JPH09189569A - Method and apparatus for support of regular operation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enhance the accuracy of the detection of an abnormality in an apparatus constituting a plant and to specify even the kind of the abnormality in the regular operating test of the plant. SOLUTION: An operating signal which operates an apparatus constituting a plant 1 is taken into from an operating-signal input part 2, a plant signal related to an apparatus, to be tested, which is specified by the operating signal is taken into by a plant-data intake part 3, the structure of a neural-network computing part 7 is decided according to the apparatus to be tested, a plurality of kinds of feature amounts are found by a computing part 5 on the basis of a change in the time of the plant signal related to the apparatus to be tested, and the respective feature amounts are input to the neural-netowrk commuting part 7. Then, on the basis of the output of the neural-network computing part 7, whether the apparatus to be tested is normal or abnormal is judged and/or the kind of an abnormality is specified by an abnormality judgement part 10. Thereby, since the number of respective constituent units and the weighting factor of the input layer, the intermediate layer and the output layer of a neural network are decided at every apparatus to be tested, the accuracy of a judgment is enhanced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a regular operation support method and apparatus for discriminating normality / abnormality of equipment constituting a plant in a regular operation test of a plant, and in particular, it is highly accurate in discriminating the presence / absence of abnormality, The present invention relates to a regular operation support method and apparatus suitable for specifying.

[0002]

2. Description of the Related Art In various plants such as a thermal power plant and a nuclear power plant, a regular operation test is regularly performed to confirm that the equipments constituting the plant operate normally. In the regular operation test, the operator operates the equipment to be tested, and by observing the operation result,
Determine normal / abnormal. For example, in a test on a pump, which is one of the equipments that make up a plant, the pump rotation speed should be gradually increased to the rated rotation speed, and the inspector should monitor changes in the discharge pressure and flow rate of the pump at this time. Has confirmed that this pump operates normally.
However, in this method, since the normality / abnormality is judged based on the subjective evaluation criteria of the inspector, there is a possibility that the judgment result may be different between an experienced inspector and an inexperienced inspector, and an abnormality of the device may be overlooked. There is a problem that there is.

Therefore, in order to solve such a problem, a method of automatically determining abnormality using a computer can be considered. For example, in the abnormality detection method described in Japanese Patent Laid-Open No. 2-96613, an average value of process values is obtained as a feature amount, and abnormality is detected based on the average value and the change in the process value.

[0004]

DISCLOSURE OF THE INVENTION Problems to be Solved by the Invention
In the method using a single feature quantity, which is an average value, as in the conventional technique described in Japanese Patent No. 6613, when the process value is changed greatly like the regular operation test, and the change tendency differs depending on the type of abnormality. Has a problem that it is difficult to detect abnormalities with high accuracy and determine the type of abnormality.

An object of the present invention is to provide a regular operation support method and apparatus which have a high abnormality detection sensitivity in a regular operation test of a plant and can even specify the type of abnormality.

[0006]

The above object is to take in an operation signal for operating equipment constituting a plant, take in a plant signal relating to a device under test specified by the operation signal, and to respond to the device under test. Determine the structure of the neural network, obtain a plurality of types of feature quantities from the time change of the plant signal related to the test target device, input each feature quantity to the neural network, and output the test target from the output of the neural network. This is achieved by determining the normality / abnormality of the device and / or identifying the type of abnormality.

According to the present invention, the structure of the neural network for outputting the normal / abnormality determination data and the abnormality type specifying data for each of a large number of plant constituent devices has an optimal structure for each test target device. Therefore, it is possible to improve the determination accuracy.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram of a regular operation support device according to the first embodiment of the present invention. In FIG. 1, 1 is a plant which is the subject of a regular operation test. 2
Is an operation signal input unit that takes in an operation input signal input by an operator operating a control panel (not shown) into the regular operation support device. Reference numeral 3 is a plant data acquisition unit for acquiring data from the plant 1. A plant data storage unit 4 stores the data captured by the plant data capture unit 3. Reference numeral 5 is a feature amount calculation unit for calculating a feature amount from the captured data. Reference numeral 6 is a feature quantity storage unit for storing the feature quantity calculated by the feature quantity calculation unit 5. Reference numeral 7 denotes a neural network computing unit for digitizing the state of the plant component equipment from the plurality of feature amounts calculated by the feature amount computing unit 5. Reference numeral 8 denotes a learning result storage unit that stores parameters necessary for determining the structure of the neural network of the neural network operation unit 7. An input device 9 is used by an operator to input data necessary for the operation of the regular operation support device. Reference numeral 10 is an abnormality determination unit for determining normality / abnormality and specifying the type of abnormality based on the output of the neural network operation unit 7. 1
Reference numeral 1 denotes an abnormality determination data storage unit that stores abnormality determination data that the abnormality determination unit 10 refers to when executing. 12
Is a display control unit that controls the display of the type of abnormality determined by the abnormality determination unit 10. Reference numeral 13 is a display device that displays the type of abnormality determined by the abnormality determination unit 10.

