JP7618971B2 - Abnormality diagnosis device and abnormality diagnosis system - Google Patents

Description

The present invention relates to an abnormality diagnosis device and an abnormality diagnosis system.

A method is known in which abnormality diagnosis is performed based on data obtained by performing a fast Fourier series transform on vibration data output from a vibration sensor that detects vibrations in an electric motor, etc.

For example, in the method (Method 1) described in Patent Document 1, a diagnostic model is generated by applying multivariate statistical process control (MSPC) to statistics (e.g., average, maximum, etc.) of the Fourier spectrum for vibration data under normal conditions. Then, in Method 1, this diagnostic model is used to perform an abnormality diagnosis on the vibration data to be diagnosed.

In addition, in the method (Method 2) described in Patent Document 2, an abnormality diagnosis is performed on the vibration data of the diagnosis target based on the correlation coefficient between the Fourier spectrum of the vibration data under normal conditions and the Fourier spectrum of the vibration data of the diagnosis target.

JP 2004-279056 A JP 2018-091033 A

However, in method 1, even if the shape of the Fourier spectrum for the vibration data to be diagnosed is significantly different from that of normal data due to an abnormality in the motor, if the statistics such as the average and maximum values are within the normal range, the data may be diagnosed as normal even though it is abnormal. Note that the frequency with which such a misdiagnosis occurs in method 1 does not depend on the number of samples of normal data.

In addition, method 1 requires more samples of normal data than method 2, and if the number of normal data samples is insufficient, the discrimination performance deteriorates significantly.

In addition, method 2 can perform a diagnosis that takes into account the shape of the Fourier spectrum for the vibration data of the diagnosis target, thereby suppressing the above-mentioned misdiagnosis. However, method 2 generally has inferior discrimination performance compared to method 1.

The objective of the present invention is to provide an anomaly diagnosis device that can suppress misdiagnosis that occurs regardless of the number of normal data samples, and can suppress deterioration of discrimination performance due to an insufficient number of normal data samples.

One invention for achieving the above object is an abnormality diagnosis device including a first processing unit that calculates a first index indicating the degree of similarity between diagnostic data indicating the frequency spectrum of vibrations of a target device and a first model constructed based on normal data indicating the frequency spectrum of vibrations of the target device when the target device is normal, a second processing unit that calculates a second index indicating the degree of abnormality of the target device based on the normal data and abnormal data indicating the frequency spectrum of vibrations of the target device when the target device is abnormal, and the diagnostic data, and a diagnosis unit that diagnoses whether the target device is abnormal based on the first index and the second index. Other features of the present invention will be made clear by the description in this specification.

The present invention provides an anomaly diagnosis device that can suppress misdiagnosis that occurs regardless of the number of normal data samples, and can suppress deterioration of discrimination performance due to an insufficient number of normal data samples.

FIG. 1 is a diagram showing an overview of an abnormality diagnosis system. FIG. 2 is a diagram illustrating an example of a hardware configuration of an abnormality diagnosis device. FIG. 2 is a diagram illustrating an example of a functional block realized in the abnormality diagnosis device. FIG. 11 is a diagram illustrating an example of diagnostic data. FIG. 11 is a diagram illustrating an example of processed diagnostic data. FIG. 11 is a diagram showing an example of diagnostic data and normal data. FIG. 13 is a diagram illustrating an example of a calculation result of a correlation coefficient. FIG. 13 is a diagram showing an example of a calculation result of Q statistics. 4 is a flowchart showing an example of a process executed by the abnormality diagnosis device. FIG. 2 is a diagram illustrating an example of a functional block realized in the abnormality diagnosis device. 4 is a flowchart showing an example of a process executed by the abnormality diagnosis device.

First Embodiment
<Abnormality diagnosis system>
1 is a diagram showing an overview of an abnormality diagnosis system 1 according to the present embodiment. The abnormality diagnosis system 1 is a system for detecting abnormalities, such as a bearing abnormality, of an electric motor 62 driven by an inverter 61 in a production facility such as a rolling facility, based on information from a vibration sensor 2 that detects vibrations of the electric motor 62. The abnormality diagnosis system 1 includes the vibration sensor 2 and an abnormality diagnosis device 3. Each of these will be described below.

