sufficiently small
value is turned to 0 by calculation of the mathematical
expression (6) below, the input data are rid of noise. "
[0108] j
5 [Math. 6] I
i
I
Vi0=f(Xi0)=Xi0 (where, Xi0>9) j
=0 (where other than the above) ... (6) j
!
I
[0109] j
Finally, a normalized input vector UiO is obtained using !
f
10 the mathematical expression (7) below, where V0 is a normalized f
ViO. The eventually acquired UiO is input to the Fl layer. j
f
[0110] J
[Math. 7] j
Ui0=Vi0/V0 ... (7) J
15 [0111] j
Fig. 4C is a block diagram showing the structure of the [
i:
f
I
Fl layer 722. The Fl layer 722 holds UiO in a short-term memory [
following acquisition using the mathematical expression (7)
above, before eventually calculating Pi to be input to the F2
20 layer 723. |
i
I
[0112] I
t
I
The F2 layer includes processing function blocks 75, 76, I
j
77, 78, 79 and 80. The processing function blocks 75, 76, 77, |
i
78, 79 and 80 perform the mathematical expressions (8), (9), f I
25 (10), (11), (12) and (13) shown below, respectively. These [
39 I
!•
8-! f
I
l
c
mathematical expressions are listed as [Math. 8] through [Math.
13] below, where "a" and "b" denote a constant each, f ( ) denotes
the function indicated by the mathematical expression (6) , and
Tj represents the goodness of fit calculated in the F2 layer
5 722. It should be noted that in the mathematical expressions
(9), (11) and (12), the denominators are each normalized.
[0113] |
[Math. 8] |
I
Wi=UiO+aUi ... (8) I
I
10 [0114] j
[Math. 9] • {
Xi=Wi/W ... (9) f
[0115] {
[Math. 10] |
15 Vi=f(Xi)+bf(Qi) ... (10) I
[0116] j
[Math. 11] {
Ui=Vi/V ... (11) I
[0117] j i
20 [Math. 12] f
|
Qi=Pi/P ... (12) I
I.
[0118] |
f
[Math. 13] { 1 Pi=Ui+Sg(yi)Zij {
25 where, g(yi)=d(Tj=max(Tj)) {
40 j
F
=0 (where other than the above) ... (13)
[0119]
Figs. 5Aand5B show some results of classifying the process
quantities input to the plant into categories by use of the j
5 algorithm of the ART module 720 shown in Figs. 4A through 4C.
Fig. 5A is a graph showing typical results of the classification. j
As an example, this figure depicts two items of measured data j
in a two-dimensional graph. The vertical and the horizontal j
axes of the graph indicate standardized measured data of each j
10 item. |
[0120] j
i
I
The measured data are divided into a plurality of l
categories 750 (indicated as circles in Fig. 5A) by the ART |
I
I
module 720 shown in Fig. 4A. 15 [0121] i
Although Fig. 4A shows the measured data of two items I
in a two-dimensional graph, this is not limitative of the I
invention. Alternatively, the measured data of three or more f
I
items may be classified into categories using multidimensional f
20 coordinates. The results of the classification in Fig. 5A have been
I
obtainedby the classif icationmeans 400 innormal state learning f
mode indicated in Fig. 2A. These results are stored in the {
|
classification result database 250. Also, in diagnosis mode I
i
25 indicated in Fig. 2B, the results of the classification are I
41 f
f
f i
f
i
acquired by the state change detection unit 520.
[0122]
Fig. 5B shows typical time-dependent changes upon
occurrence of an abnormal event in the measurement signal 1 |
5 acquired from the plant 100. The horizontal axis of the figure
denotes time and the vertical axis indicates the measurement •
signal, category numbers, and the rate of occurrence of new j
categories (frequency of generation). Data Dl and D2 j
correspond to the items A and B, respectively. |
i
10 [0123] j
In Fig. 5B, the items A and B are shown to be approximately i
constant at first. The data Dl (item A) drops immediately before i
i
time tl, and the data D2 (item B) rises immediately after time j
tl. The data D2 (item B) drops thereafter, and then the data j
15 Dl (item A) and data D2 (item B) both rise in the end. |
[0124] j
Until time tl, the classified categories are numbered I
|
I
1 through 4 . These are reference-time categories (i.e., normal I I
categories) . By contrast, past time tl, the categories of the I
20 items A and B are numbered 5 through 7 . These are new categories i
indicating that the state of the plant has changed. j
{
[0125] j
With the state change, the rate of occurrence of new t
I
categories (frequency of generation) rises immediately after I
25 time tl. The rising rate of the occurrence is diagnosed as i
42 I
i
j
indicative of a state change exceeding a threshold. Here, the j
rate of occurrence of new categories is calculated using a moving
average of the number of new categories generated over a
predetermined time period. j
5 [0126] |
Such a state change is detected in diagnosis mode of Fig.
