JP3392552B2 - Vibration waveform identification method and identification device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vibration waveform identifying method and an identifying apparatus for determining shaft vibration of a rotating body such as a pump, which is frequently used in various plants such as nuclear power plants.

[0002]

2. Description of the Related Art For example, in a nuclear power plant, the size is about 3
Since 00 pumps are used, ensuring their soundness is a very important matter for the operation of the entire plant. Among them, especially for pumps directly related to core cooling, a device has been developed for constantly monitoring the vibration state during operation.

These monitoring devices generally respond to the signals of a vibrometer such as a displacement meter, a speedometer, and an accelerometer.
In addition to the amplitudes of all frequency components, the fast Fourier transform (Fast F
Fourier analysis (ourier Transform, FFT) etc. makes the rotation synchronization component (1N component) highly sensitive to anomalies.
Also, the amplitude of the higher harmonic component (2N, 3N, ... Component) is monitored.

Further, for some pumps, a rotation pulse signal is measured, and a vibration vector expressing the amplitude and phase of a rotation synchronization component and a harmonic component is monitored with the pulse generation position as a phase reference. . This is because the number of synchronization components and harmonic components that grow due to the occurrence of anomalies and their amplitudes differ depending on the degree of anomalies.
This is because diagnosis is performed in response to the fact that the direction of vibration differs depending on the location of the abnormality.

[0005]

However, when looking at the axial vibration, for example, the phase relationship between the X-direction vibration and the Y-direction vibration, or the phase relationship between the rotation synchronization component and the harmonic component is evaluated. In order to do so, there was the problem that multiple vibration vectors had to be compared.

On the other hand, the locus of the vibration waveform measured in the two directions of X and Y drawn on the XY plane contains all of the above information, but a human can visually judge from this information. However, it is the actual situation that the feature is identified, and it is extremely difficult to quickly identify the one showing a similar trajectory from this large amount of data.

The object of the present invention is to characterize the axial vibration waveform measured in two directions X and Y based on the locus drawn on the XY plane without performing complicated calculation processing such as fast Fourier transform. An object of the present invention is to provide a vibration waveform identifying method and an identifying apparatus capable of automatically identifying an existing waveform having a similar physical pattern.

[0008]

Identification method of the vibration waveform according to the invention of claim 1, wherein in order to achieve the above object, according to an aspect of the the vibration waveform of the shaft vibration signal of the rotating body, measured from the two directions of X and Y The trajectory pattern obtained when displayed on the XY plane is input to the neural network that has been pre-learned with the trajectory pattern of the existing waveform, and the existing waveform most similar to the current vibration state is input. Is selected.

In the vibration waveform identifying method according to the second aspect of the present invention, the vibration waveform displayed on the XY plane is once obtained by averaging data obtained by a plurality of rotations of the rotating body to be measured.
It is characterized in that it is used as the vibration waveform data of the transposition .

In the vibration waveform identifying method according to the third aspect of the present invention, the pattern of the locus is obtained by dividing the XY plane into a grid pattern and determining the proportion of the region where the locus exists in the entire X region. It is characterized in that it is a histogram in the X direction obtained for each Y region, and a histogram in the Y direction obtained for each X region as a percentage of the region in which the locus exists among all the Y regions.

According to a fourth aspect of the present invention, in the vibration waveform identifying method, the pattern of the locus is "1" when the XY plane is divided into a plurality of small areas and a locus exists in each small area. "
In addition, when there are no bits, the number of bit patterns is equal to the number of areas having "0" as a value.

According to a fifth aspect of the present invention, there is provided a vibration waveform discriminating apparatus including a signal input section for inputting signals from two vibrometers and a rotary pulse meter for measuring axial vibrations of a rotating body in two directions X and Y. , together with obtaining the vibration waveform based on the rotation pulse signal, pattern of the locus drawn on the XY plane obtained by dividing the vibration waveform of this advance into a plurality of small areas - providing a pre-processing unit for determining the emission.

Also, the pattern of the trajectory of the existing waveform is newly
A waveform learning unit that learns the coupling coefficient of the neural network so that only the output layer unit corresponding to the waveform shows a constant value when input to the input layer unit of the laural network, and the pattern of the current trajectory. A waveform identifying section is provided for selecting a most similar existing waveform by inputting a waveform into a previously learned neural network.

Further, it is characterized by comprising an output display section for outputting information on the selected existing waveform stored in the waveform database, and a control section for controlling execution of the above processing.

