JP7298414B2 - Abnormality predictive diagnostic system for rotating machine, Abnormality predictive diagnostic method for rotating machine - Google Patents

Description

The present invention relates to a technique for diagnosing the waveform of a current flowing through a rotating machine and detecting signs of abnormality in the structural system of the rotating machine.

Patent Documents 1 and 2 are publicly known as diagnostic methods for detecting a structural abnormality of a rotating machine based on current.

The diagnostic method of Patent Document 1 calculates the difference between the power supply frequency level of the motor and the rotational frequency band wave level of the device to be diagnosed based on the frequency analysis result of the frequency analysis means. The difference value calculated here is collated with the judgment criteria, and the presence or absence of a rotating system abnormality in the equipment to be diagnosed is judged based on the collation result.

The diagnosis method of Patent Document 2 calculates the harmonic content rate of each order of the current harmonics flowing through the electric motor and inverter, and diagnoses the operating state and abnormal deterioration of the electric motor and the like. This calculation uses a specific exponent value divided by the total distortion factor of current harmonics up to a predetermined order, and the contribution rate of the selected principal component of the current harmonics.

JP 2016-90546 JP 2016-118928

The diagnostic methods of Patent Documents 1 and 2 have the following problems.

(1) The diagnostic method of Patent Literature 1 diagnoses a motor (rotating machine) using a difference value between an electric frequency level and a rotation frequency waveband level.

However, various disturbances (noises) are often superimposed on rotating machines that are actually operating in the field, and there is no correlation between the power supply frequency level and the rotation frequency band wave level. There is a risk of erroneous determination (e.g., determination of abnormality for a normal motor, determination of normality for an abnormal motor, etc.).

(2) The principal component contribution rate used in the diagnostic method of Patent Document 2 is obtained by a principal component analysis method for analyzing the relevance of multidimensional events. This principal component analysis method has the problem of taking too much calculation time due to the eigenvalue problem.

Therefore, the diagnostic method of Patent Document 2 requires a computer with a certain degree of high performance. On the other hand, computers with reduced performance (for example, laptop computers) are often used in actual sites, and in this respect, there is a possibility that adaptation to the site may be difficult.

The present invention was made in order to solve such conventional problems, and provides a technology capable of detecting the occurrence of an abnormality sign of a rotating machine with high precision on site without requiring a high-spec computer. and

(1) One aspect of the abnormality sign diagnostic system according to the present invention is a learning unit that acquires current waveform data of the rotating machine in a normal state in advance as learning data, and creates learning parameters based on the acquired learning data;
a diagnostic unit that acquires electric waveform data of the rotating machine to be diagnosed as diagnostic data and diagnoses the acquired diagnostic data based on the learning parameters,
The learning unit acquires the learning parameter as a vector of binary values of spectra at two frequencies separated by a rotation frequency from a power supply frequency related to the operation of the rotating machine in a normal state based on the learning data,
The diagnostic unit obtains a diagnostic parameter that is a vector of binary values of spectra at two frequencies separated by the rotation frequency from the power supply frequency based on the diagnostic data,
calculating the inner product value of the two parameters and the angle formed by the vector of the two parameters as a diagnostic value;
It is characterized in that the occurrence of the sign of abnormality is determined based on the calculated diagnostic value.

(2) Another aspect of the abnormality sign diagnostic system according to the present invention is a learning unit that acquires current waveform data of the rotating machine in a normal state in advance as learning data, and creates learning parameters based on the acquired learning data. ,
a diagnostic unit that acquires electric waveform data of the rotating machine to be diagnosed as diagnostic data and diagnoses the acquired diagnostic data based on the learning parameters,
The learning unit performs frequency band quantization on the learning data, divides the spectrum waveform into small sections, takes an average value or a maximum value for each small section, and obtains quantum waveform data as one point of the quantized waveform. and while using the vector of the acquired quantum waveform data as the learning parameter,
The diagnosis unit performs frequency band quantization on the diagnosis data, divides the spectrum waveform into small sections, and obtains quantum waveform data as one point of the quantized waveform by taking an average value or a maximum value for each small section. and the vector of the acquired quantum waveform data as a diagnostic parameter,
calculating the inner product value of the two parameters and the angle formed by the vector of the two parameters as a diagnostic value;
It is characterized in that the occurrence of the sign of abnormality is determined based on the calculated diagnostic value.

