JP7647138B2 - Abnormality diagnosis system and abnormality diagnosis method, frequency fluctuation correction processing device and correction processing method, abnormality diagnosis program, and frequency fluctuation correction processing program - Google Patents

Description

The present invention relates to a technology that sequentially learns the operating state of a diagnostic target and diagnoses abnormalities based on the learning results.

Abnormality diagnosis of equipment such as rotating machines is performed based on constant Q conversion of acceleration data from an acceleration sensor installed in the equipment to be diagnosed (see Patent Documents 1 and 2).

As a basic diagnostic technique, when the subject to be diagnosed continues to operate normally, there is little acceleration data during abnormal times, so the normal/abnormal state of the subject is diagnosed by learning only the acceleration data during normal times and quantifying how much the current acceleration data differs from normal times.

Patent Publication 2017-198619 Patent Publication 2017-198620

(1) Conventional methods for diagnosing abnormalities have the potential to reduce diagnostic accuracy due to the following reasons:

In other words, as mentioned above, because the system learns the state of the diagnostic target during normal operation when the diagnostic target is running stably, unstable acceleration when the diagnostic target is not operating (hereinafter referred to as the "transition period") may be judged to be abnormal.

In addition, the output of the acceleration sensor may not be stable immediately after the acceleration sensor is installed or immediately after the diagnosis target is repaired. For example, the acceleration sensor is installed on the diagnosis target using magnetic attraction, adhesives, bolts, etc., but the sensor output does not stabilize until the installation condition is settled. In addition, when the diagnosis target is repaired, the target device is often bolted to the base, and the sensor output does not stabilize until the bolt tightening is settled.

Acceleration data from such transitional periods when the sensor output is unstable can be larger or smaller than during stable operation, so learning data from those periods will result in a decrease in diagnostic accuracy.

Furthermore, as shown in Figure 1, when the acceleration data of the diagnostic target for a certain period (for example, one week or one month) is learned, annual fluctuations occur based on that period. Due to the influence of this annual fluctuation, even if the diagnostic target is normal, the threshold may be exceeded and an abnormality may be falsely detected. This annual fluctuation is thought to be caused by the expansion and contraction of the diagnostic target due to temperature, and in the case of pumps, the amount of air dissolved in water changes.

(2) Therefore, a method has been proposed in which, when learning and diagnosing acceleration data obtained by an acceleration sensor, the frequency data converted from the acceleration data is divided into octave frequencies, feature values such as average values are calculated for each divided section, and sections determined to be transitional periods based on the calculated feature values are excluded from the learning and diagnosing targets.

As shown in Figure 2(a), this method does not cause any problems if the base 3 of the rotating machine 1 is strong. However, as shown in Figure 2(b), if the rotating machine 1 is installed on an unstable base 3, the acceleration data from the acceleration sensor 2 will be affected not only by the vibrations of the rotating machine 1, but also by the vibrations of the base 3.

As a result, as shown in Figure 3, the vibration of the base 3 causes the frequency to fluctuate, and there is a risk that the original frequency of the rotating machine 1 will not appear in the data (learning data/frequency data) that is the subject of learning and diagnosis.

Furthermore, due to the fluctuations, there is a risk that the learning data and diagnostic data will change significantly depending on which section the frequency component falls in at the position of the separator. For example, if a component with a large rotation frequency falls at the position of the separator, the feature amount will differ significantly depending on which section it falls in.

The present invention was made to solve these conventional problems, and aims to improve diagnostic accuracy by suppressing the effects of fluctuations in the amplitude of acceleration data.

(1) One aspect of the present invention is an anomaly diagnosis system that sequentially learns an operating state of an object to be diagnosed and diagnoses an anomaly,
a correction processing unit that performs fluctuation correction based on a moving average of an acceleration amplitude obtained from the acceleration data of the diagnosis object;
a sequential learning processing unit that sequentially learns the result of the fluctuation correction and generates a learning result model;
and a diagnosis processing unit that uses the learning result model to determine the presence or absence of the abnormality based on current acceleration data of the diagnosis target.

(2) Another aspect of the present invention is a device for correcting fluctuations in acceleration amplitude of acceleration data of a diagnosis target when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising:
The acceleration data is frequency-converted to obtain frequency data, and a moving average unit is provided for performing a moving average process of the frequency data to correct the fluctuation.

