Determination method, determination system, and determination program

Abstract

To determine whether or not a measurement value is suitable for a diagnosis.SOLUTION: In a determination method for performing a determination whether or not a measurement value measured for a rotating device in operation and provided for a diagnosis of the rotating device is suitable for the diagnosis, the determination is performed by using a classifier that classifies whether or not the measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a determination method, a determination system, and a determination program. [Background technology]

[0002] Techniques for diagnosing the condition of rotating equipment such as a pump based on measurements taken on the rotating equipment have been studied. For example, Japanese Patent Laid-Open Publication No. 2022-124145 (Patent Document 1) discloses a diagnostic device that uses a trained model to output output data including diagnostic information on the condition of the pump based on input data including data on the operating status of the pump. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-124145 Summary of the Invention [Problem to be solved by the invention]

[0004] The invention described in Patent Document 1 assumes that pump measurements are performed normally and does not consider verifying the validity of the measurements. However, the reliability of a pump condition diagnosis based on invalid measurements is questionable. For example, if a measurement containing an error due to a measurement mistake or the like is used to diagnose a pump without noticing the existence of the error, the pump may be detected as abnormal even when it is normal. In this case, it may be necessary to dispatch a maintenance technician for an abnormality that does not exist, resulting in unnecessary work. Therefore, to improve the accuracy of diagnosis, it is desirable to ensure that the measurements are appropriate for the diagnosis, but the invention described in Patent Document 1 did not consider this point.

[0005] Therefore, it is desirable to realize a determination method, a determination system, and a determination program for determining whether or not a measurement value is suitable for diagnosis. [Means for solving the problem]

[0006] The determination method according to the present invention is a determination method for determining whether or not a measurement value measured on a rotating device in operation and used to diagnose the rotating device is suitable for the diagnosis, and is characterized in that the determination is made using a classifier that classifies whether or not the measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.

[0007] The determination system according to the present invention comprises a measuring device that obtains measurement values ​​of rotating equipment in operation to be used to diagnose the rotating equipment, and a computing device, wherein the computing device is capable of realizing a determination function that determines whether the measurement values ​​are suitable for the diagnosis, and wherein the determination function makes a determination using a classifier that classifies whether the measurement values ​​are suitable for the diagnosis when an explanatory variable based on the measurement values ​​is given.

[0008] The judgment program according to the present invention, when executed on a computer, can realize a judgment function of judging whether or not a measurement value measured on a rotating device in operation and used to diagnose the rotating device is suitable for the diagnosis, and is characterized in that the judgment function makes a judgment using a classifier that classifies whether or not the measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.

[0009] According to these configurations, it is possible to determine whether or not the measurement value is suitable for diagnosis, thereby improving the reliability of the diagnosis using the measurement value.

[0010] Preferred embodiments of the present invention will be described below, but the scope of the present invention is not limited to the preferred embodiments described below.

[0011] In one aspect of the determination method according to the present invention, the measurement values ​​preferably include at least one of vibration acceleration, discharge pressure, axial displacement, and rotation frequency of the rotating device.

[0012] According to this configuration, the suitability of diagnosis is determined based on measured values ​​that are widely used in diagnosing rotating machines, making the present invention easy to use.

[0013] In one aspect, the determination method according to the present invention includes a first step of extracting features from the measurement values, and a second step of determining whether the measurement values ​​are suitable for the diagnosis based on the features, and it is preferable that the explanatory variables include the features.

[0014] According to this configuration, by extracting the feature amount, it may be possible to reduce the number of factors that need to be taken into consideration when making a judgment, which may make the judgment easier.

[0015] In one aspect of the determination method according to the present invention, the first step preferably includes identifying at least one of the standard deviation of the measurement values, the amount of fluctuation in the standard deviation of the measurement values, the moving average value of the measurement values, the amount of fluctuation in the moving average value of the measurement values, and the amount of fluctuation in the measurement values.

[0016] According to this configuration, the feature quantities are extracted using values ​​that are widely used in diagnosing rotating equipment, so the feature quantities extracted in the first stage can also be used in diagnosing rotating equipment.

