JPH07190849A - Vibration diagnosing device for bearing of rotation shaft - Google Patents

Abstract

PURPOSE:To enable sure diagnosing of various kinds of failures including noise vibration, and prevent an erroneous diagnosis. CONSTITUTION:Provided are a frequency analyzer 1 to analyze in spectrum the envelope waveform of vibration signals in a rotation shaft system and take out the spectrum component, a failure recognition means 2 constituted of a neural network which studies the relation between the spectrum component and the kind of failure by inputting in advance the spectrum components of failure relating to the rotation shaft system and outputs the signals of each different operation result for the input of spectrum components obtained from a certain kind of failure relating to the rotation shaft system, a complex failure of overlapping a plurality of failures and the noise vibration other than the rotation shaft system, and a diagnosing means 3 to judge the kind of failure based on the amplitude of the signals of the operation results obtained from the failure recognition means.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of a rotary bearing vibration diagnosing device for diagnosing the type of vibration (type of failure) occurring in a rotary bearing system of a rotary machine.

[0002]

2. Description of the Related Art Various kinds of failures can be considered as failures of a rotary bearing system of a motor, and the appearance of spectrum varies depending on the kind of the failure. For example, if the rotary bearing is scratched, the pitch frequency of the abnormal vibration can be calculated from the scratched position and the structure of the rotary bearing. It can also be seen that multiple harmonic components are generated in. Furthermore, the pitch frequency and its harmonic components can be observed from the spectrum of the envelope waveform during this abnormal vibration.

Therefore, the conventional motor bearing vibration diagnosing apparatus utilizes the characteristic of abnormal vibration, sets a threshold value for each of a specific spectrum component or a plurality of spectrum components appearing at the time of the abnormal vibration, and determines the spectrum component of vibration. When a threshold value exceeds a preset threshold value, it is determined that there is a failure.

[0004]

By the way, the rotating machine to be diagnosed is installed in various environments, but when it is affected by a phenomenon other than a failure related to the rotary bearing system, that is, noise vibration from the environment. There are many. Moreover, many of these noise vibrations have peaks that are very similar to the failure spectrum, which may cause erroneous diagnosis.

Therefore, conventionally, as a means for preventing such erroneous diagnosis, it is possible to reduce the probability of erroneous diagnosis by setting a large threshold value of the fault spectrum component or repeating the diagnosis a plurality of times. Has been done.

However, such a diagnosis method is not sufficient as a countermeasure against noise vibration, and since a diagnosis is repeated a plurality of times, a diagnosis result cannot be obtained quickly, and the efficiency of the diagnosis work is poor. I have a problem. Further, even if a failure occurs, not only the type of the failure cannot be specified accurately, but also there is a problem that when the failure is a composite failure in which a plurality of types of failures of the rotary bearing overlap, it cannot be determined that the failure is the compound failure.

The present invention has been made in view of the above circumstances, and provides a rotary bearing vibration diagnosis device capable of surely diagnosing various kinds of failures including noise vibration and thereby preventing erroneous diagnosis. With the goal.

Another object of the present invention is to provide a rotary bearing vibration diagnosing device which informs the contents of various diagnostic results and contributes to quick maintenance improvement.

[0009]

In order to solve the above problems, the invention according to claim 1 is directed to a rotary bearing for spectrally analyzing the envelope waveform of vibration of a rotary shaft system and diagnosing a failure from the obtained spectrum component. In a vibration diagnosis device, the spectrum component of a failure related to the rotating shaft system is input in advance, the relationship between the spectrum component and the kind of failure is stored, and a certain kind of failure related to the rotating shaft system,
Fault recognition means for outputting signals of different computation results to the input of the spectrum component obtained by a complex fault in which a plurality of types of faults overlap and noise vibration other than the rotating shaft system, and a computation result obtained from the fault recognition means Is a rotary bearing vibration diagnosing device which is provided with a diagnosing means for judging the type of failure based on the magnitude of the signal.