In the vibration waveform identifying apparatus according to the sixth aspect of the present invention, the vibration waveform displayed on the XY plane is averaged once by averaging data obtained by a plurality of rotations of the rotating body to be measured.
It is characterized in that it is used as the vibration waveform data of the transposition .

In the vibration waveform discriminating apparatus according to the invention of claim 7, the patterns of the loci obtained by the preprocessing section are the X-direction histogram and the Y-direction histogram given in claim 3. It is characterized by According to an eighth aspect of the invention, the vibration waveform identifying apparatus is characterized in that the trajectory pattern obtained by the pre-processing unit is the bit pattern given in the fourth aspect.

[0017]

[Action] first aspect of the present invention, the axial vibration in the rotating body with measured in two directions of X and Y, and the vibration waveform displayed in the XY plane is divided into a plurality of small areas in a lattice pattern, the Input the pattern of the locus obtained from time to the neural network that has been learned in advance by the locus of the existing waveform ,
The vibration waveform is identified by selecting an existing waveform that is most similar to the current vibration state.

According to a second aspect of the present invention, the vibration waveform displayed on the XY plane is data obtained by a plurality of rotations of the rotating body.
The vibration waveform data and to Rukoto for one revolution are averaged to be determined by averaging the trajectories of a plurality of rotation
You can

According to a third aspect of the present invention, the pattern of the locus obtained from the axial vibration of the rotating body is divided into a lattice shape on the XY plane, and the proportion of the region where the locus exists in the entire X region is determined. Each Y
The histogram in the X direction obtained for each area and the ratio of the area where the locus exists in the entire Y area are set as the histogram in the Y direction obtained for each X area. The characteristic pattern of this trajectory
By using a histogram as the input, the identification time can be swiftly evaluated because the similarity between the number of harmonic components and the amplitude is evaluated.

According to the fourth aspect of the invention, the pattern of the locus obtained from the axial vibration of the rotating body is divided into a plurality of small areas on the XY plane, and when the locus exists in each small area, " 1 ", and if there is not, it is input to the neural network as bit patterns corresponding to the number of areas having" 0 "as a value. As a result, even when the phases in the X direction and the Y direction are difficult to distinguish by the histogram of the locus,
The vibration waveform can be easily identified.

According to a fifth aspect of the present invention, the two vibrometers cause axial vibrations of the rotating body in two directions of X and Y, and the rotary pulse meter gives one pulse per one rotation in response to a measurement execution command from the control unit. The signal is output to the signal input unit, and the signal input unit transmits the signal to the preprocessing unit. Before the processing unit, with obtaining the vibration waveform based on the rotation pulse signal, the locus drawn on the XY plane obtained by dividing the vibration waveform of this advance into a plurality of small areas pattern -
Ask for

The waveform learning unit learns the trajectory pattern of the existing waveform previously stored in the waveform database by the neural network. In the waveform discriminating unit, the pattern of the current locus is input to the previously learned neural network in the waveform learning unit to select the most similar existing waveform and output it to the output display unit. The output display section also displays information about existing waveforms stored in the waveform database.

According to a sixth aspect of the present invention, in the preprocessing section, X
The vibration waveform displayed on the Y plane can be changed by rotating the rotor multiple times.
The data obtained is averaged and the vibration waveform data for one rotation is
As a result, the characteristics based on the vibration waveform can be obtained.

According to the seventh aspect of the invention, the patterns of the loci obtained by the preprocessing section are a given X-direction histogram and Y-direction histogram, so that the number and amplitude of harmonic components are emphasized. The quick vibration waveform evaluation is performed. In the invention according to claim 8, the pattern of the locus obtained by the pre-processing unit is a bit pattern to identify the vibration waveform when the phases in the X and Y directions are different.

[0025]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to the drawings. In the first embodiment, as shown in the block diagram of FIG. 1, an apparatus for identifying a vibration waveform has an X-axis vibrometer 2 and a Y-axis that are provided in two directions of X and Y facing a rotation axis 1 of a rotating body. The vibrometer 3, a single rotation pulse meter 4, a signal input section 5 for inputting these signals, a control section 6 for issuing a measurement execution command, a preprocessing section 7, a waveform learning section 8, and a waveform identification section 9, It is composed of a waveform database 10 and an output display section 11.