2B by the state change detection unit 520. The state change I
detection unit 520 may select one of two methods for diagnosis: j
a method for determining that the state has changed if the moving j
10 average of the number of new categories generated over a j
predetermined time period exceeds a threshold, or a method for j
j
determining that the state has changed if the generated category j
i
i
is a new category and for diagnosing each sample if any one
!
t
of the normal categories is selected on the assumption that j
I
15 no state change has occurred. I
i"
[0127] |
j
Fig. 6A shows the screen of the image display device 940
in Fig. 2 on which to set the data items to be extracted by
the processing data extraction means 300. The data items to
20 be used for diagnosis may be determined in accordance with the |
I
I
purpose of the diagnosis (i.e., details of the abnormal event |
i
I
desired to be detected). !
i
[0128] t
i
This screen 940 can be scrolled vertically and |
|
25 horizontally, and is provided with tabs (e.g., tabs of groups I
43 J
f
I
!
i
c
1 and 2 in the example of Fig. 6A) . The screen shows the process
quantities (data items) A, B, C, D, etc., of the plant together
with their process numbers PID. Also, the maximum value,
minimum value, etc., of each process quantity (data item) are
5 displayed. The operator may select a combination of the process
quantities to be managed in groups by taking into account the
correlations between the process quantities, the purpose of
the diagnosis (details of the abnormal event desired to be ,
detected) , etc. In Fig. 6A, the tab of group 1 shows that the
10 process quantities (data items) A, C and D have been selected (
to be monitored relative to one another in one group. [0129] |
Fig. 6B is a block diagram showing the structure of the j
i
classification means 400. With regard to the process j
i
15 quantities in each of the groups established in Fig. 6A, the !
I
classification means 400 is configured to include the data 1
I
preprocessing unit and ART module explained in reference to (
i
E
Fig. 4A. That is, the classification means 400 has the data processing unit and the ART module carry out their series of }
20 processing as many times as the number of the groups involved. j
[0130] |
Figs. 7A, 7B and 7C show how data are stored in the measurement signal database 230, reference signal database 240, i
i
and classification result database 250 in Fig. 1, respectively. |
25 These figures may each be considered a display screen of the !
i
44 !
I
I
I
I
I
c
image display device 940 in Fig. 1. As with the screen of Fig. I
i
6A and other screens, the screen 940 may be scrolled vertically !
and horizontally and are furnished with tabs as needed. j
[0131] |
5 As shown in Fig. 7A, the measurement signal database 230 j
stores the values of a plurality of data items (items A, B, j
C, etc.) measured in the plant 100 for each sampling period j
i
(times of day on the vertical axis) . The display screen 940 j
chronologically displays the values of these data items (A, 10 B, C, etc.) on the vertical axis. The use of scroll boxes 56A j
and 56B that can be moved vertically and horizontally provides j
i
scrolling display of extensive data. I
[0132] j
In the case of the reference signal database 240 in Fig. j
15 7B, selecting a tab 57a or 57b indicative of a data sheet of j
reference 1 or 2 provides collective display of only the items classified for each reference. I
[0133] j
The processing data extraction means 300 shown in Fig. I
j
20 1 extracts from the measurement signal database 230 a group j
I
of data to be used for diagnosing the plant 100. For example, f
Fig. 7B shows that there are two groups of reference signal {
data (for references 1 and 2) and that the group of data in
effect when "Reference 1" is selected is made up of items A,
25 C and D. The data corresponding to the data items of group
45 I
I
I
|
i
I
i
i
I
i
I
j
G
1 set in Fig. 6Abelong to reference 1, and the data corresponding j
to the data items of group 2 in the same figure belong to reference j
2.