(3) One aspect of the abnormality sign diagnosis method according to the present invention is a learning step of acquiring in advance current waveform data of the rotating machine in a normal state as learning data, and creating a learning parameter based on the acquired learning data;
a diagnostic step of acquiring electric waveform data of the rotating machine to be diagnosed as diagnostic data, and diagnosing the acquired diagnostic data based on the learning parameters;
In the learning step, based on the learning data, the learning parameter is acquired as a vector of binary values of spectra at two frequencies separated by a rotation frequency from a power supply frequency related to the operation of the rotating machine in a normal state,
In the diagnosis step, a diagnostic parameter is acquired as a vector of binary values of spectra at two frequencies separated by the rotational frequency from the power supply frequency based on the diagnostic data;
a step of calculating an inner product value of the two parameters and an angle formed by vectors of the two parameters as diagnostic values;
a step of determining occurrence of the abnormality sign based on the calculated diagnostic value;
It is characterized by having

(4) Another aspect of the abnormality sign diagnosis method according to the present invention is a learning step of acquiring in advance current waveform data of the rotating machine in a normal state as learning data, and creating a learning parameter based on the acquired learning data. ,
a diagnostic step of acquiring electric waveform data of the rotating machine to be diagnosed as diagnostic data, and diagnosing the acquired diagnostic data based on the learning parameters;
The learning step performs frequency band quantization on the learning data, divides the spectrum waveform into small sections, and obtains quantum waveform data as one point of the quantized waveform by taking the average value or maximum value for each small section. and while using the vector of the acquired quantum waveform data as the learning parameter,
In the diagnostic step, the diagnostic data is quantized in a frequency band, divided into small sections of the spectrum waveform, and the average value or the maximum value is taken for each small section to obtain quantum waveform data as one point of the quantized waveform. and a step of using the obtained vector of quantum waveform data as a diagnostic parameter;
a step of calculating an inner product value of the two parameters and an angle formed by vectors of the two parameters as diagnostic values;
a step of determining occurrence of the abnormality sign based on the calculated diagnostic value;
It is characterized by having

ADVANTAGE OF THE INVENTION According to this invention, the abnormality sign of a rotating machine can be detected with high precision in the field, without requiring a high-spec computer.

1 is a configuration diagram of an abnormality sign diagnostic system for a rotating machine according to a first embodiment; FIG. Flow diagram of the same learning stage. Flow chart of the diagnosis stage. FIG. 4 is an image diagram showing division of time-series current waveform data; The graph which shows an example of the structural system diagnosis result of a rotating machine based on time-series current waveform data. FIG. 10 is an explanatory diagram of a trimmed average that excludes outliers in Example 2; The graph which shows the value of the power supply frequency sideband (+ rotation frequency) of a rotating machine. Graph showing a comparison of mean and 25% trimmed mean. FIG. 11 is a configuration diagram showing a waveform data exponentiation unit according to the third embodiment; The graph which shows the comparison of a sine waveform and a 9th power sine waveform. (a) is a graph before waveform quantization in Example 5, and (b) is a graph after waveform quantization.

Hereinafter, an abnormality sign diagnostic system for a rotating machine according to an embodiment of the present invention will be described. This anomaly predictor diagnostic system calculates parameters that occur when a structural system is abnormal from the current waveform flowing through a rotating machine (motor/generator), and determines the presence or absence of an anomaly predictor from the degree of divergence from normal parameters. Embodiments of the abnormality sign diagnosis system will be described below based on Examples 1 to 5. FIG.