[Vibration sensor]
The vibration sensor 2 is provided in a drive device 6 for driving a load. The drive device 6 has a power supply facility 60, an inverter 61, and an electric motor 62. The inverter 61 has a switching circuit that operates at a predetermined switching frequency, and drives the electric motor 62. In this embodiment, the electric motor 62 corresponds to a "target device (or target equipment)" that is the target of an abnormality diagnosis.

The vibration sensor 2 detects the vibration of the electric motor 62. The vibration sensor 2 may be of a contact type or a non-contact type. The vibration sensor 2 outputs time series data of the detected vibration as vibration data to an acquisition unit 40 (described later) of the abnormality diagnosis device 3.

[Abnormality diagnosis device]
2 is a diagram showing an example of a hardware configuration of the abnormality diagnosis device 3. The abnormality diagnosis device 3 is a computer including a CPU (Central Processing Unit) 30, a memory 31, a storage device 32, an input device 33, a display device 34, and a communication device 35.

The CPU 30 executes programs stored in the memory 31 and the storage device 32 to realize various functions of the abnormality diagnosis device 3.

Memory 31 is, for example, a RAM (Random Access Memory) and is used as a temporary storage area for programs, data, etc.

The storage device 32 is a non-volatile storage device 32 that stores the first model M1, the second model M2, the third model M3, and the abnormality diagnosis program P executed or processed by the CPU 30, which will be described later.

The abnormality diagnosis program P is a program for implementing various functions possessed by the abnormality diagnosis device 3, and includes, for example, an OS (Operating System), etc.

The input device 33 is a device that accepts commands and data input by the user, and includes input interfaces such as a keyboard and a touch sensor that detects the touch position on a touch panel display.

The display device 34 is, for example, a display device. The communication device 35 exchanges data and various programs with the vibration sensor 2 and other computers via a network.

Figure 3 is a diagram showing an example of a functional block realized in the abnormality diagnosis device 3. When the CPU 30 of the abnormality diagnosis device 3 executes the abnormality diagnosis program P, the abnormality diagnosis device 3 realizes an acquisition unit 40, a diagnostic data generation unit 41, a first processing unit 42, a second processing unit 43, a third processing unit 44, and a diagnosis unit 45.

The acquisition unit 40 acquires the vibration data output by the vibration sensor 2. The acquisition unit 40 may constantly acquire the vibration data output while the drive device 6 is in operation, or may acquire the vibration data output at a fixed time at a specified cycle. When the acquisition unit 40 acquires the vibration data, it outputs the vibration data to the diagnostic data generation unit 41.

The diagnostic data generating unit 41 generates diagnostic data based on the vibration data. The diagnostic data is data that indicates the frequency spectrum of the vibration of the electric motor 62. The diagnostic data will be described below with reference to FIG. 4. FIG. 4 is a diagram illustrating an example of diagnostic data.

The diagnostic data generating unit 41 generates diagnostic data a _in by performing a Fourier transform shown in the following (Equation 1) on the vibration data x(t) in a range from time _ti to a predetermined time interval T [sec].

In the diagnostic data a _in , the subscript i indicates that the diagnostic data is based on the vibration data x(t) from time _ti to time _ti +T. The subscript n indicates that the diagnostic data corresponds to a frequency component of n/T [Hz], where n is an integer from 0 to a predetermined upper limit n _max .

In this embodiment, the time interval T is set to 50 μsec, and n _max is set to 200. Then, the diagnostic data a _in (n=0 to 200) is data extracted at intervals of 50 Hz from a frequency spectrum in the range of 0 to 10 kHz.

The diagnostic data generation unit 41 performs the above-mentioned Fourier transform processing using a fast Fourier transform algorithm.

The diagnostic data generating unit 41 of this embodiment processes the diagnostic data shown in FIG. 4 obtained by Fourier transforming the vibration data, and generates the processed diagnostic data. This will be described in detail below with reference to FIG. 4 and FIG. 5.

The diagnostic data shown in FIG. 4 is unprocessed diagnostic data, and the unprocessed diagnostic data has sharp peaks near the frequencies of 4 kHz, 6 kHz, 8 kHz, and 10 kHz. These peaks are noise caused by the switching frequency of the switching circuit of the inverter 61. In other words, such noise is not caused by an abnormality in the electric motor 62.