The operation of the above configuration is that the signal input section 5 receives the measurement execution command from the control section 6 and
2 of X and Y on the rotation axis 1 by the Ke axis vibrometers 2 and 3
A plurality of rotations of the output signal measured from the direction and one pulse signal per one rotation of the rotation pulse meter 4 are input. At this time, the rotation pulse signal from the rotation pulse meter 4 is used as a trigger to input the vibration data, so that the data can be accurately measured the designated number of times.

The preprocessing unit 7 divides the input vibration data of a plurality of rotations in time length into a plurality of blocks each having a time length of one rotation, and calculates an average vibration amplitude between the blocks. By taking it, the average vibration waveform data for one rotation is generated. Also, the average values m _X and m _Y of the vibrations in the X and Y directions within one rotation, the maximum values Max _X and Max _Y , and the minimum values Min _X and Min _Y are calculated.

Next, the maximum value on the X-axis is Max _X , the minimum value is Min _X , and there are n _X intervals, and the maximum value on the Y-axis is Max _Y , and the minimum value is Min _Y , and there are n _Y intervals. ,
A trajectory of the obtained average vibration waveform for one rotation is drawn on the XY plane divided into the grid-like small regions. It is desirable that the number of data per rotation is as large as possible. However, even if the number of data is small, the data points are displayed as straight lines so that the pattern can be easily evaluated.

In practice, by assigning one small area to one memory element, the contents of each memory element are set to "0" in advance, and data points or straight lines connecting them are included in the small area. If it passes, the content of the memory element is set to "1". Then, the number h _{Yj of} X regions in the same j-th Y region where a locus exists is obtained, and the locus existence probability normalized by n _X is used as a histogram in the Y direction. Similarly, a histogram in the X direction is also obtained.

The histogram diagram of FIG. 2 shows an example of the locus of the vibration waveform and also an example of the histogram obtained when dividing into n _X = n _Y = 16. In FIG. 2, in the rightmost (j = 16th) vertical 1-column Y region n _Y = 16 that includes the maximum value Max _X of the X axis, the number of regions h _X16 in which the locus exists is 6 Yes, the 16th value of the histogram in the X direction is 6/16 = 0.375.

The pre-processing unit 3 outputs (n _X + n _Y ) values of such a standardized histogram as a characteristic pattern of the locus of the vibration waveform. The waveform learning unit 8 receives a learning execution command from the control unit 6, and has a characteristic pattern of a locus obtained in a known vibration state registered in advance in the waveform database 10.
Learning by a neural network as shown in FIG.

As shown in the schematic view of FIG. 3A, the neural network has an input layer (first layer) and an intermediate layer (second layer).
Layer) and an output layer (third layer), each layer includes a plurality of units represented by circles in FIG.
The output of each unit is a function of the sum of weighted values of the input signal as represented by the following equation, and the weight called the coupling coefficient is learned by prior training.

When the output of the first unit of the k-th layer (for example, the intermediate layer when k = 2) is represented by O ^k _i , the output O
^k _i is calculated by the following equations (1) and (2).

[0034]

[Equation 1]

Here, n _k is the number of units in the (k-1) -th layer, and w ^k-1 _j ^k _i is the output from the j-th unit in the (k-1) -th layer to the i-th unit in the k-th layer. The coupling coefficient, f (z), is a stepwise threshold function or an S-shaped sigmoid function (1+
e- ^cz ) ^-1 .

In FIG. 3B, x ^k _i −h ^k _i = Z,
It shows a structure of one unit when a sigmoid function is used. The sigmoid function is Z = (-∞, +
The value of (0, 1) is taken in the range of (∞), and Z = 0 is the intermediate value f.
Since the threshold value is (Z) = 0.5, h ^k _i is x ^k
called the threshold for _i . Among the parameters of the above equations (1) and (2), the coupling coefficient w ^k-1 _j ^k _i and the threshold value h ^k _i are learned by a feature pattern known in advance. Is performed as follows.

The number of units in the output layer is set to the number of existing waveforms to be identified, and the characteristic patterns of trajectories in various known vibration states are input to the input layer unit, and the waveform pattern -)), The coupling coefficient w ^k-1 _j ^k _i and the threshold value h ^k _i are learned so that only the output layer unit corresponding to (1) outputs “1” and all the other output layer units output “0”. ..