The abnormality sign diagnostic system of the first embodiment will be described with reference to FIGS. 1 to 4. FIG. 1 in FIG. 1 indicates the abnormality sign diagnostic system. Here, a current sensor 5 is attached to a power cable 3/ground wire (earth wire) 4 connected to a rotating machine 2 to be diagnosed.

The current waveform of the rotating machine 2 is measured by the current sensor 5 and input to the diagnosis device 6 as current waveform data. That is, the current sensor 5 and the diagnostic device 6 are connected so as to be able to input data, and the diagnostic device 6 determines whether or not an abnormality sign has occurred in the rotating machine 2 based on the input current waveform data.

The diagnostic device 6 is mainly composed of a computer such as a notebook computer, and includes normal computer hardware resources (for example, CPU, main memory such as RAM/ROM, auxiliary memory such as HDD/SSD, etc.).

As a result of cooperation between the hardware resources and software resources (OS, applications, etc.), the diagnostic device 6 learns the current waveform data of the rotating machine 2 in a normal state in advance, as shown in FIGS. A learning unit 7 for creating reference data by using the reference data, and a diagnosis unit 8 for diagnosing the rotating machine 2 using the reference data and determining the occurrence of the sign of abnormality are implemented.

The processing contents of the learning section 7 are executed by a time-series data collection section 10, a waveform data division section 11, a parameter calculation section 12, and a reference data generation section 13, as shown in FIG. On the other hand, the processing contents of the diagnosis unit 8 are executed by a time-series data collection unit 10, a waveform data division unit 11, a parameter calculation unit 12, a diagnosis unit 14, and an abnormality determination unit 15, as shown in FIG. The parts 7 and 8 will be roughly classified and explained below.

<<Processing content of the learning unit 7>>
The processing contents of the learning unit 7 will be described based on FIG. The processing contents of this learning unit 7 are based on the premise that the rotating machine 2 is in a normal state, and by learning the current waveform data in advance, create reference data of a single vector value.

(1) Time-series data collection unit 10
Measured data of the current sensor 5 attached to one phase or three phases of the three-phase power supply of the rotating machine 2 is input to the time-series data collection unit 10 and collected. That is, the sensor value of the current sensor 5 for the rotating machine 2 operating in a normal state is digitally sampled, and time-series current waveform data for a measurement time of 10 seconds at a sampling frequency of 100 KHz, for example, is obtained as learning data.

Although one or more pieces of time-series current waveform data may be obtained here, it is preferable to obtain a plurality of pieces of data (for example, 10 pieces of data). Note that the optimum number of data depends on the operation status of the rotating machine 2 to be diagnosed.

(2) Waveform data division unit 11
The waveform data dividing unit 11 divides the learning data collected by the time-series data collecting unit 10 . The division here is generally equal division (for example, division into eight).

However, as shown in FIG. 4, it is preferable to obtain a plurality of learning data (for example, 1 learning data to 8 learning data) by cutting out the learning data in overlapping sections of N data lengths. This makes it possible to acquire learning data for a plurality of cycles. Note that if sufficient learning data is collected, the processing of the waveform data division unit 11 may be omitted from the viewpoint of maintaining waveform resolution.

(3) Parameter calculator 12
The parameter calculator 12 performs a Fourier transform (using a Hanning window) on each of the current waveform data obtained by the two units 11 and 12, converts it into a frequency spectrum, and obtains the absolute value after the transform.

After that, based on the power supply frequency F _L and the rotation frequency F _r related to the operation of the rotating machine 2, two values of "(power supply frequency F _L )±(rotation frequency F _r )" are obtained, and further the frequency spectrum _X1, Find _X2 . Assuming that the two values of the frequency spectrum obtained here are vector X, it can be expressed as Equation (1).