There are two processing modes of the regular operation support device shown in FIG. 1, a "learning mode" and a "recognition mode".
The "recognition mode" is a normal / abnormal state of the equipment based on the feature amount obtained from the operation input signal for operating each of the various equipments that make up the plant and the plant signal related to the operation of the equipment to be operated. This is a mode for performing the determination of and the process of specifying the type of abnormality. The "learning mode" is a mode for performing processing for learning the relationship between the feature amount and the normality / abnormality of the device and the type of abnormality.

FIG. 2 is a flow chart showing the processing procedure in the "recognition mode" of the regular operation support device shown in FIG. When the operation input signal of the device under test is input from the plant 1 to the operation signal input unit 2 (step 101), the regular operation support device according to the present embodiment starts operation by using this operation input signal as a trigger. When the operation starts,
First, a device to be subjected to a regular operation test corresponding to this operation input signal (for example, a valve operation signal, a pump start signal, etc.) is specified, and a neural network structure corresponding to the device is specified as described later in detail. (The number of each of the input layer, the intermediate layer, and the output layer, and the weighting coefficient of the unit in each layer are stored in the learning result storage unit 8) (step 10).
2).

Then, the plant data capturing section 3 captures a plant signal related to the operation of the device under test, and the plant data storing section 4 stores this plant signal (step 103). From the plant signal, the characteristic amount calculation unit 5 calculates the characteristic amount (described in FIG. 3) used to determine whether or not there is an abnormality in the device under test (step 10).
4) The feature quantity is output to the neural net having the structure determined in step 102 (step 105). The abnormality determination unit 10 determines whether there is an abnormality and identifies the type of abnormality based on the output result of the neural network (step 1
06), the judgment result is displayed (step 107).

FIG. 3 is an explanatory diagram of the characteristic amount used in the valve closing test when the device under test is a valve. In FIG. 3, a signal 801 indicated by a broken line is a valve opening / closing operation signal output to the valve by the control panel based on an operation input signal input by the operator. A signal 802 indicates the change over time of the valve opening when the valve is actually operated by the operation signal 801. The change 802 in the valve opening indicates the operation signal 8
A value closer to 01 changes indicates that the valve is operating normally. In this valve closing test, the feature amount calculated by the feature amount calculation unit 5 is (I) the difference between the operation signal value of the valve opening decrease rate and the actually measured value (II) the operation signal value of the valve opening and the actually measured value Sum of errors (III) Sum of errors squared (IV) This is the delay time until the valve opening decreases. These feature quantities are calculated with reference to the input time point of the valve operation input signal.

As in the valve closing test described with reference to FIG. 3, the characteristic amount used for abnormality determination is defined in advance for each test item. The characteristic amount calculation unit 5 calculates the characteristic amount corresponding to the test item from the captured operation input signal and plant signal, and outputs it to the neural network calculation unit 7 (step 105 in FIG. 2).

FIG. 4 is a schematic diagram showing the structure of the neural network. As shown in FIG. 4, the neural network operation unit 7 is composed of an input layer, an intermediate layer, and an output layer. 7
Reference numeral 01 is an input layer of the neural network, 702 is an intermediate layer, and 703 is an output layer. Each layer is composed of a unit 704 indicated by “◯”, and units are connected by 705. Input data 706 is input to each unit 704 of the input layer 701. The feature amount calculated by the feature amount calculation unit 5 is used as this input data. Reference numeral 707 is teacher data used when learning the neural network. 7
Reference numeral 08 represents the output of the neural network.