Therefore, the diagnostic data is processed so that the contribution of noise caused by the switching frequency of the switching circuit is not reflected in the abnormality diagnosis process described below. Specifically, the diagnostic data in this embodiment is data that indicates the frequency spectrum of the vibration of the electric motor 62, and in which the frequency spectrum of the frequency band corresponding to the switching frequency of the switching circuit of the inverter 61 is suppressed.

Fig. 5 is a diagram for explaining an example of the processed diagnostic data. In the processed diagnostic data shown in Fig. 5, the frequency spectrum corresponding to the frequency band of 400 Hz around each of the frequencies of 4 kHz, 6 kHz, 8 kHz, and 10 kHz is set to 0, compared to the unprocessed diagnostic data shown in Fig. 4. That is, in the processed diagnostic data a _in (n=0 to 200), the components of n=76 to 85, 116 to 124, 156 to 164, and 196 to 200 are set to 0. Hereinafter, the processed diagnostic data will be simply referred to as diagnostic data.

The first processing unit 42 calculates a first index indicating the degree of similarity between the diagnostic data and the first model M1. The first model M1 is a model constructed based on normal data. The normal data is data indicating the frequency spectrum of the vibration of the electric motor 62 when the electric motor 626 is normal. This will be explained in detail below.

First, we will explain normal data. Figure 6 is a diagram showing an example of diagnostic data and normal data. In Figure 6, an example of diagnostic data is shown by a solid line, and an example of normal data is shown by a dashed line.

In the following description, normal data is expressed as b _jn (j=1 to j _max , n=0 to n _max ). Here, the subscript j is a subscript for distinguishing between multiple normal data, each of which is based on vibration data output at a different time. The subscript n indicates that the frequency component has an amplitude of n/T [Hz], similar to the diagnostic data. The normal data b _jn is data obtained by applying a Fourier transform to the vibration data output by the vibration sensor 2 when the electric motor 62 is normal, in the same manner as in the above (Equation 1).

Next, the first model M1 will be described. The first model M1 in this embodiment is the average value of multiple normal data b _jn , and is expressed as M1 _n (n=0 to n _max ). Here, the subscript n indicates that the frequency component has an amplitude of n/T [Hz], similar to the diagnostic data. In other words, the first model M1 _n is the value obtained by dividing the sum of multiple normal data b _jn for j by j _max .

（式２）において、

と定義される。 Next, the first index indicating the similarity will be described. In this embodiment, the first processing unit 42 calculates a correlation coefficient r _i indicating the similarity between the diagnostic data a _jn (n=0 to n _max ) and the first model M1 _n (n=0 to n _max ) as the first index. The correlation coefficient r _i is defined by the following (Equation 2).

In formula 2,

It is defined as:

The correlation coefficient is a value between 0 and 1, and the closer to 1 the correlation coefficient is, the higher the similarity between the diagnostic data and the first model M1. In other words, the closer the correlation coefficient r is to 1, the more likely it is that the motor 62 is operating normally at the time when the disturbance data on which the diagnostic data is based is output. In particular, if the correlation coefficient is about 0.6 or more, it can be considered that the motor is operating normally. Furthermore, when the correlation coefficient is 1, the diagnostic data a _jn (n=0 to n _max ) and the first model M1 _n (n=0 to n _max ) match.

The first processing unit 42 calculates correlation coefficients for the diagnostic data sequentially output from the diagnostic data generating unit 41 according to the above procedure. Fig. 7 is a diagram showing an example of the calculation result of the correlation coefficient. Fig. 7 shows the correlation coefficients calculated for each of the diagnostic data sequentially output with sample numbers 1 to 17. The dashed line in the figure indicates the position where the correlation coefficient is 0.6. The sample number here corresponds to the subscript i of the diagnostic data a _in .

In the above, the correlation coefficient has been exemplified as the first index indicating the similarity in this embodiment, but the present invention is not limited to this. As another example, the inner product (cosine similarity) may be used. In this case, the inner product n _i of the diagnosis data a _jn (n=0 to n _max ) and the first model M1 _n (n=0 to n _max ) is defined by the following (Equation 5).