(3) Yet another aspect of the present invention is an anomaly diagnosis method for diagnosing an anomaly by sequentially learning an operating state of an object to be diagnosed by a computer, comprising:
a correction processing step of performing fluctuation correction based on a moving average of acceleration amplitude obtained from the acceleration data of the diagnosis object;
a sequential learning processing step of sequentially learning the result of the fluctuation correction to generate a learning result model;
and a diagnosis processing step of determining the presence or absence of an abnormality based on current acceleration data of an object to be diagnosed using the learning result model.

(4) Yet another aspect of the present invention is a method for correcting fluctuations in acceleration amplitude of acceleration data of a diagnosis target by a computer when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising:
The method includes a moving average step of performing a moving average process of a frequency based on frequency data obtained by frequency-converting the acceleration data, thereby correcting the fluctuation.

(5) The present invention can also be configured as a program that causes a computer to function as the abnormality diagnosis system/frequency fluctuation correction processing device.

The present invention makes it possible to reduce the effects of fluctuations in the amplitude of acceleration data and improve the accuracy of abnormality diagnosis.

11 is a graph showing a change over time in the degree of deviation indicating fluctuations in the operational data of the diagnosis target. 1A is an image showing a rotating machine to be diagnosed placed on a solid foundation, and FIG. 1B is an image showing the rotating machine placed on an unstable foundation. Graph showing the fluctuation of the frequency analysis results. 1 is a configuration diagram showing a state in which a rotating machine is diagnosed by an abnormality diagnosis system according to an embodiment of the present invention; FIG. 1 is a configuration diagram of an abnormality diagnosis system according to a first embodiment. FIG. 4 is a block diagram of a frequency fluctuation correction processing unit of the same. 4 is a flowchart showing the process of abnormality diagnosis. 1 is a graph showing an Nth-order frequency shift. FIG. 11 is a configuration diagram of a frequency fluctuation correction processing unit according to a second embodiment. FIG. 11 is a configuration diagram of a frequency fluctuation correction processing unit according to a third embodiment.

The following describes an abnormality diagnosis system according to an embodiment of the present invention. This abnormality diagnosis system performs moving average processing during frequency analysis of the object to be diagnosed, thereby correcting frequency fluctuations and improving the accuracy of abnormality diagnosis. Here, a rotating machine (electric motor/generator) will be described as an example of the device to be diagnosed.

In Figure 4, 1 indicates the rotating machine to be diagnosed, which is connected to rotating equipment 4 such as a pump, turbo blower, or water turbine. An acceleration sensor 2 is installed on the bearings of this rotating machine 1, and a data measuring device 5 is also installed to measure the output signal of the acceleration sensor 2 as acceleration data.

In this installation state, the abnormality diagnosis system 6 analyzes the acceleration data measured by the data measuring instrument 5 and diagnoses the operating state of the rotating machine 1. At this time, the acceleration sensor 2, the data measuring instrument 5, and the abnormality diagnosis system 6 are connected by wire/wireless, respectively, and data can be transmitted and received. Below, the abnormality diagnosis system 6 will be specifically described based on Examples 1 to 3.

A first embodiment of the abnormality diagnosis system 6 will be described with reference to Figures 5 to 7. Here, acceleration data of the rotating machine 1 measured by a measuring instrument 5 is input to the abnormality diagnosis system 6.

(1) Example of the Configuration of the Abnormality Diagnosis System The abnormality diagnosis system 6 is configured by a computer and includes the hardware resources of a normal computer (e.g., a CPU, primary storage devices such as RAM and ROM, and secondary storage devices such as an HDD and an SSD).

As a result of the cooperation between these hardware resources and software resources (OS, applications, etc.), the abnormality diagnosis system 6 is equipped with a memory unit 7, a frequency fluctuation correction processing unit 8, a transient period removal unit 9, a learning period selection unit 10, a sequential learning processing unit 11, a diagnosis processing unit 12, and a diagnosis result transmission unit 13, as shown in FIG. 5.

The memory unit 7 is constructed in the primary storage device/secondary storage device of the computer, and stores the acceleration data input to the abnormality diagnosis system 6, as well as the output data of each of the units 8 to 12.

The correction processing unit 8 receives the acceleration data from the memory unit 7 as input, and implements a frequency conversion unit 15 and a moving average unit 16 as shown in FIG. 6. The frequency conversion unit 15 converts the input acceleration data into frequency data, and outputs the converted frequency data to the moving average unit 16.

The frequency data output by the frequency conversion unit 15 is input to the moving average unit 16, which performs moving average processing of the frequency intensity of the input frequency data to perform frequency fluctuation correction.