From the neural network, a code number indicating the normal / abnormal mode is output. Normal / abnormal mode refers to the normal, abnormal, and abnormal types of equipment that are defined for each test item. For example, in the case of the test item “valve A / closed test”, the normal / abnormal mode is “normal”, “sticking”,
There are three types of "play". A code number defined by a binary number is assigned to each normal / abnormal mode. For example, "Normal" is "00" and "Stuck" is "1".
“01” corresponds to “0” and “play”. The neural network outputs the code numbers of these normal / abnormal modes with the feature amount calculated by the feature amount calculation unit 5 as an input.

In the regular operation support device according to this embodiment,
The input layer 701, the intermediate layer 702, and the input layer 702 of this neural network
The number of units forming each layer of the output layer 703 and the weighting factor in the coupling between the units are calculated in step 102 of FIG.
In, it is decided for each test item. That is, the structure of the neural network is changed every time the test item is changed.
The structure information corresponding to the test item is stored in the learning result storage unit 8.

FIG. 5 is a block diagram of the learning result storage unit 8. The learning result storage unit 8 stores, for each test item, the number of units of the input layer, the intermediate layer, and the output layer, and the weighting coefficient at that time. The weighting coefficient is obtained when the neural network operation unit 7 obtains the relationship between the feature amount and the code number indicating the normal / abnormal mode by learning.

In step 105 of FIG. 2, the abnormality determining unit 1
0 determines the normal / abnormal mode based on the output value of the neural network operation unit 7, but at this time, the abnormality determination data in the abnormality determination data storage unit 11 is used. In FIG.
The structure of the abnormality determination data storage unit 11 is shown. In the example shown in FIG. 6, the output value corresponding to the test item “valve A / close test” is “00” when the valve is normal, “10” when the valve is stuck, and “01” when the valve is loose.

FIG. 7 is a flow chart showing the processing procedure of the abnormality judging section 10. The abnormality determination unit 10 takes in the output value of the neural network and the operation input signal from the neural network operation unit 7 (step 201), and binarizes the output value of the neural network (step 202). This is because the output of the neural network operation unit 7 is an analog value, and this analog value is associated with the binary code number. For example, the neural network calculation unit 7
Outputs a value of "0.01,0.99", the abnormality determination unit 10
Binarizes this into "0,1" according to a predetermined threshold value.

Next, the corresponding test item in the abnormality determination data storage section 11 is retrieved from the operation input signal (step 20).
3) In the next step 204, the output value of the corresponding test item (FIG. 6) is compared with the output value of the neural network binarized in step 202. Then, the normal / abnormal mode is determined by comparing the two (step 205), and the determination result is output (step 206), or if the determination cannot be made, information that cannot be determined is output (step 20).
7). For example, in the case of the test item “valve A / closed test” of the abnormality determination data shown in FIG.
If the binarized value of the output value of “01” is “01”, the normal / abnormal mode is determined to be “play” corresponding to “01”.

FIG. 8 is a diagram showing an example of a screen for displaying the determination result of the abnormality determining section 10. In the present embodiment, together with the abnormality determination result, the graph of the characteristic amount value used for the abnormality determination and the test result is displayed together with the past data of the same determination result.

Next, the processing in the "learning mode" of the regular operation support device according to this embodiment will be described. The “learning mode” refers to a process of learning the relationship between the normal / abnormal mode of the device and the corresponding feature amount when the corresponding feature amount is known. "Learning mode" is "online learning mode"
And "Offline learning mode". The "online learning mode" refers to a process of learning when an operation input signal and a plant signal are fetched from the plant. The “offline learning mode” refers to a process of learning based on the characteristic amount accumulated in the storage device.

FIG. 9 is a flowchart showing a processing procedure in the "online learning mode". Before operating the regular operation support device in this online learning mode, the operator inputs from the input device 9 shown in FIG. 1 that the online learning process is to be performed and the device to be learned and its normal / abnormal mode. To do. Then, when the operation input signal is input (step 301), the structure of the neural network is determined based on the operation input signal (step 302). Next, the plant data acquisition unit 3 acquires the plant signal and stores it in the plant data storage unit 4 (step 303). Next, the feature amount calculation unit 5
Calculates a feature amount from the operation input signal and the plant signal (step 304). The neural network operation unit 7 receives the normal / normal input from the input device 9 in the next step 305.
The relationship between the abnormal mode and this feature quantity is learned (step 3
06), the learning result is stored in the storage unit 8 (step 30).
7).