[0134] j
5 Thus the measurement signal database 230 chronologically
stores the measured values of all data items in each data group
as shown in Fig. 7A, whereas the reference signal database 240 j
chronologically stores the measured values of the data items i
i
extracted by the processing data extraction means 300 in a j
10 plurality of data groups as indicated in Fig. 7B. j
[0135] ;
Also, Fig. 7C shows a display screen showing some data |
i
i.
j
to be stored in the classification result database 250 in Fig. j
1. On the left side of Fig. 7C, times of day and the data j
15 applicable thereto are displayed in relation to the numbers I
i
of the classified categories. On the right side of Fig. 7C, j
! i
the relations between category numbers andweights are displayed. |
I I
In this manner, the classification result database 250 stores j
i
the results of the classification of data in groups, the grouped j
1
20 data being held in the reference signal database 240. j
[0136] j
I
Fig. 8A is a block diagram showing the structure of the |
I
abnormal event determination unit 510. The abnormal event I
i
determination unit 510 determines whether an abnormal event {
i
25 has occurred in the plant, using the state change determination I
46 I
|
1
i
j
i
i
c !
result 15 from the state change detection unit 520, aging j
j
degradation determination result 16 from the aging degradation f
i
detection unit 530, and operating condition change j
determination result 17 from the operating condition change I
5 detection unit 540. {
i
[0137] j
As summarized in Fig. 13, the state change determination f
result 15 is "1" when a state change is determined to have occurred f f
and is "0" when no state change is detected. The aging r
i
10 degradation determination result 16 is "1" when a change in f
t
l
the state of the plant is determined to exceed the range of f I aging degradation and is "0" when the state change is determined i
I.
|
to fall in the range of aging degradation. The operating }
|
condition change determination result 17 is "1" when no change f
I
15 is detected in the operating conditions and is "0" when a change j
l
is detected in the operating conditions. |
I
[0138] j
Digital signals representing these results are input to
an AND gate 514 shown in Fig. 8A. In turn, the AND gate 514
20 gives an abnormal event determination result 10. That is, the
abnormal event determination result 10 is "1" indicating the
determination that an abnormal event has occurred if a change
•
in the state of the plant is detected, if the change in the
state of the plant is determined to exceed the range of aging j
i% |
25 degradation, and if no change is detected in the operating i
47 I
i
I
f I
I
c
conditions. The abnormal event determination result 10 is "0" j
under all other conditions. 1
[0139] i
[
Fig. 8B shows how the screen 94 0 may display some data |
5 to be stored in the diagnosis result database 260. The state I
change determination result, operating condition change I
|
i
determination result, aging degradation determination result, f
I
and abnormal event determination result are shown stored I
chronologically in the database and displayed on the screen. 1
10 Also, the diagnosis result database 260 stores the relations !
between times of day and the numbers of the categories into
which the data applicable to these times of day are classified,
along with the relations between the category numbers and weights
(as shown in Fig. 7C) , the relations being the results of the >
i
15 operation of the state change detection unit 52 0. [
I
[0140] j
I
Fig. 8C shows a typical screen display of some diagnosis }
I
results 12 shown plotted chronologically by the image display j
device 940, the results 12 being stored in the diagnosis result
20 database 260. It is possible not only to display numerically
the chronological results as shown in Fig. 8B, but also to plot |
them graphically as illustrated in Fig. 8C. It is also possible |
I
i"
not only to display "0" s" and "1" s" but also to indicate moving I
averages of the results over a predetermined time period. In {
I
25 Fig. 8C, a curve 551 denotes amoving average of the state change ! I
48 f
t I
f
f
j
j
I €
I
determination result, a curve 552 represents a moving average !
of the operating condition change determination result, a curve [
I
553 stands for a moving average of the aging degradation f
i I
determination result, and a curve 554 indicates amoving average j
{
5 of the abnormal event determination result. {
[0141] I
Fig. 9A is an explanatory view explaining how the aging
degradation detection unit 530 operates. As shown in the top
left graph of Fig. 9A, measured values of aging degradation
!•" 10 change gradually. That is, where the horizontal axis denotes j
I
time and the vertical axis represents process quantities, the i
i
items A and B of aging degradation have tendencies to change i
slowly. On the other hand, as shown in the top right graph, j
measured values may exceed the range of aging degradation from f
i
15 time to time. In many of these cases, an abnormal event or I
I
a change in the operating conditions has occurred. The graph {
i
!
in the top right also shows time on the horizontal axis and |
process quantities on the vertical axis. |
[0142]
I
20 It is also possible to recognize the above-described I
{
time-dependent changes by a determination based on the amount |
of change in measured values as well as on the amount of change j
A z in the weight of a given category. Since there exist j
f
correlations between these parameters, aging degradation can |
25 be determined using the amount A z . In the bottom right and f
4 9 I
|
|
|
f
f I
f
i
c I
bottom left graphs of Fig. 9A, the horizontal axis denotes the
item A and the vertical axis represents the item B, with the
amount of change A z in the weight of the category shown between
different times of day. In the bottom left graph plotting aging !