[0017] In one aspect of the determination method according to the present invention, the classifier is preferably generated by machine learning.

[0018] This configuration makes it easier to make highly reliable determinations.

[0019] Further features and advantages of the present invention will become more apparent from the following description of exemplary and non-limiting embodiments, which is given with reference to the drawings. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 1 is a schematic diagram illustrating a configuration of a determination system according to an embodiment. [Figure 2]FIG. 1 shows an example of a measurement set suitable for diagnosis. [Figure 3] FIG. 1 illustrates an example of a non-diagnostic measurement set. [Figure 4] FIG. 10 illustrates another example of a non-diagnostic measurement set. [Figure 5] FIG. 10 illustrates another example of a non-diagnostic measurement set. [Figure 6] FIG. 4 is a diagram showing a feature set extracted from the measurement set of FIG. 3. [Figure 7] FIG. 1 illustrates a classifier according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0021] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] The present invention will be described in detail below with reference to the accompanying drawings, in which:

[0024] A determination method, a determination system, and a determination program according to the present invention are applied to a determination system 1 that determines a pump.

[0022] [Configuration of the Determination System] The determination system 1 according to this embodiment includes a measurement terminal 2 and a server device 3 (FIG. 1). The measurement terminal 2 and the server device 3 are configured to be able to communicate with each other via a network N such as the Internet.

[0023] The measurement terminal 2 has a probe 21 that measures the pump and a measurement computer 22. The probe 21 is a device that can measure the pump's vibration acceleration, axial displacement, and rotational frequency (all of which are examples of measured values), and each measured value is input to the measurement computer 22. A plurality of probes 21 may be provided depending on the measured values ​​to be acquired. Furthermore, a measured value of the discharge pressure is input to the measurement computer 22 from a pressure gauge (not shown) installed at the pump's discharge port. The pressure gauge is typically installed at the discharge bend of the pump. The measurement computer 22 is a well-known computer (such as a personal computer) and transmits each measured value to the server device 3. Note that hereinafter, a set of measured values ​​of the vibration acceleration, discharge pressure, axial displacement, and rotational frequency measured by the measurement terminal 2 is referred to as a "measurement set."

[0024] When measuring vibration acceleration, the probe 21 can be an acceleration sensor. It is preferable to provide a plurality of probes 21 each including an acceleration sensor. In this case, typically, one probe 21 is installed at a position where it can measure vibration acceleration in the direction of the rotation axis of the pump, and the other probe 21 is installed at a position where it can measure vibration acceleration in a direction perpendicular to both the discharge direction of the pump and the direction of the rotation axis.

[0025] When measuring axial displacement, the probe 21 can be an axial displacement sensor. It is preferable to provide a plurality of probes 21 each including an axial displacement sensor. In this case, typically, two probes 21 are installed at positions that are perpendicular to the direction of the rotational axis of the pump and can measure axial displacement in two directions that are perpendicular to each other.

[0026] When measuring the rotation frequency, the probe 21 can be an optical sensor. In this case, a reflector (not shown) is attached to the rotating shaft of the pump, and the probe 21 is installed at a position where it can detect the light reflected from the reflector. The probe 21 for measuring the rotation frequency is preferably installed near the probe 21 for measuring the shaft displacement.

[0027] Each measurement value included in the measurement value set may be a value at a specific moment, or may be a group of values ​​measured over a predetermined period of time. In this embodiment, a measurement is performed for 180 seconds at a time, and the measured waveforms of vibration acceleration, discharge pressure, shaft displacement, and rotational frequency during that 180 seconds are used as measurements to diagnose the pump and are subject to suitability judgment. However, setting the measurement period to 180 seconds is merely an example. Alternatively, only a portion of the measurements measured during a predetermined measurement period (for example, only the measurements taken during 150 seconds of the 180-second measurement period) may be used for diagnosis.