The second processing unit 43 calculates a second index indicating the degree of abnormality of the electric motor 62 based on the second model M2 and the diagnostic data. The second model M2 is a model constructed based on normal data and abnormal data. The abnormal data is data indicating the frequency spectrum of the vibration of the electric motor 62 when the electric motor 62 is abnormal.

In this embodiment, the second processing unit 43 calculates the second index using the second model M2 constructed by multivariate statistical process management. In this embodiment, the Q statistic is used as the second index. The smaller the Q statistic, the lower the degree of abnormality of the motor 62, and the larger the Q statistic, the higher the degree of abnormality of the motor 62. In particular, if the Q statistic is about 1.0×10 ⁴ or less, the motor 62 can be considered normal.

The second processing unit 43 calculates Q statistics for the diagnostic data sequentially output from the diagnostic data generating unit 41. Fig. 8 is a diagram showing an example of the calculation result of the Q statistics. Fig. 8 shows the Q statistics calculated for each of the sequentially output diagnostic data with sample numbers 1 to 17. The dashed line in the figure indicates the position where the Q statistics is 1.0 x ^104. The sample number here corresponds to the subscript i of the diagnostic data a _in .

In the present embodiment, the second model M2 is constructed using multivariate statistical process control, but this is not limiting. As another example, the second model M2 may be constructed using a machine learning model such as a support vector machine.

The third processing unit 44 calculates a third index indicating the degree of similarity between the statistics of the diagnostic data and the statistics of the third model M3. The third model M3 is a model constructed based on normal data. As with the first model M1 described above, the third model M3 is constructed by calculating the average value of multiple normal data. The third model M3 may be the same model as the first model M1. Alternatively, the third model M3 may be constructed using a set of normal data different from the set of normal data used when constructing the first model M1.

The statistics here include at least one of the average value, the maximum value, and the standard deviation. Here, the average value, the maximum value, and the standard deviation of the diagnostic data are the respective values calculated for the diagnostic data a _in using the subscript n as the sample number. The same applies to the average value, the maximum value, and the standard deviation of the third model M3.

In this embodiment, the average value is used as the statistic. The average value of the diagnostic data a _in (n=1 to n _max ) is A, and the average value of the third model M3 _n (n=1 to n _max ) is a. The ratio of A to a (A/a) is the third index. The closer the third index is to 1, the higher the similarity, and the farther it is from 1, the lower the similarity.

In particular, when the average value is used as a statistical quantity as in this embodiment, if the third index (A/a) is smaller than 0.5 or larger than 2, there is a high possibility that a bearing abnormality has occurred in the electric motor 62.

The diagnosis unit 45 diagnoses whether or not the electric motor 62 is abnormal based on the first to third indices. For example, the diagnosis unit 45 may diagnose whether or not the electric motor 62 is abnormal based on the results obtained by weighting the normalized values of each of the first index, the second index, and the third index.

Specifically, the first index (correlation coefficient), the second index (Q statistic), and the third index (similarity of statistics) are normalized to the range of 0 to 1 and are designated as α, β, and γ, respectively.

In this normalization, the correlation coefficient r _i calculated by the first processing unit 42 may be used as α. The Q statistics is adjusted so that β is closer to 1 as the degree of abnormality is lower (the Q statistics is smaller), and β is adjusted so that β is closer to 0 as the degree of abnormality is higher (the Q statistics is larger). The similarity of the statistics is adjusted so that γ is closer to 1 as the similarity of the statistics is closer to 1, and γ is closer to 0 as the similarity of the statistics is farther from 1.

Then, the weights for α, β, and γ are _wα , _wβ, and _wγ , respectively, and based on the weighted amount W (= _wαα + _wββ + _wγγ ), it is diagnosed whether or not the motor 62 is abnormal.

Here, if the weights of _wα , _wβ, and _wγ are determined so that the sum of these is 1, the weighted amount W will be equal to or greater than 0 and equal to or less than 1. Therefore, a threshold value, for example, 0.5, may be set in advance, and the diagnosing unit 45 may diagnose the weighted amount W as normal when it is greater than the threshold value, and diagnose the weighted amount W as abnormal in other cases.