Next, the invention according to claim 2 is, in a rotary bearing vibration diagnosis apparatus for spectrally analyzing an envelope waveform of vibration of a rotary shaft system and diagnosing a failure from a spectrum component obtained, in advance relating to the rotary shaft system. The spectrum component of the fault to be input is stored and the relationship between the spectrum component and the type of fault is stored, and a certain type of fault related to the rotary shaft system, a complex fault in which a plurality of types of faults overlap, and noise other than the rotary shaft system. Fault recognizing means for outputting different calculation result signals to the input of the spectrum component obtained by vibration, and diagnosis for judging the type of fault based on the magnitude of the calculation result signal obtained from the fault recognizing means. Bearing and a message creating and outputting means for explaining the diagnosis result based on the determination result of this diagnosis means It is a vibration diagnostic equipment.

In the invention corresponding to claim 3, a neural network is used as the failure recognizing means of the invention corresponding to claims 1 and 2, and the combination is made by learning the relationship between the spectrum component and the kind of failure. As weighting factors: a. The coupling weight coefficient between a neuron in the input layer that inputs a spectral component that appears only in a certain type of fault and a certain neuron in the output layer prepared to represent the certain type of fault is positive. Set a value, b. The coupling weight coefficient between a neuron in the input layer that inputs a spectral component that appears only when a certain type of fault and a neuron in the output layer that is prepared to represent another type of fault are negative. Set the value, c. The coupling weight coefficient between the neuron that inputs the spectral component of a fault that is not related to the type of fault related to the rotating axis system and the neuron in the output layer has a constant value regardless of the type of fault. If a negative value is set, a value close to zero is set when the value of the spectral component changes at random, and a fault related to the rotating shaft system, a complex fault and The signals of different calculation results are output for the spectral components obtained by the noise vibration.

[0012]

Therefore, in the invention corresponding to claim 1, by taking the above-mentioned means, the failure recognizing means inputs the spectrum component of the failure related to the rotating shaft system in advance and detects the failure of the spectrum component and the failure. Stores the relationship with the type, and signals of different operation results for certain types of faults related to the rotary axis system, complex faults in which multiple types of faults overlap, and spectrum component input due to noise vibration other than the rotary axis system. I am trying to output
It is possible to appropriately determine a certain type of fault, a complex fault, and noise vibration other than that of the rotating shaft system from the signal of the calculation result, and it is possible to reliably prevent erroneous diagnosis.

Next, in the invention according to claim 2, the diagnostic means determines the type of the failure based on the signal of the operation result obtained from the failure recognition means, and the message is created based on the determination result of the diagnostic means. The output means notifies the type of failure and the information necessary for maintenance, so that necessary measures can be taken promptly.

Further, in the invention corresponding to claim 3, as the coupling weight coefficient by learning the relationship between the spectrum component and the kind of the fault, a positive value, a negative value and an approximate value are obtained according to various fault states. By setting a value close to zero, it is possible to appropriately output signals of different calculation results for the types of faults related to the rotary shaft system, complex faults, and spectrum components obtained by the noise vibration,
As a result, the diagnosis result can be output reliably and without erroneous diagnosis.

[0015]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the device of the present invention. This diagnostic device includes a frequency analyzer 1 that extracts a spectrum component of an envelope waveform from a vibration signal observed by a sensor, and a certain type of failure related to a rotating shaft system based on the spectrum component extracted by the frequency analyzer 1. Relates to a malfunction related to the rotating shaft system that recognizes noise vibrations, etc. other than complex failures in which multiple types of failures overlap and failures related to the rotating shaft system, and outputs different calculation result signals, resulting in incorrect recognition. Failure recognition means 2 capable of removing noise vibration components
And a diagnostic means 3 for determining a fault related to the rotary shaft system, that is, a certain type of fault and a plurality of types of complex faults from the signal of the calculation result of the fault recognizing means 2.