[0028] The server device 3 (an example of a computing device) is a known server-type computer, and is capable of diagnosing the state of the pump based on the measurement values ​​measured by the measurement terminal 2. The server device 3 also has installed therein a determination program that determines whether the measurement values ​​measured by the measurement terminal 2 are suitable for diagnosing the pump.

[0029] [Judgment method] The determination method according to this embodiment determines whether or not a set of measurement values ​​obtained by the measurement terminal 2 is suitable for diagnosing a pump.

[0030] A set of measurement values ​​can be obtained, for example, in the following manner. First, while the pump is stopped, the probe 21 of the measurement terminal 2 is installed on the pump. At this time, it is confirmed that the installation state (posture, fixation, etc.) of the probe 21 is correct. Next, the pump is started and waited until the water discharge state stabilizes. The stability of the discharge state can be confirmed by visually checking the discharged water or observing the waveform of the discharge pressure. Once the discharge pressure stabilizes, the measurement terminal 2 begins acquiring measurement values. After acquiring measurement values ​​for a predetermined period (180 seconds in this embodiment), the acquisition of measurement values ​​is stopped. If an obvious abnormality in the operating state of the pump is found during this predetermined period (e.g., water running out on the primary side of the pump), the pump is restored to normal operating conditions and measurement is repeated. After acquiring the measurement values, the probe 21 can be removed from the pump, or the probe 21 can be left attached to the pump if there is a possibility that another measurement will be performed in the near future.

[0031] FIG. 2 shows an example of a measurement set suitable for pump diagnosis, while FIG. 3 shows an example of a measurement set unsuitable for pump diagnosis. In FIG. 2, the measurement waveforms of vibration acceleration, discharge pressure, shaft displacement, and rotational frequency all show roughly consistent shapes throughout the entire 180-second measurement period. In contrast, in FIG. 3, the waveforms of each measurement change at 150 seconds after the start of measurement. In this embodiment, the measurement set used for diagnosis is assumed to be measured for 180 seconds while the pump is operating and discharging water. FIG. 2 shows an example where the pump discharge state was maintained for 180 seconds. In contrast, FIG. 3 shows an example where the operating state of the pump changed due to the water running out on the primary side of the pump 150 seconds after the start of measurement.

[0032] The diagnostic program that uses a measurement value set to diagnose a pump is constructed on the assumption that the measurement value set is input under predetermined conditions for 180 seconds, i.e., while the pump is operating and discharging water. Therefore, a correct diagnostic result will be obtained when the measurement value set in Figure 2 is used for diagnosis, but the accuracy of the diagnostic result cannot be guaranteed when the measurement value set in Figure 3 is used for diagnosis. The determination method according to this embodiment determines whether the measurement value set is suitable for diagnosis.

[0033] More specifically, the diagnostic program according to this embodiment performs diagnosis using measurements acquired during normal pump operation. Therefore, measurements are inappropriate for standby operation (operation with no water on the primary side of the pump), mixed operation (operation in which a mixture of water and air is discharged), and coasting operation (when the pump is not supplied with power but is rotating by coasting).

[0034] 4 and 5 show other examples of measurement sets that are not suitable for diagnosing a pump. In the example of FIG. 4, the value of the axial displacement (y-axis) gradually increases over the 180-second measurement period. The cause of this behavior in the axial displacement cannot be found on the pump side; rather, this behavior is observed when the probe 21 that measures the axial displacement gradually tilts, for example. In the example of FIG. 5, there is a period during the 180-second measurement period where there is no rotational frequency measurement value. Such measurement results can be obtained when the probe 21 is installed incorrectly (for example, is about to come off, is misaligned, etc.).

[0035] However, there may be cases where the waveform of a measurement value clearly contains an abnormality without any errors that could impair the validity of the measurement value. For example, if an event occurs that significantly changes the state of the pump while the measurement value is being taken (such as a broken V-belt), the waveform of the measurement value will change significantly from the time that the event occurred. In this case, even if the measurement method is valid, there is a high possibility that the measurement value obtained will be unsuitable for diagnosis. However, in this case, the user who takes the measurement value will be able to recognize the occurrence of an abnormality in the pump with their own five senses, so it is unlikely that the obtained measurement value will cause any practical problems because it is unsuitable for diagnosis.