This learning is performed, for example, by the back propagation method described in the publication. (For example, "Neural Network Information Processing-Introduction to Connectionism or Toward Soft Symbols-" by Hideki Aso, Publishing of Industrial Books, 198
See 8. ) The waveform identifying unit 9 inputs the characteristic pattern of the locus of the vibration waveform output from the pre-processing unit 7 into the previously learned neural network, and corresponds to the output layer unit where the output value is maximum. The existing waveform is selected as the one most similar to the vibration waveform just input. This result is output from the output display unit 11 together with the information stored in the waveform database 10 regarding the selected existing waveform, and the vibration waveform is identified.

Next, the second embodiment will be described. According to the above-described first embodiment, for example, the waveform diagram of FIG. 4 shows a signal x _{1 obtained} by combining the amplitudes of 1N, 2N, and 5N variously,
x ₃ , x ₅ , x ₇ and these 4 signals and the phase of each component are 90
° Waveforms of two revolutions of different signals x ₂ , x ₄ , x ₆ , and x ₈ are shown with time taken on the horizontal axis. From the waveform of FIG. 4, for example, a waveform similar to the waveform of x ₁ is x ₂ , and the intuitive similarity of the waveform for identifying this similarity depends on the number and amplitude of harmonic components first rather than the phase relationship. It is presumed to do.

The waveform diagram of FIG. 5 shows the same waveform as that of FIG. 4 for one rotation when the 1N component and the 2N component are combined. Moreover, the locus and the histogram of FIG. 6 show the axial vibration locus drawn when the upper waveform is the axial vibration waveform measured from the X direction and the lower waveform from the Y direction.

FIG. 6 shows histograms in the X and Y directions of the locus, but it can be said that there is no significant difference.
That is, it is considered that using a histogram as the characteristic pattern of the locus emphasizes the number of harmonic components and the degree of similarity in amplitude. Therefore, for example, as shown in A and B in the locus and the histogram diagram of FIG. 7, when two phases are different in the X direction and the Y direction with a signal including only the rotation synchronization component, these two vibration states are plotted in the histogram It was difficult to distinguish by.

The second embodiment discriminates the waveform in accordance with the phase difference in the first embodiment. That is, instead of the histogram shown in FIG. 2, the following pattern is used as the characteristic pattern of the locus generated by the preprocessing unit 7 in FIG.

First, as in the case of FIG. 2, a locus is drawn so that the maximum and minimum values of the XY plane coincide with the maximum and minimum values of the waveform data. At this time, the XY plane may be divided into a plurality of small areas in the same manner as in FIG. 2, but the dividing method is arbitrary and does not need to be in the grid shape.

Considering the same locus as in FIG. 7, FIG.
As shown in the locus diagram of (a) and the bit pattern diagram of (b), if the XY plane is divided into 36 small regions in a grid pattern, a locus exists inside each of these small regions. Then, if "1" is given and "0" is not given, a 36-bit pattern is generated for one trajectory.

In the case of FIG. 8, "1" is a black corner and "0"
Is represented by white corners, the bit pattern is as shown in FIG. 8B for the loci of A and B in FIG. 8A.
This bit pattern is input to the neural network that has input layer units for all areas.

It is considered that the method using the bit pattern has a large number of input layer units of the neural network and requires a long time for learning as compared with the method using the histogram described above. Therefore, it is effective to apply the identification method using the bit pattern only to the vibration waveform which is difficult to identify by the histogram method.

It is also known that a low frequency vibration such as a semi-harmonic component (N / 2 component) may appear in the axial vibration. Here, for example, when the N / 2 component is dominant. When the locus is displayed with the center of rotation as the origin of the coordinates and the locus for the first rotation is on the left half plane, the locus for the next one rotation shifts to the right half plane.

Therefore, the loci of two rotations are averaged to be 1
When the locus for rotation is obtained, a locus approaching the Y axis is obtained, unlike the actual case. In this way, the data for one rotation obtained by averaging from the data for multiple rotations
The amplitude of the data may be underestimated.

However, in the second embodiment, as the waveform for drawing the locus in the pre-processing section 7, instead of the average vibration waveform for one rotation, all waveforms that do not average the time for two or more rotations are used. It uses data. As a result, when the average vibration waveform for one rotation is generated, the waveform data that is not averaged is generated only when the maximum amplitude (maximum value-minimum value) becomes, for example, half or less of the value before averaging. -It is also considered effective to use the trajectory of data.