The frequency analyzer 1 includes a filter 11 for extracting a signal of a required frequency band among analog signals from the sensor, and an A / D converter for converting the analog signal output from the filter 11 into a digital signal. 1
2. Envelope conversion of the digital signal converted by the A / D conversion unit 12 to obtain the characteristic of the beat vibration of the rotary shaft system, that is, the envelope waveform, and the envelope waveform of the envelope conversion unit 13 A digital Fourier transform unit 14 for analyzing and extracting a spectral component is provided, and the spectral component is sent to the fault recognizing means 2.

The fault recognizing means 2 is composed of a neural network as shown in FIG. In this neural network, assuming that the number of spectrum components used for diagnosis is N and the number of fault types to be diagnosed is M, the number of neurons of the neuron 21 forming the input layer is N and the number of neurons of the neuron 22 forming the output layer is M. It is composed of a two-layered layered neural network.
In this neural network, a connection weight coefficient is set between the input neuron 21 and the output layer neuron 22, and this connection weight coefficient is set based on an operation called learning.

Specifically, the spectrum component of the vibration that should be observed or theoretically observed at the time of a past failure is input to the neuron i in the input layer, and at this time, the bit string data and the teacher signal generated from the output layer are generated. The bit weight data (teaching signal) representing the type of failure observed at the time of the failure is compared from the unit 23, and the connection weighting coefficient is learned while varying so that these bit data are equal, and stored in the network. That is, the relationship between the spectrum component of vibration and the failure type is learned and stored in the network in the form of a coupling weight coefficient.

At the time of actual fault diagnosis, the vibration spectrum component obtained from the frequency analyzer 1 is given to the i-th neuron i in the input layer, and the signal of the operation result obtained from the required neuron on the output layer side is used as a diagnostic means. 3 to determine the type of failure.

The diagnostic means 3 has a threshold value set in advance. When the output value of the failure recognition means 2 exceeds a predetermined threshold value, it is determined that there is a failure. A failure type is assigned to each of the 22, ..., The failure represented by the neuron 22 that outputs the largest signal is determined to be the first failure type, and the failures are output in the order of the output values of the other neurons 22. Judged as a type,
When signals with values of approximately the same magnitude are output from a plurality of neurons, it is determined as a composite fault in which a plurality of faults related to the plurality of neurons overlap each other.

Next, the operation of the apparatus configured as described above will be described.

Now, the signal from the sensor is introduced into the filter 11 where it removes electrical vibration noise in the low frequency range. The signal after removing this low frequency noise component is
A spectrum of vibration that has been digitally converted by the A / D conversion unit 12, further obtained by the envelope conversion unit 13 and then subjected to frequency analysis of the envelope waveform by the digital Fourier conversion unit 14 to remove high frequency noise having no periodicity. The component is taken out and supplied to the input layer 21 of the neural network which constitutes the fault recognition means 2.

In this neural network, the output of the neuron 21 is multiplied by a connection weighting coefficient obtained by learning in advance, and each neuron 2 in the next layer, for example, the output layer 2
Enter 2. Each neuron 22 constituting this output layer
Calculates the total of all the signals sent by multiplying the connection weighting factors, and the total value as the calculation result is used as the diagnostic means 3
Send to.

Here, learning of the connection weight coefficient in the neural network will be described. The coupling weight coefficient is determined by, for example, a general learning method (back propagation method). Specifically, a. A teacher signal corresponding to a certain failure is assigned to each neuron 22 in the output layer.

B. A spectrum value used for learning is assigned and input to each neuron 21 in the input layer.

C. The output of each neuron 21 of the input layer is connected to each neuron 22 of the output layer, and the output of the neuron 21 is multiplied by the connection weight coefficient for each connection to be used as the input of each neuron 22 of the output layer.

D. Each neuron 22 in the output layer receives the multiplication signal from each neuron 21 having a connection relationship, and sets the total value of the multiplication signals as an output value.