[0036] This determination method includes a first step of extracting feature quantities from each measurement value, and a second step of making a determination based on the measurement values ​​and feature quantities. Note that unless otherwise specified, the server device 3 executes the operations related to each of the following steps.

[0037] The first step is to extract features from each measurement value. Examples of extracted features include, but are not limited to, the standard deviation of the measurement values, the variation in the standard deviation of the measurement values, the moving average of the measurement values, the variation in the moving average of the measurement values, and the variation in the measurement values. The standard deviation can be calculated as the standard deviation of the measurement values ​​over a moving window of a predetermined time length (e.g., 1 second). The variation in the standard deviation can be calculated as the difference between the maximum and minimum values ​​of the standard deviation during the measurement period. The moving average can be calculated as a moving average over, for example, 5 seconds. The variation in the moving average can be calculated as the difference between the maximum and minimum values ​​of the moving average during the measurement period.

[0038] It is preferable to use the maximum fluctuation amount of the standard deviation as a feature of the vibration acceleration. It is also preferable to measure the vibration acceleration separately in the direction of the pump's rotation axis and in a direction perpendicular to both the pump's discharge direction and the direction of the rotation axis. The vibration acceleration tends to reflect vibrations with a relatively low frequency.

[0039] For the discharge pressure, it is preferable to use the fluctuation amount of the moving average value of the discharge pressure as the feature quantity. The fluctuation amount can be determined, for example, by obtaining a curve by calculating the median value of the discharge pressure over a moving window of a predetermined time length, and then determining the transition of the difference between the maximum and minimum values ​​of the curve over a moving window of the predetermined time length. It is also preferable to set multiple calculation periods for the fluctuation amount in question. That is, it is preferable to use both the long-term fluctuation amount (for example, the above-mentioned predetermined time length is 18 seconds) and the short-term fluctuation amount (for example, the above-mentioned predetermined time length is 9 seconds) of the moving average value of the discharge pressure as the feature quantity related to the discharge pressure.

[0040] For the shaft displacement, it is preferable to use the amount of fluctuation in the moving average value of the shaft displacement as the feature quantity, and it is more preferable to use the maximum amount of fluctuation in the moving average value of the shaft displacement as the feature quantity. Furthermore, for the shaft displacement, it is preferable to separately measure the shaft displacement in multiple directions intersecting the rotating shaft of the pump. The shaft displacement tends to reflect vibrations with a relatively high frequency.

[0041] Regarding the rotation frequency, it is preferable to use the amount of fluctuation in the rotation frequency as the characteristic quantity, where the amount of fluctuation in the rotation frequency is specified as the difference between the maximum and minimum values ​​of the rotation frequency.

[0042] In the first stage, feature quantities are extracted from each of the measured values ​​(vibration acceleration, discharge pressure, axial displacement, and rotational frequency) included in the measurement value set to be evaluated. Since the computational processes involved in the extraction of the feature quantities exemplified above are publicly known, a description thereof will be omitted here. Hereinafter, the set of extracted feature quantities will be referred to as a feature quantity set. Figure 6 shows an example in which the following feature quantities have been extracted from the measurement value set shown in Figure 3: the maximum fluctuation in the standard deviation of vibration acceleration in the z-axis direction, the maximum fluctuation in the standard deviation of vibration acceleration in the y-axis direction, the long-term and short-term fluctuations in the moving average value of discharge pressure, the maximum fluctuation in the moving average value of x-axis displacement, the maximum fluctuation in the moving average value of y-axis displacement, and the difference between the maximum and minimum values ​​of the rotational frequency.

[0043] The feature values ​​extracted in the first stage may be feature values ​​used in diagnosing the pump. In this case, the process of extracting feature values ​​in the first stage also serves as the process of extracting feature values ​​for diagnosing the pump, thereby reducing the total amount of calculations required in the series of steps from determining the suitability of the measurement value set to diagnosing the pump.