Since the frequency spectrum is always a positive value at this time, "x ₁ , x ₂ ≧0" is established. This vector X is created for "(the number of learning data)×(the number of divisions of the learning data)". Let this be a learning parameter group.

(4) Reference data generator 13
The reference data generation unit 13 acquires a single reference data (vector X) shown in Equation (2) by calculating the learning parameter group obtained by the parameter calculation unit 12, that is, the average value of a plurality of vectors X. do. The reference data acquired here is stored in the storage device. It should be noted that the information is read out from the storage device during the processing of the diagnostic unit 8 and used to calculate the diagnostic value.

<<Processing content of diagnostic unit 8>>
The diagnostic unit 8 calculates a diagnostic value using the reference data, and determines whether or not there is an anomaly sign of the rotating machine 2 based on the diagnostic value. That is, the current waveform of the rotating machine 2 at the time of diagnosis is measured by the current sensor 5, and is input to the time-series data collection unit 10 and collected as time-series current waveform data in the same manner as at the time of learning. The time-series current waveform data acquired here is used as diagnostic data.

However, since the diagnosis unit 8 executes diagnosis immediately after obtaining diagnostic data, diagnosis starts with data collected for a certain period of time. This data may be treated as one data, or may be treated as multiple data. At this time, the waveform data division unit 11 and the parameter calculation unit 12 basically perform the same processing as the learning unit 7 does. The parameter value of the sideband wave of the power supply frequency of the diagnostic unit 8 obtained as a result is set as the vector X' (diagnostic parameter) shown in Equation (3).

(1) Diagnosis unit 14
The diagnosis unit 14 calculates an inner product value using the vector X′ of Equation (3) and the reference data (vector X) using the formula (4). The inner product value here is obtained by dividing the diagnostic data.

Further, if the magnitudes of the vector X' and the reference data (vector X) are represented by Equation (5), "cos θ" is calculated as shown in Equation (6) using the angle θ formed by the two vectors.

Furthermore, the angle θ can be calculated by calculating the inverse cosine "cos ^-1 (cos θ)" of Equation (7) using the calculated "cos θ".

The inner product value of Equation (5) obtained here and the angle θ indicated by Equation (7) are calculated for the number of divisions of the diagnostic data. By calculating the average value of the calculation results, a representative inner product value and angle can be obtained. The information obtained here (representative inner product value and angle) is used as a diagnostic value.

(2) Abnormality determination unit 15
Using the diagnostic value obtained by the diagnostic unit 14, the state of the rotating machine 2 at the time of diagnosis is grasped from the normal data (reference data) and the difference. At this time, the diagnostic value deviates from the normal data as it increases, indicating that the degree of abnormality increases. Therefore, the sign of abnormality may be detected simply by comparing the diagnostic value with the past data.

However, if a threshold value is set, it is only necessary to determine whether or not the diagnostic value exceeds the threshold value, which facilitates abnormality diagnosis. Here, an example of a threshold determination method will be described.

First, in order to easily determine the threshold value, the degree of abnormality is calculated as a new determination value. This degree of anomaly is calculated from the mean value/deviation value (μ _1, σ ₁ ) of the learning parameter group (vector X group) obtained when learning the learning data, and the diagnostic parameter group (vector X ' group) using the mean value and deviation value (μ _{2 ,} σ ₂ ), and the formula (8).

In equation (8), “μ _1, σ ₁ ” indicates the average and deviation of the learning parameter group (vector X group), and “μ _2, σ ₂ ” indicates the average and deviation of the diagnostic parameter group (vector X′ group). "K" indicates a judgment coefficient, and " _kc , _kd " indicate a caution area coefficient and a danger area coefficient.

Next, the determination coefficient K, the caution zone coefficient k _c , and the dangerous zone coefficient k _d are set according to the operating status of the rotating machine 2 and the degree of importance of the rotating machine 2, for example (K=1, 2, k _c =3, k _d =6) and so on. In addition, if the value of the abnormal value exceeds "20%", it can be set as a caution area, and if it exceeds "70%", it can be set as a danger area. Prompt stop level. In addition, when the result of the calculation of formula (8) exceeds "100%", the degree of abnormality remains "100%".