I
5 degradation, the amount of change A z in the weight of the f
i
category is small. Where measured values contain a change |
exceeding the range of aging degradation, the amount of change I
Az in the weight of the category appears pronounced. I
i"
[0143] 1
If
10 In the normal state, too, data tendencies vary with the j
aging degradation of the plant. The degrees of change in data }
tendencies in the normal state may then be recorded as a normal
I
range. As long as changes in data tendencies fall within that j
range, these changes are determined to be within the range of {
I
15 aging degradation. |
i
[0144] (
f
Fig. 9B is an explanatory view explaining criteria for I
I
i
determining whether or not the range of aging degradation is J
i
exceeded. In this figure, the horizontal axis denotes the width f
I
20 of time change and the vertical axis represents the amount of
weight change. The maximum values obtained in normal state
learning mode are plotted to form a line as criteria for I
I
determination. Whether the range of aging degradation is J
exceeded is determined by whether the relation between the width f
I
25 of time change and the amount of weight change (both actually |
I
50 [
I
I
c i
t
measured) has exceeded a predetermined boundary. j
[0145] S
i
With this working example, the determination is made based j
on the relations summarized between the width of time change i
5 and the amount of weight change in the normal state. Here, !
the amount of weight change with regard to the width of time
change is calculated using the following mathematical
expression (14) :
j
[0146] |
|
10 [Math. 14] i
I
Max(S(Wi(t+At)-Wi(t)2) ... (14) f
[0147] 1
i
In the mathematical expression (14) above, "i" is a symbol |
t
! I
that identifies the data item (l<i<n, where "n" is the total }
15 number of data items) , and Wi (t) is the weight of the data item j
i at time t. }
[0148] [
If the maximum value of the amounts of weight change in
the normal state is exceeded in diagnosis mode, then it is
f
20 determined that the range of aging degradation is exceeded. j
Whereas this working example determines the maximum value of
the amounts of weight change in the normal state to be a threshold, I
l
it is also possible to set the average of the amounts of weight I
change in the normal state to be a threshold. j
i
25 [0149] [
51 f
|
c
Fig. 10A is an explanatory view explaining how a first i
working example of the operating condition change detection j
i
unit 540 operates. This example performs the same ART j
!
processing in Figs. 4B and 4C carried out by the classification j
5 means 400 and state change detection unit 520 (i .e ., processing i
by the data preprocessing unit 710c and ART module 720d) . In i
i
i
t
this case, it should be noted that the input to the ART module I
is not the process quantities of the plant but the operating }
condition data thereof. That is, whether or not the operating j
10 conditions have changed is determined using the result of the j
ART-based classification of the data in groups made up of solely j
the operating condition data. Candidates of the operating |
condition data include "generated output," "temperature," and |
other external environmental factors not directly associated {
15 with equipment characteristics. I
[0150] I
If a new category is created upon classification of the I
!
operating condition data by ART, there is a high possibility |
I"
I
that a change in the operating conditions has occurred. J
|
20 Conversely, if no new category is created upon classification I
of the operating condition data by ART and if a new category
is created upon classification of the measured data, then the
I
change of the state is not caused by any change in the operating I
I
conditions. In this manner, it is determined whether the I
I
25 operating conditions have changed. {
52 I
I
I
I
[0151] !
Fig. 10B is an explanatory view explaining how a second
j
working example of the operating condition change detection
unit 540 operates.
5 [0152] |
The second working example also uses ART. As shown in I
Fig. 10B, this ART setup involves showing the items A and B
on the vertical and horizontal axes, respectively, and I
classifying reference-time data into several normal categories .
10 In this operation, if a new category is created, the similarity
between target data and each of the normal categories is
calculated. Following the calculations, the normal category
to which the target data is most similar (i.e., with the highest
similarity) is extracted.