E. The neuron 22 whose output value has been determined
Is given the following teacher signal. That is, for a vibration spectrum of a certain type of fault in the rotating axis system, a teacher signal “1” is output to a neuron 22 having an output layer that represents that type of fault, and to another neuron 22 that constitutes another output layer. Gives a teacher signal "0", and these neurons 22
Find the error between the output value and the teacher signal.

F. This error value is multiplied by the output value of the input layer neuron 21, and is further multiplied by a constant α (> 0, <1 value), and the obtained value is the correction amount of the coupling weight coefficient between the input layer and the output layer. Then, the value is added to the value of the coupling weight coefficient at that time.

In the series of processes of b to f, the neurons 22 in the output layer are processed for all the prepared data.
Until the sum of the squares of the errors between the output value of 1 and the teacher signal of the teacher signal generator 23 becomes less than or equal to a preset threshold value. By this learning operation, the neurons 2 in the input layer
It is possible to determine the connection weighting coefficient between 1 and the neuron 22 in the output layer.

As a result, by the above learning operation,
You can set the weighting factors that have the following characteristics:

A. A positive value is given to the coupling weight coefficient between the neuron 21 of the input layer that inputs the spectral component that appears only when the fault A of a certain type and the neuron 22 of the output layer prepared to represent the fault A of that type Is set.

B. The coupling weight coefficient between the input layer neuron 21 that inputs the spectral component that appears only when a certain type of fault A and the neuron 22 that has the output layer prepared to represent another type of fault B are negative. The value of is set.

C. In the case of the connection weight coefficient between the neuron 21 which inputs the spectral component unrelated to the specific kind of fault and the neuron 22 of the output layer, the following two setting methods are applied. That is, c-1. A negative value is set when the value of the spectrum component has a constant value for any kind of failure.

C-2. When the value of the spectral component changes at random, a value close to zero is set.

Therefore, if the above-mentioned coupling weighting factors are set in the coupling line between the input layer and the output layer of the neural network, the spectral component of the frequency analyzer 1 will have a spectral component related to the type of failure. The more included, the larger the output value from the neuron 22 having the output layer. On the other hand, when the spectrum component of a pattern different from that at the time of learning is input, the output of the neuron 22 does not have a very large value. This makes the neural network less susceptible to the influence of noise vibration as compared with the conventional case where the determination is simply made by using the spectral component.

When the spectrum analysis result output from the frequency analyzer 1 is transmitted to the neural network, a part of the spectrum analysis result is extracted and input to the neural network to remove a component in which noise is likely to be superimposed. You can Furthermore, a real-valued sequence of 0 to 1 is output as the output value of the neural network, but if the diagnostic means 3 selects one having a large output value, a meaningless output of the neural network, such as noise vibration, is generated. Can be removed.

Next, a concrete example when the device of the present invention is applied to an actual device will be described.

(Specific Example 1) FIG. 3 is a diagram showing a structural example in which the device of the present invention is applied to a vibration signal preprocessing device for a small motor. This pre-processing device pre-processes the observed vibration spectrum after the neural network forming the failure recognizing means 2 is made to learn beforehand the relationship between the type of failure and the spectrum component.

This device comprises a frequency analyzer 1, a fault recognizing means 2 having a neural network structure, a diagnosing means 3, and an interface 31 for sending the value of the diagnosis result to a host computer. Among them, the filter 11 of the frequency analyzer 1 has a characteristic of cutting a signal component of 1.5 KHz or less. As shown in FIG. 4, the neural network serving as the fault recognizing means 2 uses 64 input layer neurons 32 and 4 output layer neurons 33 to 36. Then, this neural network is subjected to the following preprocessing prior to use. (A) Collect the failure vibrations of bearing balls, outer rings, and inner rings and normal vibrations of a small motor of the same type as the diagnosis target motor and normal vibrations, perform frequency analysis of the envelope processing result, and extract spectral components. (B) Next, the type of failure for each spectral component created in (a) above is investigated, and in the case of a ball injury, the output layer neuron 33, in the case of an outer ring injury, the output layer neuron 34, of the inner ring injury. In this case, the neural network is made to learn each spectral component value and the type of failure so that the output layer neuron 35 outputs "1" from the output layer neuron 35 and the output layer neuron 36 outputs under normal conditions.