Since the state level of the rotating machine 2 can be monitored by calculating the degree of abnormality in this way, it is possible to detect the occurrence of an abnormality sign of the rotating machine 2 by monitoring and confirming the state of the rotating machine 2 at the actual site. Become. An example of diagnosis of the structural system of the rotating machine 2 during simulation will be described with reference to FIG.

Here, a rotating body (not shown) of the load is connected to the rotating machine 2, and a weight is attached to the rotating body in stages to simulate a structural abnormality of the rotating machine 2, that is, an unbalanced abnormality. Diagnosis is made based on time-series current waveform data to determine whether there is an anomaly sign.

Specifically, the "normal" state indicates a state in which no imbalance abnormality has occurred, that is, a state in which the rotor and the weight are not attached. "Small Abnormality", "Medium Abnormality", and "Large Abnormality" each indicate a state in which the weight is increased stepwise to increase the imbalance abnormality.

Then, according to the judgment result in FIG. 5, the "normal" state is indicated with an abnormality degree of "0%", and the "abnormality small" and "abnormality medium" states are indicated with an abnormality degree of "50%". The state of "abnormality large" was shown as the degree of abnormality "100%", and it was confirmed that the judgment could be performed correctly.

At this time, according to the abnormality predictive diagnosis system 1, the calculation time is suppressed by adopting the inner product value of the reference data and the diagnosis parameter as the diagnosis value. As a result, high-precision diagnosis can be performed using a computer with limited performance, such as a notebook computer, that is used in the actual field, without the need for a high-spec computer.

In addition, since the waveform data dividing unit 11 divides the learning data and the diagnosis data, the data can be artificially increased even with a small number of measurements. Therefore, it is possible to obtain parameters with more variation than when they are extracted from the learning data/diagnostic data as they are. This point also enables more accurate diagnosis.

The abnormality sign diagnostic system 1 of the second embodiment will be described with reference to FIGS. 6 to 8. FIG. Here, the processing of the reference data generation unit 13 and the diagnosis unit 14 has been modified.

In other words, the time-series current waveform data of the learning data and the diagnostic data are difficult to obtain stable values compared to the generally used time-series vibration waveform data, and outliers (values far from the average value) often take

Therefore, in the present embodiment, the reference data generation unit 13 and the diagnosis unit 14 are caused to perform processing for removing outliers. In Example 1, the mean of the parameters was obtained after calculating the learning parameters and the diagnostic parameters. In this regard, in the second embodiment, the processing is changed to obtain the trimmed average of both parameters.

6, the reference data generator 13 calculates the average value of the learning parameter group, and the diagnosis unit 14 calculates the average value of the diagnostic parameter group. Individual values of this parameter may contain outliers, so simply calculating the average value may be affected by outliers.

Therefore, as shown in FIG. 6, a trimmed average is adopted in which data near unintended minimum and maximum values are removed to obtain an average value. Here, the number of data to be removed is represented by "% (% of the total number of data)".

For example, if "10%" trimming is performed on 100 pieces of data, the average of 80 pieces of data is calculated by removing 10% (10 pieces) from each of the maximum value side and the minimum value side. The ratio of the number of data to be removed here (the ratio of outliers) varies depending on the number of acquisitions of learning data and diagnostic data, the operating status of the rotating machine 2, and the like, but for example, it may be set to "25% trimmed average".

An example will be described with reference to FIGS. 7 and 8. FIG. FIG. 7 shows the transition of the power supply wavenumber sideband (+rotational frequency) of the spectrum waveform of the time-series current waveform data in the normal state of a rotating machine 2, where P indicates an outlier.