A method for determining the weights w _α , w _β, and w _γ can be, for example, determined according to the number of normal data used in constructing the first to third models (M1 to M3). This will be described in detail below.

When constructing the first to third models (M1 to M3), the amount of normal data required varies depending on each model. Here, the amount of normal data required is the amount of normal data required to obtain calculation results of reliable indices (first to third indices). Here, the calculation results of reliable indices are calculation results of indices that are reproduced with a certain level of uniqueness or higher.

For example, the amount of normal data required to construct the second model M2 is generally greater than the amount of normal data required to construct the first model M1 or the third model M3.

In addition, the reliability of the Q statistic calculated using the number of normal data required to construct the second model M2 is generally higher than the reliability of the correlation coefficient calculated using the same number of normal data.

Therefore, when the number of normal data used in constructing the second model M2 is equal to or greater than the number required for constructing the second model M2, it is preferable to weight the normalized value β of the Q statistic the most. As a specific example, weights such as ( _wα , _wβ , _wγ )=(0.25, 0.50, 0.25) may be used.

On the other hand, if the number of normal data used is less than the number required to construct the second model M2, but is equal to or greater than the number required to construct the first model M1, the reliability of the calculated Q statistic is lower than the reliability of the calculated correlation coefficient.

Therefore, in such a case, it is preferable to give the smallest weight to the normalized value β of the Q statistic. As a specific example, weights such as ( _wα , _wβ , _wγ )=(0.40, 0.20, 0.40) may be used.

<Abnormality Diagnosis Processing>
An example of the process executed by each functional block will be described below with reference to Fig. 9 etc. Fig. 9 is a flow chart showing an example of the process executed by the abnormality diagnostic device 3.

First, the acquisition unit 40 acquires the vibration data output by the vibration sensor 2 (S10). Next, the diagnostic data generation unit 41 generates diagnostic data _{a_in} based on the vibration data (S11).

Next, the first processing unit 42 calculates a first index indicating the similarity between the diagnostic data and the first model M1 (S12). The first model M1 is constructed in advance based on multiple normal data. In addition, the first index indicating the similarity in this embodiment is a correlation coefficient.

Next, the second processing unit 43 calculates a second index indicating the degree of abnormality of the motor 62 based on the second model M2 and the diagnostic data (S13). The second model M2 is constructed in advance by applying multiple normal data and multiple abnormal data to multivariate statistical process control. In addition, the second index in this embodiment is a Q statistic.

Then, the third processing unit 44 calculates a third index indicating the similarity between the statistics of the diagnostic data and the statistics of the third model M3 (S14). The third model M3 is constructed in advance based on a plurality of normal data. In addition, the third index indicating the similarity in this embodiment is the ratio (A/a) of the average value A of the diagnostic data to the average value a of the third model M3.

Then, the diagnosis unit 45 diagnoses whether or not the electric motor 62 is abnormal based on the first to third indexes, and outputs the diagnosis result (S15). The diagnosis unit 45 diagnoses whether or not the electric motor 62 is abnormal based on the result obtained by weighting the normalized values of each of the first index, the second index, and the third index, for example.

The order in which S12, S13, and S14 are executed is arbitrary and is not limited to the above example. In addition, S12, S13, and S14 may be executed in parallel.

Second Embodiment
10 is a diagram showing an example of functional blocks of this embodiment realized in the abnormality diagnosis device 3. The functions of this embodiment realized in the abnormality diagnosis device 3 are different from the functions of the first embodiment realized in the abnormality diagnosis device 3 in a first processing unit 52, a second processing unit 53, a third processing unit 54, and a diagnosis unit 55. The acquisition unit 50 and the diagnosis data generation unit 51 are similar to the acquisition unit 40 and the diagnosis data generation unit 41 of the first embodiment, and therefore a description thereof will be omitted.

The first processing unit 52 outputs a first diagnosis result indicating whether the electric motor 62 is abnormal or not based on the first index. Specifically, the first processing unit 52 outputs the first diagnosis result by comparing the first index with a predetermined first threshold. In this embodiment, the first processing unit 52 diagnoses that the electric motor 62 is normal when the correlation coefficient is greater than the first threshold, and diagnoses that the electric motor 62 is abnormal otherwise. For example, 0.6 can be used as the first threshold.