15 [0153] I
The similarity S is calculated using the mathematical |
expression (15) given below, for example, so that the normal |
f
category with the smallest value S (i .e., with the lowest degree
of similarity) is extracted. [
20 [0154] j
[Math. 15] I
Sj=S(Di-Wij)2 ... (15) |
[0155] j
In the expression (15) above, "i" is a symbol that
25 identifies the data item (l<i<n, where "n" is the total number 53
I
i
i
c
!
of data items) , and " j " is a symbol that identifies the category j
i
(l<j<m, where "m" is the total number of normal categories) . j
i
Furthermore, Sj stands for the similarity, Di denotes the value j
i
of the data item i of the target data, and Wij represents the I
5 weight of the data item i. j
[0156]
Next, the degree of contribution Ci of each data item i
is calculated using, for example, the following mathematical |
expression (16): I
10 [0157] |
t
[Math. 16] i
l
s
Ci=|Di-Wij| ... (16) [0158] j
The higher its degree of contribution, the farther away j
15 a given data item is from any normal category. Such a data I
item can become the cause of the creation of a new category.
If this data item is the operating condition data, then a state |
change can be attributed to a change in the operating conditions . I
[0159] I
\ I
20 The example of Fig. 10B shows how much the coordinates f
of a new category are changed relative to the two-dimensional j
I
coordinate positions of normal categories on the vertical and ?
horizontal axes. In this figure, the change (degree of |
contribution) on the horizontal axis is seen larger than that jf
25 on the vertical axis, which reveals that the degree of 5 4 f
|
s
I
i
I
}.
I
contribution of the item A is the higher. j
[0160] j
Fig. IOC is an explanatory view explaining a third working i
i
example of the operating condition change detection unit 540. j
5 This example is a combination of the functions explained above t
in reference to Figs. 10A and 10B.
[0161] I
A result 542 determined by reference to Fig. 10A and a ("
contribution processing result 543 determined by reference to f
I
10 Fig. 10B are input to an operating condition change determination j
i
unit 544. In turn, the operating condition change {
determination unit 544 performs the ANDor OR operation digitally >
on the input before outputting an operating condition change determination result 545. I
j
15 [0162] |
What follows is an explanation of how the diagnosis device j
200 of this invention works when applied to a thermal power I
plant. Fig. 11 is a block diagram showing a typical thermal >
|
power plant. j
I
20 [0163] |
In Fig. 11, the thermal power plant 100 includes a gas l
turbine generator 110, a control device 120, and a data I
transmission device 130. The gas turbine generator 110 J
includes a generator 111, a compressor 112, a combustor 113, I
25 and a turbine 114. 55 j
|
I
I
c
[0164] i
Upon power generation, the air taken into the compressor j
112 is compressed thereby and forwarded to the combustor 113. !
The combustor 113 mixes the compressed air with a fuel and burns !
5 the mixture. A high-pressure gas produced by the burning
mixture is used to rotate the turbine 114 that in turn causes
the generator 111 to generate electric power.
[0165]
The control device 12 0 controls the output of the gas
10 turbine generator 110 in keeping with the demand for electric
power. As its input data, the control device 120 admits f
operation data 102 measured by sensors (not shown) attached I
to the gas turbine generator 110. The operation data 102, measured in each sampling period, are composed of state i
15 quantities such as intake air temperature, fuel input, turbine f
exhaust gas temperature, turbine rotation frequency, j
generator-generatedelectricity, and turbine shaft vibrations . J
I
t
Furthermore, meteorological information such as atmospheric I
Ii
temperature is measured as well. f
20 [0166] I
I
}.
The control device 120 uses the operation data 102 to I
•i
calculate a control signal 101 for controlling the gas turbine {
generator 110. I
[0167] j
25 The signal data transmission device 130 transmits to the
56 j
|
I
j
f- t
I
c
diagnosis device 200 the measurement signal 1 that includes j
the operation data 102 measured by the control device 120 as I
well as the control signal 101 calculated by the control device ?
120. !
5 [0168] l
Fig. 12 is an explanatory view explaining typical results of diagnosis of the thermal power plant discussed above in
reference to Fig. 11, the diagnosis having been made by the diagnosis device 200 of the present invention. |
10 [0169] (
i
Fig. 12 shows the results in effect when a generator output }
"a" and an atmospheric temperature "b" have been selected as j
the data items of the operating condition data for group 1, and the generator output "a," atmospheric temperature "b, " and 15 a fuel flow rate "c" have been selected as the data items for group 2. [0170] j
The data from between time T0 and time Ti are used as normal j
I
data for learning the normal state in learning mode. Even in [
20 the normal state, efficiency drops due to aging degradation I
and the fuel flow rate "c" needed to acquire the same output I
rises. i
[0171] j
From time Ti, diagnosis is started using the diagnosis
25 device 200. Between time Ti and time T2, the normal state j
57 j
j?
i
j
i
€ I
continues but the fuel flow rate "c" rises under the influence 1
I
of aging degradation. As a result of the change in the fuel j
flow rate "c," operating the state change detection unit 520 f
produces a new category. This causes a state change f
j».