In this embodiment, the frequency width determined in advance according to the measurement time of the acceleration data is used as the width of the moving average, and moving average processing of the frequency intensity is performed. The fluctuation correction result, i.e., the frequency data after the fluctuation correction, is output to the memory unit 7.

As shown in FIG. 5, the transient period removal unit 9 receives the fluctuation correction result (frequency data after the fluctuation correction) stored in the memory unit 7 as an input, and determines whether the operating state of the rotating machine 1 is in a transient period. If it is determined that the operating state is in a transient period, the frequency data is excluded from the target data (learning data and diagnosis data) for sequential learning and diagnosis. The transient period removal result is output to the memory unit 7.

The learning period selection unit 10 automatically selects the period of sequential learning using the learning data that is determined not to be in a transitional period as input. The learning period data selected here is output to the storage unit 7.

The sequential learning processing unit 11 receives as input the fluctuation correction results and learning period data stored in the memory unit 7. Here, the normal operating state of the rotating machine 1 is sequentially learned based on the frequency data after the fluctuation correction according to the learning period, and a learning result model is generated. The learning result model generated as a result of this sequential learning is output to the memory unit 7.

The diagnostic processing unit 12 receives as input the current acceleration data of the rotating machine 1 and the learning result model stored in the memory unit 7. Here, the learning result model is used to diagnose the state of the rotating machine 1 from the current acceleration data, and the diagnostic result data is output to the memory unit 7.

The diagnostic result transmission unit 13 functions as a communication control unit that receives the diagnostic result data stored in the memory unit 7 and transmits the input diagnostic result data to a monitoring control system or a cloud system.

(2) Processing Contents of the Abnormality Diagnosis System 6 Processing contents (S01 to S09) of the abnormality diagnosis system 6 will be described with reference to Fig. 7. Here, a learning stage (S01 to S07) for sequentially learning the normal state of the rotating machine 1 and a diagnosis stage (S08, S09) for determining whether the current state of the rotating machine 1 is abnormal or not are executed.

S01: Processing begins with the rotating machine 1 operating normally, the output signal of the acceleration sensor 2 is measured as acceleration data by the data measuring device 5, and the measured acceleration data is stored sequentially in the memory unit 7.

S02: The correction processing unit 8 performs frequency fluctuation correction on the acceleration stored in the memory unit 7. Here, the frequency conversion unit 15 converts the input acceleration data into frequency data using a method such as constant Q transformation (see Patent Documents 1 and 2) or Fourier transformation. The moving average unit 16 performs frequency fluctuation correction on the converted frequency data.

In this case, even if a frequency shift occurs due to vibrations caused by an unstable base 3 as shown in FIG. 2(b), the shift rarely exceeds "1 Hz." Therefore, the width of the "1 Hz" range is determined in advance as the width of the moving average, and moving average processing of the frequency intensity is performed.

This type of frequency fluctuation correction using moving average processing is performed as a frequency analysis of the input acceleration data. Since the frequency resolution differs depending on the measurement time of the acceleration data, the width of the moving average is determined taking into account the measurement time.

For example, if the measurement time is "10 seconds", the frequency resolution will be in increments of "0.1 Hz". Therefore, to perform a moving average of the time series frequency data in a range of "1 Hz", the moving average width is set to "0.1 Hz" and 10 samples are divided, and the average value of each sample is graphed to perform the moving average processing.

S03: The transient period removal unit 9 checks whether the corrected frequency data includes data on a transient period during operation of the rotating machine 1. For example, the corrected frequency data is divided, the frequency peak value of each divided section is found, and if there is a difference between the peak values of each frequency that is equal to or greater than a threshold value (1 to 2 Hz), it is determined to be a transient period. If it is determined to be a transient period as a result of the check, it is excluded from the learning data and diagnostic data, and the transient period removal result is output to the memory unit 7, and the stored data in the memory unit 7 is updated.

S04: The learning period selection unit 10 acquires learning data (learning data that has not been determined to be in a transitional period) from the memory unit 7 after the processing of S03, and selects the period during which the sensor output of the acceleration sensor 2 is stable as the learning period.

For example, since the effective value will be higher after a sensor is installed or repaired, the moving average of the effective value of the acceleration data is calculated and the slope of this moving average is obtained. The standard deviation of this slope is calculated, and the point at which the slope falls within the range of "±δ" for one week can be selected as the start of the learning period.