In addition, in this embodiment, the first diagnostic result is set to 1 if it is normal, and 0 if it is above normal. This also applies to the second and third diagnostic results described below.

The second processing unit 53 outputs a second diagnosis result indicating whether or not the motor 62 is abnormal based on the second index. Specifically, the second processing unit 53 outputs the second diagnosis result by comparing the second index with a predetermined second threshold. In this embodiment, the second processing unit 53 diagnoses the motor 62 as normal when the Q statistics is smaller than the predetermined second threshold, and diagnoses the motor 62 as abnormal in other cases. For example, 1.0× ¹⁰⁴ can be used as the second threshold.

The third processing unit 54 outputs a third diagnostic result indicating whether the motor 62 is abnormal based on the third index. Specifically, the third processing unit 54 outputs the third diagnostic result by comparing the third index with a predetermined third threshold. In this embodiment, the third processing unit 54 diagnoses the motor 62 as normal when the ratio (A/a) between the average value A of the diagnostic data described in the first embodiment and the average value a of the third model M3 is within a predetermined range, and diagnoses the motor 62 as abnormal otherwise. The predetermined range can be, for example, 0.5 to 2.0.

The diagnostic unit 55 outputs a fourth diagnostic result indicating whether or not the motor 62 is abnormal based on the first to third diagnostic results. The diagnostic unit 55 may output the fourth diagnostic result based on the sum of the first diagnostic result, the second diagnostic result, and the third diagnostic result. As described above, the first to third diagnostic results are 0 (abnormal) or 1 (normal). Therefore, the sum of the first to third diagnostic results is a value between 0 and 3. The diagnostic unit 55 may diagnose the motor 62 as normal if this sum is, for example, 2 or 3.

Alternatively, the diagnostic unit 55 may output a fourth diagnostic result based on the results obtained by weighting each of the first diagnostic result, the second diagnostic result, and the third diagnostic result according to the number of normal data.

As a specific example, assume that the first, second, and third diagnostic results are 0 (abnormal), 1 (normal), and 1 (normal), respectively. The weights for the first, second, and third diagnostic results are _wα , _wβ , and _wγ , respectively. In this case, the weighted amount W is W=0× _wα +1× _wβ +1× _wγ = _wβ + _wγ . Then, based on this weighted amount W, it is diagnosed whether or not the electric motor 62 is abnormal.

Here, as a method for determining the weights w _α , w _β, and w _γ and a method for performing an abnormality diagnosis, the method described in the first embodiment can be used.

<Abnormality Diagnosis Processing>
An example of the process executed by each functional block will be described below with reference to Fig. 11 etc. Fig. 11 is a flowchart showing an example of the process executed by the abnormality check device 3.

In the flowchart shown in FIG. 11, steps S22, S24, and S26 are similar to steps S12, S13, and S14, respectively, described in the first embodiment using FIG. 9, and therefore will not be described in detail.

After calculating the first index in S22, the first processing unit 52 outputs a first diagnosis result indicating whether or not the electric motor 62 is abnormal based on the first index (S23).

Next, the second processing unit 53 calculates the second index in S24, and then outputs a second diagnosis result indicating whether or not the electric motor 62 is abnormal based on the second index (S25).

Next, the third processing unit 54 calculates the third index in S26, and then outputs a third diagnosis result indicating whether or not the electric motor 62 is abnormal based on the third index (S27).

Next, the diagnostic unit 55 outputs a fourth diagnostic result indicating whether or not the electric motor 62 is abnormal based on the first to third diagnostic results (S28). The diagnostic unit 55 may output the fourth diagnostic result based on the results obtained by weighting each of the first diagnostic result, the second diagnostic result, and the third diagnostic result according to the number of normal data.

Note that S23, S25, and S27 may be executed after S22, S24, and S26, respectively, and the order of S22 to S26 is not limited to that in this example.