5 determination result "d" to approach 1 past time Ti. Since the I
i
rate of increase in the fuel flow rate "c" is within the range l
of learning mode, the aging degradation detection unit 530 f
i
t
determines that the change falls within the range of aging !
degradation. With these factors taken into account, the |
10 abnormal determination unit 510 determines that no abnormal I
event has occurred. S
i.
[0172] |
i
Between time T2 and time T3, the atmospheric temperature j
I
"b" rises temporarily. Since the rise in the atmospheric f
I I
15 temperature "b" is known to lower efficiency, the fuel flow j
I
rate "c" required to obtain the same output "a" increases. As |
a result, operating the state change detection unit 520 produces f
I
a new category. Also, the aging degradation detection unit I
I I
530 determines that the rate of increase in the fuel flow rate I
I
20 "c" has exceeded the range of learning mode. Because the f
atmospheric temperature "b" has changed, the operating !
condition change detection unit 54 0 determines that operating I
k
conditions "f" have changed. With these factors taken into f
consideration, the abnormal event determination unit 510 !
25 determines that no abnormal event has occurred. |
58 I
i I
i
1
s
j c
[0173]
Following a drop in the atmospheric temperature, an
abnormal event occurs at time T4 entailing an increase in the
fuel flow rate "c." Because the rate of increase in the fuel
5 flow rate "c" exceeds the range of aging degradation "e" and
because the operating conditions "f" remain unchanged, the I
abnormal event determination unit 510 determines that an I
abnormal event has occurred. |
[0174] j
10 If solely the result from the state change detection unit i
520 were used for determining abnormal events, an abnormal event i
would be diagnosed as having occurred past time Ti. This is a false alarm. Through the use of the aging degradation I
S-
|I "
detection unit 530 and operating condition change detection f
15 unit 540, it is possible to determine that the normal state |
continues between time Ti and time T4 thereby reducing the rate }
of false alarms. [0175] j
Using the diagnosis device 200 of this invention as j
20 described above provides the effects of lowering the rate of f
false alarms. Also, because the operator is offered f
information as to whether a change in data trends is attributable >
to aging degradation, to a change in the operating conditions or to an abnormal event, the offered information helps work 25 out a plant maintenance schedule. 59 {
t
i
1
[0176] |
Whereas the diagnosis means 500 of this invention includes j
the abnormal event determination unit 510, state change i
detection unit 520, aging degradation detection unit 530, and "
i
5 operating condition change detection unit 540, it is also I
possible to have either the aging degradation detection unit I
530 or the operating condition change detection unit 540 alone
in the diagnosis means 500 so as to detect either changes of
aging degradation or those in the operating conditions.
10 [0177] I
F
Furthermore, the matters illustrated in the accompanying |
drawings and discussed in this description may be implemented |
i
to constitute a diagnosis device, to practice a diagnosis method f
I"
i
involving the use of a computer, or to devise a program made f
15 up of a series of steps. I
[Industrial Applicability] {
[0178] |
The present invention can be applied extensively to I
diverse kinds of plants and other facilities in the form of I
I
20 a diagnosis device, a diagnosis method, or a diagnosis program. }
[Explanation of Reference
j-
] [
[0179] |
100 Plant I
I
25 200 Diagnosis device 60 j
l
| 1
I
t
I
!
j
j
€
210 External input interface
220 External output interface
230 Measurement signal database
240 Reference signal database
5 250 Classification result database r
260 Diagnosis result database
300 Processing data extraction means j
400 Classification means f
500 Diagnosis means 10 510 Abnormal event determination unit 520 State change detection unit I
530 Aging degradation detection unit |
540 Operating condition change detection unit j
i
600 Alarm generation means I,
15 900 Operation control room |
910 External input device 920 Keyboard ?
930 Mouse 940 Image display device f
20 I

[CLAIMS]
[Claim 1]
A plant diagnosis device for detecting an abnormal event
of a plant being diagnosed based on measurement signals coming
5 therefrom and for reporting the detected abnormal event, the
plant diagnosis device comprising: a state change detection
unit which classifies the measurement signals from the plant
being diagnosed into categories for storage and which determines
that the state of the plant has changed if any measurement signal
10 cannot be classified into any of the categories; an abnormal
event determination unit which determines that an abnormal event
has occurred in the plant being diagnosed using a signal from
the state change detection unit as part of input to the abnormal
event determination unit; an alarm generation means for
15 reporting the output of the abnormal event determination unit
to the outside, and a false alarm inhibition unit which detects
events other than the abnormalities of the plant being diagnosed
so as to inhibit the output of the abnormal event determination
unit.