[0044] The second stage is a stage in which a determination is made based on the measured values ​​and feature quantities. In this stage, a classifier is used to make the determination. In this embodiment, a case in which a classifier generated by machine learning is used will be described as an example.

[0045] The generation of a classifier through machine learning can be achieved by a supervised learning technique. First, multiple sets of measurement values ​​are created as training data, with labels indicating whether the measurement values ​​are diagnostically appropriate. The labeling of diagnostic appropriateness to create the training data is manually performed by an operator with the skills to determine the appropriateness of a measurement value set. Note that the labeling of diagnostic appropriateness may be performed by referring to the measurement value set itself, or may be performed by referring to a feature set after a process of extracting a feature set from the measurement value set. Furthermore, there is no prohibition on operators using a computational method when labeling diagnostic appropriateness.

[0046] Next, a machine learning algorithm is applied to the obtained training data to obtain a trained model. This trained model is a classifier that classifies whether a certain set of measurement values ​​and a set of feature values ​​are suitable for pump diagnosis when the set is given as explanatory variables. Typically, a trained model is obtained using part of the obtained training data as training data, and the remaining part of the training data is used as test data to verify the trained model. At this time, if the validity of the obtained trained model is low (for example, if the probability that the classification results of the test data and the diagnostic suitability label match is low), measures can be taken to improve the validity of the trained model, such as changing the features used, performing dimensionality reduction on the explanatory variables using principal component analysis, adding or changing the training data, changing the algorithm used, or changing the parameters when applying the algorithm.

[0047] The algorithm used to generate a trained model from training data is not particularly limited. Examples of such algorithms include, but are not limited to, support vector machines, decision trees, random forests, neural networks, k-nearest neighbor algorithms, CART, and DBSCAN. Among these, decision trees are one of the preferred algorithms because they tend to produce trained models that are easy for users to understand (high explainability). Note that multiple trained models may be generated using multiple algorithms, and the trained model with the highest validity may be ultimately adopted.

[0048] When the measurement value set and the feature set extracted in the first stage are given as explanatory variables to the classifier (trained model) obtained by the above method, the measurement value set is classified as being suitable for pump diagnosis. This makes it possible to determine whether the measurement value set is suitable. The result of the determination is output to a display device such as an LCD display on the server device 3 or another terminal (not shown) that has accessed the server device 3.

[0049] If the measurement value set is determined to be suitable for diagnosis, the measurement value set is used to diagnose the pump. Note that the feature set extracted for determining the suitability of the measurement value set may also be used for diagnosing the pump. On the other hand, if the measurement value set is determined to be unsuitable for diagnosis, the user of the determination system 1 is notified that they are requested to repeat the measurement. Note that, rather than unconditionally repeating the measurement, the user who received the notification may manually perform a determination on the measurement value set that the determination system 1 determined to be unsuitable for diagnosis, and repeat the measurement only if the manual determination also determines that the measurement set is unsuitable for diagnosis.

[0050] After the judgment, the trained model may be updated (re-trained) using a set of the measurement value set, feature set, judgment result by the judgment system 1, and judgment result by the user that were used when using the judgment system 1. Note that the trained model may be updated again using an algorithm, or may be updated by manually adjusting the parameters.

[0051] Other Embodiments Finally, other embodiments of the determination method, determination system, and determination program according to the present invention will be described. Note that the configurations disclosed in the following embodiments can be applied in combination with the configurations disclosed in other embodiments, as long as no contradiction occurs.

[0052] In the above embodiment, the vibration acceleration, discharge pressure, shaft displacement, and rotation frequency are measured. However, the measurement values ​​to be judged in the present invention are not limited to those as long as they are used to diagnose rotating equipment.

[0053] In the above embodiment, a configuration has been described in which feature quantities are extracted from measured values ​​and the feature quantities are provided to a classifier as explanatory variables. However, in the present invention, feature quantities may be extracted arbitrarily. For example, the measured values ​​themselves may be explanatory variables based on the measured values. Furthermore, a determination may be made based on only either the measured values ​​or the feature quantities.