====Summary====
As described above, the abnormality diagnosis device 3 of the first and second embodiments includes a first processing unit 42 that calculates a first index indicating the similarity between diagnostic data indicating the frequency spectrum of vibrations of the target device and a first model constructed based on normal data indicating the frequency spectrum of vibrations of the target device when the target device is normal, a second processing unit 43 that calculates a second index indicating the degree of abnormality of the target device based on the normal data and abnormal data indicating the frequency spectrum of vibrations of the target device when the target device is abnormal, and the diagnostic data, and a diagnosis unit 45 that diagnoses whether the target device is abnormal based on the first index and the second index.

This makes it possible to suppress misdiagnosis that occurs regardless of the number of normal data samples, and to suppress deterioration of discrimination performance due to an insufficient number of normal data samples.

In addition, in the above-mentioned abnormality diagnosis device 3, the first processing unit 42 calculates a correlation coefficient indicating the similarity between the diagnosis data and the first model as a first index. By taking the correlation coefficient into consideration, it is possible to obtain reliable diagnosis results even when the number of normal data used is insufficient.

In addition, in the above-mentioned anomaly diagnosis device 3, the second processing unit 43 calculates the second index using a second model constructed by multivariate statistical process management. By using multivariate statistical process management, the reliability of the diagnosis result improves as the number of normal data used increases.

The abnormality diagnosis device 3 further includes a third processing unit 44 that calculates a third index indicating the degree of similarity between the statistics of the diagnosis data and the statistics of a third model constructed based on normal data, and a diagnosis unit 45 diagnoses whether the target device is abnormal or not based on the first to third indices. This allows for the adoption of an intuitive and easy-to-understand diagnosis method.

In addition, in the above-mentioned anomaly diagnosis device, the statistics include at least one of the average value, maximum value, and standard deviation. This allows for the adoption of a more intuitive and easy-to-understand diagnosis method.

In addition, in the abnormality diagnosis device 3 of the first embodiment, the diagnosis unit 45 diagnoses whether or not the target device is abnormal based on the results obtained by weighting the normalized values of each of the first index, the second index, and the third index according to the number of normal data. As a result, when constructing any of the models, it is possible to obtain a reliable diagnosis result overall by reducing the contribution of reliable indexes due to a lack of the required number of normal data.

In addition, in the abnormality diagnosis device 3 of the second embodiment, the first processing unit 42 outputs a first diagnostic result indicating whether or not the target device is abnormal based on the first index, the second processing unit 43 outputs a second diagnostic result indicating whether or not the target device is abnormal based on the second index, the third processing unit 44 outputs a third diagnostic result indicating whether or not the target device is abnormal based on the third index, and the diagnosis unit 45 outputs a fourth diagnostic result indicating whether or not the target device is abnormal based on the first to third diagnostic results. Since the diagnosis results by each diagnosis method can be confirmed, if the fourth diagnostic result is abnormal, it becomes easier to identify the cause of the diagnosis of abnormality.

In addition, in the abnormality diagnosis device of the second embodiment, the diagnosis unit 45 outputs a fourth diagnosis result based on the results obtained by weighting each of the first diagnosis result, the second diagnosis result, and the third diagnosis result according to the number of normal data. As a result, when constructing any model, it is possible to obtain a reliable diagnosis result overall by reducing the contribution of reliable indicators due to a lack of the required number of normal data.

In addition, in the above-mentioned abnormality diagnosis device 3, the target device has a switching circuit that operates at a predetermined switching frequency, and the diagnosis data is data that indicates the frequency spectrum of the vibration of the target device, in which the frequency spectrum of the frequency band corresponding to the predetermined switching frequency is suppressed. As a result, even if switching noise is included in the vibration data, it is possible to suppress erroneous diagnosis caused by switching noise.

The system also includes a processing unit that calculates an index indicating the degree of similarity between diagnostic data indicating the frequency spectrum of the vibration of the target device and a model constructed based on normal data indicating the frequency spectrum of the vibration of the target device when the target device is normal, and a diagnostic unit that diagnoses whether the target device is abnormal or not based on the index. This makes it possible to suppress deterioration of discrimination performance due to an insufficient number of normal data samples.

The system also includes a processing unit that calculates an index indicating the degree of abnormality of the target device based on a model constructed based on normal data indicating the frequency spectrum of the vibration of the target device when the target device is normal and abnormal data indicating the frequency spectrum of the vibration of the target device when the target device is abnormal, and diagnostic data indicating the frequency spectrum of the vibration of the target device, and a diagnostic unit that diagnoses whether the target device is abnormal based on the index. This allows the shape of the frequency spectrum to be taken into consideration when diagnosing an abnormality, thereby suppressing erroneous diagnosis that occurs regardless of the number of samples of normal data.