FIG. 8 sorts the transitions of FIG. 7 and shows the “mean value” and the “25% trimmed mean value”. high accuracy is also obtained. Therefore, according to this embodiment, the accuracy of the diagnostic value and the degree of abnormality is improved, and in this respect, erroneous detection of the occurrence of an abnormality sign can be suppressed.

The abnormality sign diagnostic system 1 of the third embodiment will be described with reference to FIGS. 9 and 10. FIG. Here, a waveform data exponentiation unit 20 shown in FIG. 9 is added between the time-series data collection unit 10 and the waveform data division unit 11 .

That is, in the inverter-driven rotating machine 2, harmonic voltage waveform distortion exists as noise in the zero-crossing portion (near 0 volts) of the time-series current waveform data, which may affect the current waveform.

Therefore, in this embodiment, the waveform data exponentiation unit 20 is introduced in order to remove the noise caused by the distortion. Here, the waveform data exponentiation unit 20 performs odd power calculations on the collected learning data and diagnosis data.

Fig. 10 shows a raw waveform (sin waveform) and a 9th power waveform (sin waveform). . By applying this processing to the time-series current waveform data of the inverter-driven rotating machine 2, it becomes possible to ignore the noise inherent in the inverter.

Therefore, according to the present embodiment, it is possible to appropriately diagnose the inverter-driven rotating machine 2 and determine an abnormality sign. Note that the processing content of the waveform data exponentiation unit 20 is not limited to odd-number powers, and even-number powers may be multiplied by the sign of the original data.

The abnormality sign diagnostic system 1 of the fourth embodiment will be described. Although the system configuration of the present embodiment is the same as that of the first embodiment, the processing contents of the parameter calculation section 12 of the learning section 7 and the diagnosis section 8 are different.

That is, the parameter calculation unit 12 of the first embodiment executes a Fourier transform to obtain a spectral waveform and calculate learning parameters and the like. has been changed to

The periodogram executed here is a kind of nonparametric estimation method, and is a method of estimating the power spectral density of the fluctuation process. In the first place, when performing signal analysis that handles time-series signals, in many cases stochastic phenomena such as external noise appear on the signals.

In other words, even when spectral analysis is performed using Fourier transform, the probabilistic phenomenon is mixed in, and thus noise is constantly generated at random, which may affect the diagnosis result. Therefore, in this embodiment, instead of the Fourier transform, a method of estimating the power spectral density by a periodogram is executed.

This estimation method obtains the Fourier transform of the time-series current waveform data and appropriately scales the square of the amplitude, which is represented by Equation (9). Here we assume a sampling frequency of 'F _s ' and a signal of length L, 'x _L (n)'.

According to the present embodiment, it is possible to suppress the noise that is constantly generated at random, suppress variations in learning parameters and diagnostic parameters, and obtain stable diagnostic results.

The abnormality sign diagnostic system 1 of the fifth embodiment will be described. This embodiment has the same device configuration as the first embodiment, but differs in the processing of the parameter calculator 12 of the learning unit 7 and the diagnostic unit 8 as in the fourth embodiment.

That is, in the first embodiment, a binary value of "(power supply frequency F _L )±(rotational frequency F _r )" is obtained based on the power supply frequency F _L and the rotation frequency F _r . On the other hand, in the present embodiment, the processing is changed to treat the waveform obtained by quantizing the frequency band as a vector.

In the first place, the rotation speed and rotation frequency of the rotating machine 2 on site fluctuate due to various loads, operating conditions, and the like. Therefore, there is a possibility that the binary values obtained in the first embodiment for the rotating machine 2 may not be obtained correctly. At this time, the rotation speed of the rotating machine 2 should be obtained correctly, but there are cases where it is unknown.

Therefore, in this embodiment, quantization is performed by dividing the spectrum waveform into small intervals in the frequency direction for the frequency band around the power supply frequency. Here, the sub-intervals may overlap, and after that, the average value or the maximum value is taken for each sub-interval to obtain one point of the quantized waveform. Using the obtained quantized waveform data as "vector X _q =(x _qi , x _q2, x _{q3, .} . . )", subsequent learning/diagnosis processing is executed.