S05: If a learning period is selected in S04, the sequential learning processing unit 11 performs sequential learning for one year to generate a learning result model, and proceeds to S06. For this learning method, the methods of Patent Documents 1 and 2, for example, can be used.

At this time, depending on the CPU performance of the computer constituting the abnormality diagnosis system, for example, one piece of learning data may be learned every day, or one piece of learning data may be learned per week. If a learning period is not selected in S04, the process returns to S01.

S06, S07: As shown in FIG. 1, in order to suppress yearly fluctuations, it is necessary to continue sequential learning for at least one year. Therefore, in S05, the sequential learning processing unit 11 checks whether one year's worth of learning has been completed. If the result of the check is that one year's worth of learning has not been completed, the process proceeds to S07, where sequential learning is resumed and a learning result model is generated. On the other hand, if it has been completed, the process proceeds to S08.

After one year of study is completed, the effects of annual variation will disappear and so sequential study is no longer necessary, but sequential study may be continued.

S08, S09: The diagnostic processing unit 12 diagnoses the state of the rotating machine 1 from the current acceleration data using the learning result model, and stores the diagnosis result in the memory unit 7 (S08). However, the current acceleration data is subjected to processing similar to frequency fluctuation correction (S02) and transient period removal (S03), and the diagnostic processing unit 12 performs the above diagnosis using the data after such processing as diagnostic data.

In this case, when an abnormality occurs in equipment of the facility such as the rotating machine 1, the abnormal state often continues. Therefore, in order to improve the accuracy of diagnosis, it is preferable to diagnose that an abnormality has occurred in the rotating machine 1 when an abnormality is judged to have occurred continuously, rather than just a single diagnosis result. The diagnosis result transmission unit 13 transmits the diagnosis result data of S08 stored in the storage unit 7 to a monitoring control system, a cloud, or the like by wire/wireless (S09).
As described above, according to the abnormality diagnosis system 6 of this embodiment, the acceleration data of the rotating machine 1 is converted into frequency data, and frequency fluctuation correction is performed by performing moving average processing of the frequency intensity for each frequency range of the converted frequency data (S02).

This makes it possible to suppress the influence of factors other than the vibration of the rotating machine 1 when the acceleration data is affected by such factors. For example, even if the rotating machine 1 is installed on the unstable base 3 of FIG. 2(b), the influence of the vibration of the base 3 is suppressed, and accurate learning data and diagnostic data can be obtained.

As a result, a good learning result model can be generated through the transient period removal process in S03 and the learning period selection process in S04, improving the diagnostic accuracy in S08. In addition, accurate diagnostic data can be obtained, so the current operating status of the rotating machine 1 can be accurately grasped, which also improves diagnostic accuracy.

A second embodiment of the abnormality diagnosis system 6 will be described with reference to Figures 8 and 9. Here, in the "frequency fluctuation correction process (S02)" of the first embodiment, the width of the moving average is changed for each frequency band, thereby suppressing the influence of fluctuation more than in the first embodiment.

That is, as shown in FIG. 8, Nth order frequencies appear at doubled frequencies, so the higher the frequency, the greater the frequency deviation. Therefore, in this embodiment, the width of the moving average is changed according to the frequency band, and moving average processing of frequency intensity is performed.

Generally, the frequency range of the acceleration sensor 2 is "5 kHz to 10 kHz," so we will assume this frequency range. In this case, the frequency divisions for changing the width of the moving average can be arbitrary.

However, since (A) "0 to 1 kHz" is usually considered to be a low-frequency region such as displacement and velocity, and (B) "1 kHz" or higher is considered to be a high-frequency region such as acceleration, it is preferable to divide the frequency bands into (A) and (B). It may also be further subdivided into (a) 1 to 3 kHz, (b) 3 to 5 kHz, and (c) 5 kHz or higher.

When the frequency bands are divided into (A) and (B) as described above, the moving average unit 16 performs a low-frequency moving average in the range of "1 Hz" for frequency band (A) and a high-frequency moving average in the range of "10 Hz" for frequency band (B) as shown in FIG. 9. The width of this moving average, "1 Hz, 10 Hz," is one example and may be set arbitrarily.

In this way, according to the abnormality diagnosis system 6 of this embodiment, the width of the moving average is changed depending on the frequency band, and the moving average process of the frequency intensity is performed, so that the influence of the deviation of the harmonics of the rotation frequency in particular is suppressed, and in this respect, the diagnostic accuracy can be further improved.