FIG. 10 is a flow chart showing a learning processing procedure in the neural network computing section 7. When the operator takes in the learning command signal input from the input device 9 (step 401), the neural network structure corresponding to the test item is determined (step 402). The information about the neural network structure is stored in the learning result storage unit 8 in FIG. 5, as described above. The neural network operation unit 8 configures a neural network having the same number of units as the input layer, the intermediate layer, and the output layer stored in the learning result storage unit 8. Next, the feature amount calculated by the feature amount calculation unit 5 is input to the input layer of the neural network (steps 403, 4).
04), the code number of the normal / abnormal mode corresponding to the characteristic amount is input to the unit of the output layer as a teacher signal (step 405). The neural network learns the relationship between the input and the output by the error backpropagation algorithm (step 40).
6). This process is repeated for the number of test data (step 407). The weighting factor determined by the learning process of the neural network operation unit 7 is stored in the learning result storage unit 8 and used in the above-mentioned “recognition mode”.

FIG. 11 is a flowchart showing a processing procedure in the "offline learning mode" of the regular operation support device according to the present embodiment. In accordance with the test item input by the operator from the input device 9, the feature amount storage unit 6 reads the feature amount corresponding to this test item (step 501).
Then, the structure of the neural network is determined according to this test item (step 502). FIG. 12 is a configuration diagram of the feature amount storage unit 6. The feature amount storage unit 6 stores the past feature amount for each test item and normal / abnormal mode.

The neural network operation unit 7 inputs the feature quantity as input data of the input layer, inputs the code number of the normal / abnormal mode as a teacher signal to the output layer (step 503), and performs learning by the error back propagation algorithm (step 503). Fifty
4). Then, the weighting factor determined by this learning is stored in the learning result storage unit 8 (step 505). Through the above processing, the weighting factor of the neural network is updated based on the accumulated past data.

In the above-described embodiment, the valve has been described as an example, but the same effect can be obtained even when other devices are targeted. FIG. 13 shows an example of obtaining the characteristic amount of the pump as the target device. In the pump startup test, several types of valves are opened and closed to adjust the pump flow rate. The regular operation support device determines, based on an opening signal of a predetermined valve, that the starting test of the pump has started, and determines the structure of the neural network corresponding to the pump.

The pump start-up test is conducted for each process amount (flow rate,
The discharge pressure, suction pressure) can be classified into nine states shown in FIG. 13 depending on whether the state is an increasing state, a decreasing state, or a constant state. Then, the feature amount in each state is obtained. When the process amount is in a constant state, calculate the average value,
When it is in the increasing (decreasing) state, the increasing rate (decreasing rate) is obtained and these are used as the feature values. The number of feature quantities used in the pump start-up test is 3 x 9 = 2 because there are three types of process quantities.
It becomes 7. Then, the regular operation support device determines which state the pump is in based on the valve opening / closing signal, and obtains the characteristic amount thereof. Hereinafter, the feature amount is input to the neural network, and the process of determining normality / abnormality and identifying the type of abnormality is performed.

According to the regular operation support device according to the above-described embodiment, the device to be tested is operated, a plurality of characteristic amounts are calculated from the plant signal related to the operation of the device, and the mutual mutual operation of the plurality of characteristic amounts is performed. Since the normality / abnormality of the device is determined based on the relation or pattern of, it is possible to perform highly accurate determination and specify the type of abnormality.

Since the characteristic amount is obtained from the time change of the state amount with reference to the time point when the operation input signal is input,
It is possible to easily measure the delay time between the operation input signal and the operation of the device, and further, with reference to this delay time, it is possible to distinguish between normal and abnormal and to specify the type of abnormality. Furthermore, the time change of the input plant signal is
Since it is fixed at the reference time point as a boundary, it is possible to make the change patterns of the state quantity for normal and abnormal states similar to those in the case where the reference time point is not provided.

Further, a neural network is used to find the relationship between a plurality of feature quantities obtained from an operation input signal for operating the equipment and a plant signal relating to the operation of the equipment, and whether the equipment is normal or abnormal and the type of abnormality. Because of the configuration, it is possible to perform determination for normality / abnormality and learning for improving the specific recognition rate of the type of abnormality. Further, since the structure of the neural network is changed according to the test item, the structure of the neural network suitable for the device abnormality judgment can be selected, and the accuracy of the abnormality judgment according to the test item can be easily improved. .