Therefore, according to the configuration of the apparatus as described above, when the vibration signal of the small motor to be diagnosed is detected by the sensor and introduced into the frequency analyzer 1, here, after the low frequency component is cut, The envelope is taken, and the numerical sequence of spectral components obtained by the result of frequency analysis is processed by the neural network shown in FIG. The processing result of this neural network is
The diagnosis means 3 converts and outputs the information on the cause of vibration (kind of failure). The vibration cause candidate, which is the output of the diagnostic means 3, is sent to the host computer via the interface 31, and the necessary data processing is performed.

(Specific Example 2) FIG. 5 is a diagram showing a specific example of a bearing vibration diagnosing device for a small motor.

This apparatus includes, in addition to the frequency analyzer 1, the fault recognizing means 2 and the diagnosing means 3, a coefficient storing means 41 for storing the coupling weight coefficient of the neural network which constitutes the fault recognizing means 2, a fault type and other faults in advance. Message data storage means 42 for storing message data necessary for maintenance of the data and message creation output means 43 for explaining the diagnosis result by taking out the necessary message data from the message data storage means 42 based on the output results of the diagnosis means 3. It is composed by.

The coupling weight coefficient stored in the coefficient memory 2a is created by the following processing. (A) First, the pitch frequency relating to the scratches on the bearing of the motor is calculated from the motor of the type to be diagnosed and the structure of the bearing. From the pitch frequency obtained as a result of this calculation, the harmonics thereof, and the sampling period of vibration, the spectral component of the envelope waveform is obtained for each position of the scratch on the bearing. (B) Next, the neural network is made to learn the relationship between the flaw position and each spectrum component by using the obtained spectrum component and the flaw position information. (C) The connection weight coefficient of the neural network thus obtained is stored in the coefficient storage means 41.

Thereafter, the vibration spectrum component of the motor obtained from the sensor is input to the neural network, the connection weighting coefficient is taken out from the coefficient storage means 41 and set in the neural network, and is calculated from the output layer neurons of the neural network at that time. The resulting signal is sent to the diagnostic means 3 to determine the type of vibration. At this time, since the coupling weight coefficient is stored in the coefficient storage means 41 in advance, it can be used as a diagnostic device any time after the power is turned on.

On the other hand, the message creation / output unit 43 fetches the necessary message data from the message data storage unit 42 and outputs the necessary message data for the diagnostic result obtained from the diagnostic unit 3. At this time, the interpretation of the diagnostic result is, for example, as shown in FIG.
Interpretation shall be performed according to the algorithm shown in.

First, the sum of the absolute values of the calculation results is calculated for each failure type (ST1). Then, it is determined whether or not the magnitude of the absolute value is less than or equal to the threshold value (ST2). here,
Although the magnitude of the absolute value is less than or equal to the threshold value, it is determined whether the value of the diagnostic result signal is relatively large or small (ST3),
If it is small, it is normal (ST4), and if it is large, it is determined that the cause is unknown (ST5), and the corresponding message data is retrieved from the message data storage means 42 and output.

Next, when the magnitude of the absolute value exceeds the threshold value, the ratio between the positive calculation result and the negative calculation result is calculated (ST6), and which is larger is judged from the ratio calculation result (ST6). ST7). Here, if the positive operation result and the negative operation result are substantially equal, it is determined that the failure is a compound failure including the failure of the relevant type (ST8), and if the positive operation result is large, it is determined that only the failure of the relevant type. (ST9) If the result of the negative operation is further large, it is determined that the failure is of another type (ST10), and the corresponding message data is extracted and output.