The abnormality diagnosis system 1 of the first and second embodiments further includes an abnormality diagnosis device 3 of the first embodiment or an abnormality diagnosis device 3 of the second embodiment, which further includes an acquisition unit 40 that acquires vibration data output by a vibration sensor 2 provided in the target device, and a diagnostic data generation unit 41 that generates diagnostic data based on the vibration data, and a vibration sensor 2 provided in the target device. This makes it possible to suppress erroneous diagnosis that occurs regardless of the number of normal data samples, and to suppress deterioration of discrimination performance due to an insufficient number of normal data samples.

The above embodiments are intended to facilitate understanding of the present invention, and are not intended to limit the scope of the present invention. Furthermore, the present invention may be modified or improved without departing from the spirit of the present invention, and it goes without saying that the present invention includes equivalents.

1: Abnormality diagnosis system 2: Vibration sensor 3: Abnormality diagnosis device 30: CPU
31: Memory 32: Storage device 33: Input device 34: Display device 35: Communication device 40: Acquisition unit 41: Diagnostic data generation unit 42: First processing unit 43: Second processing unit 44: Third processing unit 45: Diagnosis unit 50: Acquisition unit 51: Diagnostic data generation unit 52: First processing unit 53: Second processing unit 54: Third processing unit 55: Diagnosis unit 6: Drive unit 60: Power supply equipment 61: Inverter 62: Electric motor

Claims (8)

a first processing unit that calculates a first index indicating a similarity between diagnostic data indicating a frequency spectrum of a vibration of a target device and a first model constructed based on normal data indicating a frequency spectrum of a vibration of the target device when the target device is normal;
a second processing unit that calculates a second index indicating an abnormality degree of the target device based on the normal data and abnormality data indicating a frequency spectrum of vibration of the target device when the target device is abnormal, and the diagnosis data;
a third processing unit that calculates a third index indicating a similarity between statistics of the diagnostic data and statistics of a third model constructed based on the normal data;
a diagnosis unit that diagnoses whether or not the target device is abnormal based on a result obtained by weighting the normalized values of each of the first index, the second index, and the third index in accordance with the number of normal data;
An abnormality diagnosis device comprising: The first processing unit is
The abnormality diagnosis device according to claim 1 , wherein a correlation coefficient indicating a similarity between the diagnosis data and the first model is calculated as the first index. The second processing unit is
3. The abnormality diagnosis device according to claim 1, wherein the second index is calculated using the second model constructed by multivariate statistical process control. 2. The abnormality diagnosis device according to claim 1, wherein the statistics include at least one of an average value, a maximum value, and a standard deviation . The first processing unit outputs a first diagnosis result indicating whether or not the target device is abnormal based on the first indicator;
The second processing unit outputs a second diagnosis result indicating whether or not the target device is abnormal based on the second indicator; and
The third processing unit outputs a third diagnosis result indicating whether or not the target device is abnormal based on the third indicator; and
2. The abnormality diagnosis device according to claim 1 , wherein the diagnosis unit outputs a fourth diagnosis result indicating whether or not the target device is abnormal, based on the first to third diagnosis results . 6. The abnormality diagnosis device according to claim 5, wherein the diagnosis unit outputs the fourth diagnostic result based on a result obtained by weighting each of the first diagnostic result, the second diagnostic result, and the third diagnostic result in accordance with the number of normal data . the target device has a switching circuit that operates at a predetermined switching frequency;
The diagnostic data is data in which a frequency spectrum of a frequency band corresponding to the predetermined switching frequency is suppressed among data indicating a frequency spectrum of vibration of the target device.
The abnormality diagnosis device according to any one of claims 1 to 6 . an acquisition unit that acquires vibration data output by a vibration sensor provided in the target device;
The abnormality diagnosis device according to any one of claims 1 to 7, further comprising: a diagnostic data generation unit that generates the diagnostic data based on the vibration data;
The vibration sensor is provided in the target device .

Images