FIG. 11 shows an example of quantizing a logarithmic spectrum waveform. Here, the spectrum waveform below "100 Hz" is averaged with 30 small interval data and 10 overlap data, and quantized.

At this time, the range of the small section differs depending on the rotating machine 2, and if the range is determined so as to include fluctuations in the rotation frequency, diagnosis can be made regardless of the fluctuations. Therefore, according to the present embodiment, even for a rotating machine 2 whose number of revolutions is unknown/unmeasurable (for example, a submersible pump), if the quantization interval is set to include the range of possible number of revolutions, the rotation frequency can be reduced. Learning and diagnostic parameters of the included sidebands can be obtained. In this respect, it is also possible to diagnose the rotating machine 2 that accompanies fluctuations in the rotation speed due to the load.

It should be noted that the present invention is not limited to the above-described embodiments, and can be modified and implemented within the scope described in each claim. For example, the configuration of the diagnostic device 6 is not limited to those shown in FIGS. 2, 3, and 6, and other configurations may be used. In addition, the trimmed average of Example 2 can also be used for the average values (μ ₁ ) (μ ₂ ) used for calculating the degree of abnormality.

1... Rotation abnormality sign diagnosis system.

DESCRIPTION OF SYMBOLS 2... Rotating machine 3... Power supply cable 4... Ground wire 5... Current sensor 6... Diagnosis device 7... Learning part 8... Diagnosis part 10... Time series data acquisition part 11... Waveform data classification part 12... Parameter calculation part 13... Reference data Generation unit 14 Diagnosis unit 15 Abnormality determination unit 20 Waveform data exponentiation unit

Claims (9)