In the above embodiment, a small motor has been described as an example, but the same can be applied to the case where vibration is generated in the rotary bearing of a rotating device other than a large motor and other motors. In addition, the present invention can be modified in various ways without departing from the scope of the invention.

[0051]

As described above, according to the present invention, the following various effects are exhibited.

According to the first and third aspects of the present invention, various kinds of faults including noise vibrations can be reliably diagnosed, and the different signal of the calculation result can be output from the fault recognizing means depending on the state of the fault. In addition, an appropriate diagnosis result can be output without erroneous diagnosis.

Next, according to the second and third aspects of the present invention, since messages necessary for maintenance are notified according to the contents of various diagnostic results, it is possible to speed up maintenance.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of a rotary bearing vibration diagnosis device according to the present invention.

FIG. 2 is a configuration diagram of a neural network as an example of the fault recognition means shown in FIG.

FIG. 3 is a configuration diagram showing a specific application example of the device of the present invention.

FIG. 4 is a configuration diagram of a neural network used in the application example of FIG.

FIG. 5 is a configuration diagram showing another embodiment of the device of the present invention.

FIG. 6 is a diagram for explaining the operation of the message creation / output means shown in FIG.

[Explanation of symbols]

1 ... Frequency analyzer, 2 ... Failure recognition means, 3 ... Diagnosis means,
11 ... Filter, 12 ... A / D converter, 13 ... Envelope converter, 14 ... Digital Fourier converter, 43 ... Message creating / outputting means.

Claims (3)

[Claims] 1. A rotary bearing vibration diagnosis apparatus for spectrally analyzing an envelope waveform of vibration of a rotary shaft system and diagnosing a fault from the obtained spectrum component, by inputting the spectrum component of the fault related to the rotary shaft system in advance. The relationship between the spectrum component and the type of failure is stored, and a certain type of failure related to the rotating shaft system, a composite failure in which a plurality of types of failures overlap, and an input of the spectral component obtained by noise vibration other than the rotating shaft system. On the other hand, a failure recognizing unit that outputs a different operation result signal, and a diagnosing unit that determines a failure type based on the magnitude of the operation result signal obtained from the failure recognizing unit are provided. Rotating bearing vibration diagnosis device. 2. A rotary bearing vibration diagnosis apparatus for spectrally analyzing an envelope waveform of vibration of a rotary shaft system and diagnosing a fault from the obtained spectrum component, by inputting the spectrum component of the fault related to the rotary shaft system in advance. The relationship between the spectrum component and the type of failure is stored, and a certain type of failure related to the rotating shaft system, a composite failure in which a plurality of types of failures overlap, and an input of the spectral component obtained by noise vibration other than the rotating shaft system. On the other hand, the failure recognizing means that outputs different operation result signals, the diagnosing means that determines the type of failure based on the magnitude of the operation result signal obtained from the failure recognizing means, and the deciding result of the diagnosing means. Rotational bearing vibration diagnosis, characterized by comprising: message output means for explaining the diagnosis result based on Location. 3. The fault recognizing means uses a neural network, and as a connection weighting coefficient by learning the relationship between the spectrum component and the type of fault, a. The coupling weight coefficient between a neuron in the input layer that inputs a spectral component that appears only in a certain type of fault and a certain neuron in the output layer prepared to represent the certain type of fault is positive. Set a value, b. The coupling weight coefficient between a neuron in the input layer that inputs a spectral component that appears only when a certain type of fault and a neuron in the output layer that is prepared to represent another type of fault are negative. Set the value, c. The coupling weight coefficient between the neuron that inputs the spectral component of a fault that is not related to the type of fault related to the rotating axis system and the neuron in the output layer has a constant value regardless of the type of fault. If the value of the spectral component changes randomly, a value close to zero is set, and a fault of a type related to the rotating shaft system, a complex fault, and The rotary bearing vibration diagnosing device according to claim 1 or 2, wherein signals of different calculation results are output for the spectral components obtained by the noise vibration.