A third embodiment of the abnormality diagnosis system 6 will be described with reference to FIG. 10. Here, frequency fluctuation correction is performed by changing (varying) the width of the moving average in the moving average unit 16 according to the frequency.

In other words, the rotational frequency is important in diagnosing abnormalities in the rotating machine 1. Since the rotating machine 1 has a rotational frequency that appears every Nth order of the rotational frequency, if the width of the moving average is fixed as in Example 2, it may be difficult to appropriately correct the fluctuations.

For example, if the rotation frequency is "10 Hz", N-th order rotation frequency components such as "20 Hz" and "30 Hz" will appear, and if the rotation frequency is "50 Hz", N-th order rotation frequency components such as "100 Hz" and "150 Hz" will appear.

As a result, if a deviation of 0.1 Hz occurs in the rotation frequency due to the influence of an unstable base 3 as shown in Figure 2(b), the frequency deviation will become larger for each Nth order.

If we assume a frequency shift of up to 1 kHz, then if the rotation frequency is 10 Hz, then up to 1 kHz, 100th order rotation frequency harmonics will be generated. Therefore, up to 1 kHz, a frequency shift of 0.1 (Hz) x 100 (order) = 10 (Hz) will occur.

In addition, when the rotation frequency is "50 Hz", the 20th rotation frequency harmonic occurs by "1 kHz", so a frequency shift of "0.1 (Hz) x 20 (order) = 2 (Hz)" occurs by "1 kHz". Given this, the fluctuation correction can be performed more effectively by changing the width of the moving average according to the rotation frequency.

Therefore, using the rotational frequency of the rotating machine 1 as a reference, as shown in FIG. 10, the width of the moving average is automatically changed according to the frequency band "1 to n" of the frequency data converted by the frequency conversion unit 15 to perform moving average processing of the frequency intensity. Specifically, the moving average processing is performed according to the following steps (S11 to S15: not shown).

S11: Set the rotation frequency of the rotating machine 1. This can be determined from the specifications of the rotating machine 1. For example, the rotation frequency is set to "10 Hz."

S12: Set the measurement time for the acceleration data of the rotating machine 1. For example, set the measurement time to "5 seconds."

S13: Set the frequency deviation of the rotating machine 1. For example, set the frequency deviation to "0.1 Hz."

S14: Calculate the width of the moving average at the current frequency position, i.e., the current position in the frequency data. Current frequency position: X [Hz]
・Rotation frequency: N [Hz]
・Measurement time: T [seconds]
- Frequency deviation: H [Hz]
Then, the width of the moving average [W] can be calculated using equation (1).
Formula (1): W=X[Hz]÷N[Hz]×H[Hz]×T[seconds]
S15: The width of the moving average is calculated by the calculation in S15, and the moving average process of the frequency intensity is performed while changing the width of the moving average.

In this way, according to the abnormality diagnosis system 6 of this embodiment, the width of the moving average corresponding to the current frequency position is automatically calculated based on the rotational frequency of the rotating machine 1, so that moving average processing can be performed that automatically takes into account low frequencies, high frequencies, harmonics, etc., and diagnosis accuracy can be easily improved.

The present invention is not limited to the above embodiment, and can be modified and implemented within the scope of the claims. For example, the correction processing unit 8 may be configured as a frequency fluctuation correction processing device (frequency conversion unit 15, moving average unit 16) in a computer separate from the abnormality diagnosis system 6.

The present invention can also be configured as an abnormality diagnosis program/frequency fluctuation correction processing program that causes a computer to function as the abnormality determination system 6/frequency fluctuation correction device.

According to this abnormality diagnosis program, the computer executes processing steps S01 to S09, S11 to S15, etc. (an abnormality diagnosis method). On the other hand, according to the frequency fluctuation correction processing program, the computer executes processing steps S02, S11 to S15 (a frequency fluctuation correction method).