In addition to this, as the characteristic of the locus, the average value m of the vibrations in each of the X and Y directions obtained by the preprocessing unit 7 within one rotation is m.
Comparing _X and m _Y with existing waveform values is also considered to be one of the simplest waveform identification methods. Furthermore, by providing a means for manually inputting the trajectory of the shaft vibration waveform expected in various abnormal states, the entered trajectory is regarded as an existing waveform and compared and compared to identify an unexperienced vibration state. You can

[0051]

As described above, according to the present invention, it is possible to discriminate the axial amplitude state of a rotating body based on an axial center locus including all information without performing complicated calculation processing such as fast Fourier transform. It is possible to do it automatically in a way that is close to intuition.
As a result, there is an effect that the search for the similar waveform and the diagnosis of the vibration state are quickly performed.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of an identification device of a first embodiment according to the present invention.

FIG. 2 is a locus of a vibration waveform and a histogram thereof.

FIG. 3 is a schematic diagram of a neural network composed of three layers, where (a) shows the configuration and (b) shows the unit structure.

FIG. 4 is a time change diagram of a vibration waveform including three periodic components.

FIG. 5 is a time change diagram of a vibration waveform including two periodic components.

FIG. 6 is a histogram diagram of a shaft vibration locus in FIG.

FIG. 7 is a histogram and a locus of a vibration waveform including one periodic component.

FIG. 8A is a locus diagram of a second embodiment according to the present invention,
(B) is a bit pattern diagram.

[Explanation of symbols]

1 ... Rotation axis, 2 ... X-axis vibrometer, 3 ... Y-axis vibrometer, 4 ... Rotation pulse meter, 5 ... Signal input section, 6 ... Control section, 7 ... Preprocessing section, 8 ... Waveform learning section, 9 ... Waveform identification section, 10 ... Waveform database, 11 ... Output display section.

─────────────────────────────────────────────────── ─── Continuation of front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) G01H 17/00 G01M 19/00 G06F 15/18 560 G21C 17/00

Claims (8)

(57) [Claims] The emissions, pattern of the trajectory of the existing waveform - - 1. A pattern of the trajectory obtained when viewing the vibration waveform of the X and the shaft vibration signal of the rotating body, measured from the two directions Y in XY plane by emissions A method for identifying vibration waveforms, characterized by inputting to a previously learned neural network and selecting an existing waveform that is most similar to the current vibration state. 2. The vibration waveform displayed on the XY plane is obtained by averaging data obtained by a plurality of rotations of the rotating body to be measured.
The method of identifying a vibration waveform according to claim 1 , wherein the vibration waveform data for one rotation is used. 3. The locus pattern is a histogram in the X direction obtained by dividing the XY plane into a grid pattern and obtaining the ratio of the region where the locus exists in the entire X region for each Y region.
The ratio of the area where the locus exists in the entire Y area is calculated as X
The method for identifying a vibration waveform according to claim 1, wherein the histogram is a histogram in the Y direction obtained for each area. 4. The pattern of the locus is divided into a plurality of small areas on the XY plane, and is set to "1" when a locus exists in each small area, and "0" when the locus does not exist. 2. The method of identifying a vibration waveform according to claim 1, wherein the number of bit patterns is the same as the number of regions that the vibration waveform has. 5. A X and the signal input unit for inputting the two signals of the vibration meter and the rotational pulse meter for measuring the axial vibration of the rotary body from the two directions Y, together with obtaining the dynamic waveform vibration based on the rotation pulse signal , X obtained by dividing the vibration waveform of this advance into a plurality of small areas
A pre-processing unit for obtaining the pattern of the trajectory drawn on the Y plane,
Learn the coupling coefficient of the neural network so that when the pattern of the locus of the existing waveform is input to the input layer unit of the neural network, only the output layer unit corresponding to that waveform shows a constant value. Waveform learning section and a neural network that has already learned the pattern of the current trajectory.
Waveform input section that selects the most similar existing waveform by inputting it to the waveform, an output display section that outputs information about the selected existing waveform stored in the waveform database, and execute the above processing. An apparatus for identifying a vibration waveform, comprising a control unit for controlling. 6. The vibration waveform displayed on the XY plane is obtained by averaging data obtained by a plurality of rotations of the rotating body to be measured.
The vibration waveform identification device according to claim 5 , wherein the vibration waveform data for one rotation is used. 7. The vibration waveform according to claim 5, wherein the pattern of the locus obtained by the preprocessing unit is the histogram in the X direction and the histogram in the Y direction given in claim 3. Identification device. 8. The vibration waveform identification device according to claim 5, wherein the trajectory pattern obtained by the pre-processing unit is the bit pattern given in claim 4.