A system for diagnosing a waveform of a current flowing in a rotating machine and determining signs of abnormality in the structural system of the rotating machine,
a learning unit that acquires current waveform data of the rotating machine in a normal state in advance as learning data and creates a learning parameter based on the acquired learning data;
a diagnostic unit that acquires electric waveform data of the rotating machine to be diagnosed as diagnostic data and diagnoses the acquired diagnostic data based on the learning parameters,
The learning unit acquires the learning parameter as a vector of binary values of spectra at two frequencies separated by a rotation frequency from a power supply frequency related to the operation of the rotating machine in a normal state based on the learning data,
The diagnostic unit obtains a diagnostic parameter that is a vector of binary values of spectra at two frequencies separated by the rotation frequency from the power supply frequency based on the diagnostic data,
calculating an inner product value of the learning parameter and the diagnostic parameter and an angle between the learning parameter and the diagnostic parameter as diagnostic values;
An abnormality predictor diagnostic system for a rotating machine, wherein occurrence of the abnormality predictor is determined based on the calculated diagnostic value. The learning unit acquires one or more of the learning data, divides the acquired learning data as necessary to acquire a learning parameter group,
While obtaining reference data by calculating the average value of the vector of the learning parameter group,
The diagnostic unit obtains diagnostic parameter groups by dividing the diagnostic data, calculates the inner product value and the angle with the reference data for each of the diagnostic parameters, and
2. The system for diagnosing a sign of abnormality of a rotating machine according to claim 1 , wherein the average value of the calculated inner product value and the calculated angle is used as the diagnosis value.
The diagnosis unit calculates the degree of abnormality based on the average value and deviation value of the learning parameter group and the average value and deviation value of the diagnostic parameter group,
3. The system for diagnosing a sign of abnormality for a rotating machine according to claim 2, wherein the sign of abnormality is determined according to the degree of abnormality. 4. The abnormality sign diagnostic system for a rotating machine according to claim 2, wherein said average value is a trimmed average obtained by removing an arbitrary range from the minimum and maximum values as an outlier. 5. The abnormality sign diagnosis system for a rotating machine according to claim 2, wherein said learning unit and said diagnosis unit exponentiate said learning data and said diagnosis data before said division. The learning unit and the diagnosis unit acquire the learning parameter and the diagnostic parameter after estimating the power spectrum of the current waveform data by a Fourier transform or a periodogram method. Anomaly prediction diagnostic system for rotating machines. A system for diagnosing a waveform of a current flowing in a rotating machine and determining signs of abnormality in the structural system of the rotating machine,
a learning unit that acquires current waveform data of the rotating machine in a normal state in advance as learning data and creates a learning parameter based on the acquired learning data;
a diagnostic unit that acquires electric waveform data of the rotating machine to be diagnosed as diagnostic data and diagnoses the acquired diagnostic data based on the learning parameters,
The learning unit performs frequency band quantization on the learning data, divides the spectrum waveform into small sections, takes an average value or a maximum value for each small section, and obtains quantum waveform data as one point of the quantized waveform. and while using the vector of the acquired quantum waveform data as the learning parameter,
The diagnosis unit performs frequency band quantization on the diagnosis data, divides the spectrum waveform into small sections, and obtains quantum waveform data as one point of the quantized waveform by taking an average value or a maximum value for each small section. and the vector of the acquired quantum waveform data as a diagnostic parameter,
calculating an inner product value of the learning parameter and the diagnostic parameter and an angle between the learning parameter and the diagnostic parameter as diagnostic values;
An abnormality predictor diagnostic system for a rotating machine, wherein occurrence of the abnormality predictor is determined based on the calculated diagnostic value. A method of diagnosing a waveform of a current flowing in a rotating machine by a computer and determining an abnormality sign in a structural system of the rotating machine,
a learning step of acquiring in advance current waveform data of the rotating machine in a normal state as learning data and creating a learning parameter based on the acquired learning data;
a diagnostic step of acquiring electric waveform data of the rotating machine to be diagnosed as diagnostic data, and diagnosing the acquired diagnostic data based on the learning parameters;
In the learning step, based on the learning data, the learning parameter is acquired as a vector of binary values of spectra at two frequencies separated by a rotation frequency from a power supply frequency related to the operation of the rotating machine in a normal state,
In the diagnosis step, a diagnostic parameter is acquired as a vector of binary values of spectra at two frequencies separated by the rotational frequency from the power supply frequency based on the diagnostic data;
calculating an inner product value of the learning parameter and the diagnostic parameter and an angle between the learning parameter and the diagnostic parameter as diagnostic values;
a step of determining occurrence of the abnormality sign based on the calculated diagnostic value;
An abnormality sign diagnosis method for a rotating machine, comprising: A method of diagnosing a waveform of a current flowing in a rotating machine by a computer and determining an abnormality sign in a structural system of the rotating machine,
a learning step of acquiring in advance current waveform data of the rotating machine in a normal state as learning data and creating a learning parameter based on the acquired learning data;
a diagnostic step of acquiring electric waveform data of the rotating machine to be diagnosed as diagnostic data, and diagnosing the acquired diagnostic data based on the learning parameters;
The learning step performs frequency band quantization on the learning data, divides the spectrum waveform into small sections, and obtains quantum waveform data as one point of the quantized waveform by taking the average value or maximum value for each small section. and while using the vector of the acquired quantum waveform data as the learning parameter,
In the diagnostic step, the diagnostic data is quantized in a frequency band, divided into small sections of the spectrum waveform, and the average value or the maximum value is taken for each small section to obtain quantum waveform data as one point of the quantized waveform. and a step of using the obtained vector of quantum waveform data as a diagnostic parameter;
calculating an inner product value of the learning parameter and the diagnostic parameter and an angle between the learning parameter and the diagnostic parameter as diagnostic values;
a step of determining occurrence of the abnormality sign based on the calculated diagnostic value;
An abnormality sign diagnosis method for a rotating machine, comprising:

Images