[0032]

According to the present invention, it is possible to accurately and easily confirm the normality / abnormality of equipment in a regular operation test and the type of abnormality when an abnormality occurs, thereby repairing the equipment. It is possible to efficiently and accurately perform, and it is possible to improve the reliability of the plant.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a regular operation support device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing a processing procedure in a recognition mode.

FIG. 3 is a schematic diagram illustrating a feature amount.

FIG. 4 is a schematic diagram illustrating the structure of a neural network.

FIG. 5 is a schematic diagram illustrating a configuration of a learning result storage unit.

FIG. 6 is a schematic diagram illustrating a configuration of an abnormality determination data storage unit.

FIG. 7 is a flowchart showing a processing procedure for abnormality determination.

FIG. 8 is a schematic diagram showing a display of an abnormality determination result.

FIG. 9 is a flowchart showing a processing procedure in a learning mode.

FIG. 10 is a flowchart showing a processing procedure for online learning in a neural network operation unit.

FIG. 11 is a flowchart showing a processing procedure of offline learning in a neural network operation unit.

FIG. 12 is a schematic diagram showing a configuration of a feature amount storage unit.

FIG. 13 is a diagram illustrating a characteristic amount of a pump startup test.

[Explanation of Codes] 1 ... Plant, 2 ... Operation signal input unit, 3 ... Plant data acquisition unit, 4 ... Plant data storage unit, 5 ... Feature amount calculation unit, 6 ... Feature amount storage unit, 7 ... Neural network calculation unit , 8 ... Learning result storage unit, 9 ... Input device, 10 ... Abnormality determination unit, 11 ... Abnormality determination data storage unit, 12 ... Display control unit, 13 ... Display device.

Claims (8)

[Claims] 1. An operation signal for operating a device constituting a plant is taken in, a plant signal related to a device under test specified by the operation signal is taken in, and a structure of a neural network is determined according to the device under test. , A plurality of types of feature quantities are obtained from the time change of a plant signal related to the test target device, each feature value is input to the neural network, and normality / abnormality of the test target device is determined from the output of the neural network and And / or a regular operation support method characterized by identifying the type of abnormality. 2. The normal, abnormal and / or
Alternatively, by comparing each code number with the output result of the neural network based on the stored contents of the abnormality determination data storage means that stores each code number corresponding to the type of abnormality, determination of normality / abnormality and / or abnormality A regular operation support method characterized by identifying the type of the operation. 3. The neural network according to claim 2, wherein the neural network learns a value obtained by learning the number of constituent units of each layer of an input layer, an intermediate layer, and an output layer and a weighting coefficient between layers for each test item of a device under test. The number of units and the weighting coefficient of each layer corresponding to the test item are determined from the stored contents of the learning result storage unit based on the test item of the device under test specified by the operation signal. Regular operation support method. 4. The neural network according to claim 3, wherein the feature quantity is input to an input layer for each test item, and the code number is input to the output layer as teacher data to perform learning. Regular operation support method. 5. A means for fetching an operation signal for operating a device constituting a plant, a means for fetching a plant signal related to a device under test specified by the operation signal, and a structure determined according to the device under test. And a feature quantity calculation means for obtaining a plurality of types of feature quantities from the time change of the plant signal related to the device to be tested and inputting each feature quantity to the neural network, and the output of the neural network to perform the test. A regular operation support device, comprising: an abnormality determination unit that determines whether the target device is normal or abnormal and / or specifies the type of abnormality. 6. The abnormality determination means according to claim 5, further comprising abnormality determination data storage means for storing code numbers corresponding to normal, abnormal and / or abnormal types, each code number and the neural network. The regular operation support device characterized by making a determination by comparing with the output result of. 7. The neural network according to claim 6, wherein the neural network stores a value obtained by learning the number of constituent units of each layer of an input layer, an intermediate layer, and an output layer and a weighting coefficient between each layer for each test item of a device under test. A regular operation support device, comprising: the learning result storage means, and determining the number of units and weighting coefficient of each layer corresponding to the test item based on the test item of the device under test specified by the operation signal. 8. The neural network according to claim 7, wherein the feature quantity is input to an input layer for each test item, and the code number is input to the output layer as teacher data to perform learning. Regular operation support device.