[0054] In the above embodiment, an example has been described in which the measurement terminal 2 that acquires measurement values ​​and the server device 3 that determines the measurement values ​​are separate devices. However, in the present invention, the acquisition of measurement values ​​and the determination of measurement values ​​may be performed using the same device. Furthermore, in the above embodiment, an example has been described in which the measurement terminal 2 and the server device 3 can communicate with each other via the network N, but when the device that acquires measurement values ​​and the device that determines measurement values ​​are separate devices, the manner in which measurement values ​​are exchanged between them is not limited.

[0055] In the above embodiment, a case where a classifier generated by machine learning is used has been described as an example. However, the method of obtaining a classifier is not limited in the present invention. For example, a classifier whose parameters are set based on a preliminary experiment or an empirical rule may be used.

[0056] Regarding other configurations, it should be understood that the embodiments disclosed in this specification are illustrative in all respects and that the scope of the present invention is not limited thereby. Those skilled in the art will easily understand that appropriate modifications are possible without departing from the spirit of the present invention. Therefore, other embodiments modified without departing from the spirit of the present invention are naturally included in the scope of the present invention. [Example]

[0057] The present invention will be further described below with reference to examples, but the present invention is not limited to these examples.

[0058] [Measurement set and feature set] The vibration acceleration, discharge pressure, axial displacement, and rotational frequency of the pump were measured, and a set of these measurement values ​​was defined as a measurement set. From each measurement value in the measurement set, the following feature quantities were extracted: the maximum fluctuation in the standard deviation of the vibration acceleration in the z-axis direction, the maximum fluctuation in the standard deviation of the vibration acceleration in the y-axis direction, the long-term fluctuation and short-term fluctuation in the moving average value of the discharge pressure, the maximum fluctuation in the moving average value of the x-axis displacement, the maximum fluctuation in the moving average value of the y-axis displacement, and the difference between the maximum and minimum values ​​of the rotational frequency. A set of these feature quantities was defined as a feature set.

[0059] [Classifier generation] A total of 161 measurement sets were created (measured), and feature sets were extracted from each measurement set. Each measurement set was manually judged as appropriate or inappropriate, and a diagnostic appropriate or inappropriate label was attached. Through the above procedure, 161 pairs of measurement sets, feature sets, and diagnostic appropriate or inappropriate labels were obtained, and these were used as training data. Using the machine learning library scikit-learn in a Python environment, a classifier (trained model) in the form of a decision tree was generated using the CART algorithm with the above training data.

[0060] The generated decision tree is shown in Figure 7. The branching condition for node N1 was whether the x-axis displacement was below a predetermined threshold. The branching condition for node N2 was whether the rotational frequency was below a predetermined threshold. The branching condition for node N3 was whether the short-term fluctuation in the moving average value of the discharge pressure was below a predetermined threshold. The branching condition for node N4 was whether the fluctuation in the rotational frequency was below a predetermined threshold. The branching condition for node N5 was whether the long-term fluctuation in the moving average value of the discharge pressure was below a predetermined threshold. The branching condition for node N6 was whether the long-term fluctuation in the moving average value of the discharge pressure was below a predetermined threshold. However, the thresholds for nodes N5 and N6 were different from each other. The branching condition for node N7 was whether the rotational frequency was below a predetermined threshold. However, the thresholds for nodes N2 and N7 were different from each other.

[0061] In this example, the branching conditions for nodes N1, N2, and N7 are based on the measurement values ​​included in the measurement value set, and the branching conditions for nodes N3, N4, N5, and N6 are based on the features included in the feature set. In this way, judgments based on measurement values ​​and judgments based on features can be used together.