20 [Claim 2] (Amended)
The plant diagnosis device according to claim 1, wherein
the false alarm inhibition unit includes an aging degradation
detection unit which obtains a weight corresponding to each
of the categories using the measurement signals from the plant
25 being diagnosed and which, if the amount of change in the weight
c
corresponding to any of the categories has exceeded a
predetermined maximum value, determines that the range of aging
degradation is exceeded, the aging degradation detection unit
further inhibiting the output of the abnormal event
5 determination unit if the range of aging degradation is not
exceeded.
[Claim 3]
The plant diagnosis device according to claim 1, wherein
the false alarm inhibition unit includes an operating condition
10 change detection unit which classifies operating condition data
about the plant being diagnosed into categories for storage
and which, when no new category is created from the
classification of the operating condition data, determines that
no operating condition change has occurred, the operating
15 condition change detection unit further inhibiting the output
of the abnormal event determination unit when the operating
conditions change.
[Claim 4]
The plant diagnosis device according to claim 3, wherein
20 the operating condition data as input to the operating condition
change detection unit include external environmental factors
not directly associated with the characteristics of equipment
making up the plant being diagnosed.
[Claim 5] (Amended)
25 A plant diagnosis device for detecting an abnormal event
of a plant being diagnosed based on measurement signals coming
therefrom and for reporting the detected abnormal event, the
plant diagnosis device comprising: a state change detection
unit which classifies the measurement signals from the plant
5 being diagnosed into categories for storage and which determines
that the state of the plant has changed if any measurement signal
cannot be classified into any of the categories; an aging
degradation detection unit which obtains a weight corresponding
to each of the categories using the measurement signals from
10 the plant being diagnosed and which, if the amount of change
in the weight corresponding to any of the categories has exceeded
a predetermined maximum value, determines that the range of
aging degradation is exceeded; an operating condition change
detection unit which classifies operating condition data about
15 the plant being diagnosed into categories for storage and which,
when no new category is created from the classification of the
operating condition data, determines that no operating
condition change has occurred; an abnormal event determination
unit which determines that an abnormal event has occurred in
20 the plant being diagnosed if the state change detection unit
determines that a state change is detected, if the aging
degradation detection unit determines that the range of aging
degradation is exceeded, and if the operating condition change
detection unit determines that the operating conditions have
25 not changed, and an alarm generation means for reporting the
CM
output of the abnormal event determination unit to the outside.
[Claim 6] (Amended)
Aplant diagnosis device comprising: a measurement signal
database which stores measurement signals from a plant being
5 diagnosed; a processing data extraction means for extracting
from the measurement signal database a diagnosis signal used
to diagnose the state of the plant being diagnosed; a reference
signal database which stores the diagnose signals; a
classification means for classifying data stored in the
10 reference signal database into categories; a classification
result database which stores the categories as normal
categories; a diagnosis means for diagnosing the state of the
plant being diagnosed as belonging to one of the states of a
normal event, of an abnormal event, of operating condition change,
15 and of aging degradation, using the latest diagnosis signal
extracted by the processing data extraction means as well as
information about the normal categories stored in the
classification result database; a diagnosis result database
which stores results of diagnosis by the diagnosis means, and
20 an image display unit to which information stored in the
diagnosis result database is output;
wherein the diagnosis means includes an abnormal event
determination unit, a state change detection unit, an aging
degradation detection unit, and an operating condition change
25 detection unit; wherein the state change detection unit has
65
the function of determining that a state change has occurred
if the latest diagnosis signal extracted by the processing data
extraction unit does not belong to any of the normal categories
stored in the classification result database; wherein the aging
5 degradation detection unit has the function of obtaining
relations between the width of time change and the amount of
change in a weight corresponding to each of the categories in
the normal state, the function further determining that the
range of aging degradation is exceeded if the amount of change
10 in the weight corresponding to any of the categories in the
normal state has exceeded a predetermined maximum value while
diagnosis is being carried out; wherein the operating condition
change detection unit has the function of determining that no
operating condition change has occurred if no new category is
15 created when operating condition data formed by external
environment factors not directly associated with equipment
characteristics are classified into categories, and wherein
the abnormal event determination unit determines that an
abnormal event has occurred if the state change detection unit
20 determines that a state change has occurred, if the aging
degradation detection unit determines that the range of aging
degradation is exceeded, and if the operating condition change
detection unit determines that no operating condition change
has occurred.