1...Rotating machine (diagnosis target)
2: Acceleration sensor 3: Base 4: Rotating equipment 5: Data measuring instrument 6: Abnormality diagnosis system 7: Memory unit 8: Frequency fluctuation correction processing unit (frequency fluctuation correction device)
9: Transient period removal section 10: Learning period selection section 11: Sequential learning processing section 12: Diagnosis processing section 13: Diagnosis result transmission section 15: Frequency conversion section 16: Moving average section

Claims (10)

An abnormality diagnosis system that sequentially learns an operating state of a diagnosis target and diagnoses an abnormality,
a correction processing unit that performs frequency fluctuation correction on the acceleration data of the diagnosis object;
a sequential learning processing unit that sequentially learns the result of the fluctuation correction and generates a learning result model;
a diagnosis processing unit that uses the learning result model to determine the presence or absence of the abnormality based on current acceleration data of the diagnosis target;
Equipped with
the correction processing unit includes a moving average unit that performs a moving average process of a frequency based on frequency data obtained by frequency-converting the acceleration data,
The moving average unit sets a frequency width as a moving average width and determines the moving average width for each frequency band of the frequency data.
An abnormality diagnosis system comprising: An abnormality diagnosis system that sequentially learns an operating state of a diagnosis target and diagnoses an abnormality,
a correction processing unit that performs frequency fluctuation correction on the acceleration data of the diagnosis object;
a sequential learning processing unit that sequentially learns the result of the fluctuation correction and generates a learning result model;
a diagnosis processing unit that uses the learning result model to determine the presence or absence of the abnormality based on current acceleration data of the diagnosis target;
Equipped with
the correction processing unit includes a moving average unit that performs a moving average process of a frequency based on frequency data obtained by frequency-converting the acceleration data,
The moving average unit sets a frequency width as a moving average width, and varies the moving average width according to a frequency band of frequency data obtained by frequency-converting the acceleration data.
An abnormality diagnosis system comprising: A device for correcting frequency fluctuations in acceleration data of a diagnosis target when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising: a moving average unit that corrects the fluctuations by performing a moving average process of the frequency based on frequency data obtained by frequency-converting the acceleration data;
The moving average unit sets a frequency width as a moving average width and determines the moving average width for each frequency band of the frequency data.
A correction processing device comprising: A device for correcting frequency fluctuations in acceleration data of a diagnosis target when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising: a moving average unit that corrects the fluctuations by performing a moving average process of the frequency based on frequency data obtained by frequency-converting the acceleration data;
The moving average unit sets a frequency width as a moving average width, and varies the moving average width according to a frequency band of frequency data obtained by frequency-converting the acceleration data.
A correction processing device comprising: A method for diagnosing an abnormality by sequentially learning an operating state of an object to be diagnosed by a computer, comprising the steps of:
a correction processing step of correcting frequency fluctuations in the acceleration data of the diagnosis object;
a sequential learning processing step of sequentially learning the result of the fluctuation correction to generate a learning result model;
a diagnosis processing step of determining the presence or absence of the abnormality based on current acceleration data of a diagnosis target using the learning result model;
having
The correction processing step includes a moving average step of performing a moving average process of a frequency based on frequency data obtained by frequency-converting the acceleration data,
The abnormality diagnosis method according to claim 1, wherein the moving average step defines a frequency width as a moving average width for each frequency band of the frequency data. A method for diagnosing an abnormality by sequentially learning an operating state of an object to be diagnosed by a computer, comprising the steps of:
a correction processing step of correcting frequency fluctuations in the acceleration data of the diagnosis object;
a sequential learning processing step of sequentially learning the result of the fluctuation correction to generate a learning result model;
a diagnosis processing step of determining the presence or absence of the abnormality based on current acceleration data of a diagnosis target using the learning result model;
having
The correction processing step includes a moving average step of performing a moving average process of a frequency based on frequency data obtained by frequency-converting the acceleration data,
The abnormality diagnosis method according to claim 1, wherein the moving average step uses a frequency width as a moving average width, and varies the moving average width according to a frequency band of frequency data obtained by frequency-converting the acceleration data.
A method for correcting frequency fluctuations in acceleration data of a diagnosis target by a computer when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising a moving average step of correcting the fluctuations by performing a moving average process of the frequency based on frequency data obtained by frequency-converting the acceleration data,
The moving average step determines a frequency width as a moving average width for each frequency band of the frequency data.
A correction processing method comprising: A method for correcting frequency fluctuations in acceleration data of a diagnosis target by a computer when diagnosing an abnormality by sequentially learning an operating state of the diagnosis target, comprising a moving average step of correcting the fluctuations by performing a moving average process of the frequency based on frequency data obtained by frequency-converting the acceleration data,
The moving average step uses a frequency width as a moving average width, and varies the moving average width according to a frequency band of frequency data obtained by frequency-converting the acceleration data.
A correction processing method comprising: The abnormality diagnosis system according to claim 1 or 2 comprises:
An abnormality diagnosis program that causes a computer to function. As the correction processing device according to claim 3 or 4 ,
A correction processing program that causes a computer to function.

Images