[0062] In the example of FIG. 7, among the prepared feature set, the feature including the standard deviation was not adopted as a branching condition. As in this example, in a classifier obtained by machine learning, not all of the prepared measurement values ​​and feature values ​​are necessarily used as explanatory variables. Furthermore, depending on the training data used, a classifier may be generated that uses feature values ​​including the standard deviation as explanatory variables, unlike the example of FIG. 7. Note that, according to the study by the inventors, the rotational frequency (measurement value) and the long-term fluctuation amount (feature value) of the moving average value of the discharge pressure were more likely to be used as explanatory variables than other measurement values ​​and feature values.

[0063] Rotating equipment has the property of exhibiting periodic behavior, such as vibrations and pressure fluctuations, due to its rotational motion. The characteristics of measurement values ​​resulting from periodic behavior can be well expressed by standard deviation, maximum value, difference between maximum and minimum values, etc., but standard deviation is particularly preferable because feature quantities such as maximum value are easily affected by temporary fluctuation factors.

[0064] [Classifier validation] The classifier was validated using a set of 69 measurements that was created (measured) separately from the training data. Features were extracted from each measurement in the validation measurement set in the same way as for the measurement set used as training data, and this set of features was used as the validation feature set. Each validation feature set was manually judged to be appropriate or inappropriate and labeled as diagnostic appropriate or inappropriate. Each validation feature set was also input into the created classifier, and an appropriate or inappropriate judgment was made by the classifier. The results of the manual judgment and the appropriate or inappropriate judgment made by the classifier were then compared (Table 1).

[0065] When the results of the classifier's suitability assessment match the results of the human-made suitability assessment, it can be said that the classifier has made an appropriate assessment. In this example, an appropriate assessment was made in 64 of the 69 cases (93%) tested. Furthermore, when the results of the classifier's suitability assessment did not match the results of the human-made suitability assessment, the number of discrepancies in which a measurement set that should be judged unsuitable for diagnosis was judged to be suitable for diagnosis (so-called "slip-through") was less than the number of discrepancies in which a measurement set that should be judged suitable for diagnosis was judged to be unsuitable for diagnosis (so-called "overdetection"). From the above, it can be said that the assessment method using the classifier according to this example has sufficient assessment accuracy for practical use as a pre-assessment before the measurement set is used for diagnosis.

[0066] Table 1: Classifier validation [Table 1] [Industrial Applicability]

[0067] The present invention can be used to determine whether or not measurements used for diagnosing rotating equipment such as pumps are suitable for diagnosis. [Explanation of symbols]

[0068] 1: Judgment system 2: Measurement terminal 21: Probe 22: Measurement computer 3: Server device

Claims

1. 1. A method for determining whether or not measurement values ​​measured on a rotating machine during operation and used for diagnosing the rotating machine are suitable for the diagnosis, comprising: A determination method using a classifier that classifies whether or not a measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.

2. The method according to claim 1 , wherein the measurement values ​​include at least one of vibration acceleration, discharge pressure, axial displacement, and rotation frequency of the rotating equipment.

3. a first step of extracting a feature value from the measurement value; and a second step of determining whether or not the measurement value is suitable for the diagnosis based on the feature value, The determination method according to claim 1 , wherein the explanatory variables include the feature quantities.

4. 4. The method of claim 3, wherein the first step includes identifying at least one of a standard deviation of the measurement values, a fluctuation amount of the standard deviation of the measurement values, a moving average value of the measurement values, a fluctuation amount of the moving average value of the measurement values, and a fluctuation amount of the measurement values.

5. The determination method according to any one of claims 1 to 4, wherein the classifier is generated by machine learning.

6. a measuring device for obtaining measurements of the rotating equipment during operation, the measurements being used to diagnose the rotating equipment; a computing device, the computing device is capable of realizing a determination function of determining whether the measurement value is suitable for the diagnosis, A determination system in which the determination function uses a classifier that classifies whether or not a measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.

7. When executed on a computer, the program can realize a determination function of determining whether or not measurement values ​​measured on a rotating machine in operation and used for diagnosing the rotating machine are suitable for the diagnosis, A judgment program in which the judgment function uses a classifier that classifies whether or not a measurement value is suitable for the diagnosis when an explanatory variable based on the measurement value is given.

Images