25 [Claim 7] (Amended)
c
A plant diagnosis method for detecting an abnormal event
of a plant being diagnosed based on measurement signals coming
therefrom and for reporting the detected abnormal event, the
plant diagnosis method comprising: classifying the measurement
5 signals from the plant being diagnosed into categories for
storage; determining that the state of the plant has changed
if any measurement signal cannot be classified into any of the
categories; obtaining upon such state change a weight
corresponding to each of the categories using the measurement
10 signals from the plant being diagnosed; determining that the
range of aging degradation is exceeded if the amount of change
in the weight corresponding to any of the categories has exceeded
a predetermined maximum value; classifying operation condition
data about the plant being diagnosed into categories for storage,
15 and determining that an abnormal event has occurred in the plant
being diagnosed if no new category is created from the
classification of the operating condition data, the determined
abnormal event being reported to the outside.
[Claim 8]
20 A plant diagnosis program for detecting an abnormal event
of a plant being diagnosed based on measurement signals coming
therefrom and for reporting the detected abnormal event, the
plant diagnosis program comprising: a state change detection
step which classifies the measurement signals from the plant
25 being diagnosed into categories for storage and which determines
6T
c
that the state of the plant has changed if any measurement signal
cannot be classified into any of the categories; an abnormal
event determination step which determines that an abnormal event
has occurred in the plant being diagnosed using a signal from
5 the state change detection step as part of input to the abnormal
event determination step; an alarm generation steps which
reports the output of the abnormal event detejntiination step
to the outside, and a false alarm inhibition step which detects
events other than the abnormalities of the plant being diagnosed
10 so as to inhibit the output of the abnormal event determination
unit.
[Claim 9] (Amended)
The plant diagnosis program according to claim 8, wherein
the false alarm inhibition step includes an aging degradation
15 detection step which obtains a weight corresponding to each
of the categories using the measurement signals from the plant
being diagnosed and which, if the amount of change in the weight
corresponding to any of the categories has exceeded a
predetermined maximum value, determines that the range of aging
20 degradation is exceeded, the aging degradation detection step
further inhibiting the output of the abnormal event
determination step if the range of aging degradation is not
exceeded.
[Claim 10]
25 The plant diagnosis program according to claim 8, wherein
€
the false alarm inhibition step includes an operating condition
change detection step which classifies operating condition data
about the plant being diagnosed into categories for storage
and which, when no new category is created from the
5 classification of the operating condition data, determines that
no operating condition change has occurred, the operating
condition change detection step further inhibiting the output
of the abnormal event determination step when the operating
conditions change.
10 [Claim 11]
The plant diagnosis program according to claim 10, wherein
the operating condition data as input to the operating condition
change detection step include external environmental factors
not directly associated with the characteristics of equipment
15 making up the plant being diagnosed.
[Claim 12] (Amended)
A plant diagnosis program for detecting an abnormal event
of a plant being diagnosed based on measurement signals coming
therefrom and for reporting the detected abnormal event, the
20 plant diagnosis program comprising: a state change detection
step which classifies the measurement signals from the plant
being diagnosed into categories for storage and which determines
that the state of the plant has changed if any measurement signal
cannot be classified into any of the categories; an aging
25 degradation detection step which obtains a weight corresponding
69
to each of the categories using the measurement signals from
the plant being diagnosed and which, if the amount of change
in the weight corresponding to any of the categories has exceeded
a predetermined maximum value, determines that the range of
5 aging degradation is exceeded; an operating condition change
detection step which classifies operating condition data about
the plant being diagnosed into categories for storage and which,
if no new category is created from the classification of the
operating condition data, determines that no operating
10 condition change has occurred; an abnormal event determination
step which determines that an abnormal event has occurred in
the plant being diagnosed if the state change detection step
determines that a state change is detected, if the aging
degradation detection step determines that the range of aging
15 degradation is exceeded, and if the operating condition change
detection step determines that the operating conditions have
not changed, and an alarm generation step which reports the
output of the abnormal event determination step to the outside.