Abnormality diagnosis method, abnormality diagnosis device, and computer-readable storage medium

Abstract

The abnormality diagnosis method for diagnosing abnormality of a device to be diagnosed comprises: a step of obtaining multivariate time series data on a plurality of measurement items from the device; a step of diagnosing abnormality of the operating state of the device using the multivariate time series data; and a step of diagnosing the cause of the abnormality when the abnormality of the operating state of the device is determined in the step of diagnosing the abnormality. The step of diagnosing the cause of the abnormality comprises: a step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality occurs; a step of extracting a feature quantity of the second interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality occurs; a step of obtaining a change in feature quantity based on the difference between the feature quantity of the first interval and the feature quantity of the second interval obtained by the same feature quantity extraction method; and a step of diagnosing the measurement item that is the cause of the abnormality based on the change in feature quantity of the plurality of measurement items.

Description

Technical Field

[0001] The present disclosure relates to an abnormality diagnosis method, an abnormality diagnosis device, and an abnormality diagnosis program. Background Art

[0002] In the state monitoring of equipment, multivariate time series data including various operation data of equipment and machinery and measurement data obtained by measuring vibration generated from the equipment and temperature of the equipment through various sensors are used. By analyzing the multivariate time series data obtained from the equipment to be diagnosed, it is possible to determine whether the equipment and machinery are operating normally. For example, in the Mahalanobis Taguchi (MT) method, an analysis output value is calculated based on the multivariate time series data, and the analysis output value is used to determine whether the equipment is abnormal.

[0003] There is also a need for a technique for analyzing, when an abnormality in a device is determined in the above analysis, what the measurement item is that is the cause of the abnormality and what kind of abnormal phenomenon (abnormal pattern) has occurred in the device.

[0004] For example, Japanese Patent Application Laid-Open No. 8-6635 (Patent Document 1) discloses a method for determining the cause of an abnormality as follows: the cause of an abnormality is determined by comparing the measured data obtained from each device of the facility to be diagnosed with the pseudo data obtained using the physical model. In this method, the calculation parameters of the physical model are changed based on the failure mode evaluation and analysis information of the device prepared in advance, thereby changing the pseudo data. Furthermore, when the measured data and the pseudo data are consistent, it is determined that the failure mode is correct.

[0005] In addition, in Japanese Patent Publication No. 2015-172945 (Patent Document 2), as a method for determining an abnormal pattern, the following method is disclosed: the result obtained by vector quantizing the measurement values ​​of various measurement items at the time when the abnormality is determined is used as a cause event, and the failure event that occurs from the same time until a certain time has passed is used as a result event, and a matrix of the frequency of the combination of the cause event and the result event is prepared. In this method, the distance between the vector quantization value of the measurement value at the time of the abnormality and the cause event is calculated, and the cause event with the closest distance is extracted, and the result event with the highest frequency is extracted.

[0006] Prior art literature

[0007] Patent Document 1: Japanese Patent Application Laid-Open No. 8-6635

[0008] Patent Document 2: Japanese Patent Application Publication No. 2015-172945 Summary of the invention

[0009] However, in the method described in Patent Document 1, simulation is performed based on a physical model. In order to determine the cause of the abnormality and the abnormal pattern when the phenomenon is reproduced, professional knowledge and technology related to the diagnostic object machine are required. Therefore, personnel who are not proficient in the diagnostic object machine may misdiagnose. In addition, when there are multiple inferred abnormal causes, it may take time to determine the abnormal cause and the abnormal pattern. Therefore, for abnormal situations where every second counts, the effectiveness may be reduced.

[0010] On the other hand, as described in Patent Document 2, in a method of creating a database based on events that actually occurred and determining abnormal patterns based on the distance between the events and the events registered in the database, it is difficult to enrich the database in devices where abnormalities rarely occur, so there is a possibility of erroneous determination of abnormal patterns.

[0011] The present disclosure is completed to solve such a problem, and its purpose is to accurately identify the measurement item that is the cause of the abnormality in a short time without requiring professional knowledge and technology related to the diagnostic object device / machine. In addition, the present disclosure aims to accurately identify the abnormal phenomenon that is the abnormal pattern in a short time without requiring professional knowledge and technology related to the diagnostic object.

[0012] The abnormality diagnosis method involved in the first form of the present disclosure is an abnormality diagnosis method for diagnosing abnormalities of equipment to be diagnosed, and comprises: a step of obtaining multivariate time series data on a plurality of measurement items from the equipment; a step of diagnosing abnormalities in the operating state of the equipment using the multivariate time series data; and a step of diagnosing the cause of the abnormality when the abnormality in the operating state of the equipment is determined in the step of diagnosing the abnormality. The step of diagnosing the cause of the abnormality includes: a step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality occurs; a step of extracting a feature quantity of the second interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality occurs; a step of obtaining a change in the feature quantity based on the difference between the feature quantity of the first interval and the feature quantity of the second interval obtained by the same feature quantity extraction method; and a step of diagnosing the measurement item that is the cause of the abnormality based on the change in the feature quantity of the plurality of measurement items.

[0013] The abnormality diagnosis method involved in the second form of the present disclosure is an abnormality diagnosis method for diagnosing abnormalities in equipment of a diagnosis object, and comprises: a step of obtaining multivariate time series data on multiple measurement items from the equipment; a step of using the multivariate time series data to diagnose abnormalities in the operating state of the equipment; and a step of diagnosing an abnormal pattern that presents an abnormal phenomenon when an abnormality in the operating state of the equipment is determined in the step of diagnosing the abnormality. The step of diagnosing an abnormal pattern includes: a step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from multivariate time series data of the first interval before abnormal determination; a step of extracting a feature quantity of the second interval by at least one feature quantity extraction method for each measurement item from multivariate time series data of the second interval after abnormal determination; a step of obtaining a variation of the feature quantity based on a difference between a feature quantity of the first interval and a feature quantity of the second interval obtained by the same feature quantity extraction method; a step of generating a variation vector with a variation of the feature quantity, a value obtained by using the variation of the feature quantity and a classification method according to a predetermined n-stage (n is an integer) classification method, or a value obtained by using the variation of a plurality of feature quantities using different feature quantity extraction methods and a classification method according to a predetermined m-stage (m is an integer) classification method as elements for each measurement item; a step of creating a database in which each abnormal phenomenon of a plurality of abnormal phenomena is registered in association with the variation vector corresponding to the abnormal phenomenon; and a step of diagnosing an unknown abnormal pattern based on the database and the variation vector generated by the generating step.

[0014] According to the present disclosure, it is possible to identify the measurement item that is the cause of the abnormality with high accuracy in a short time without requiring professional knowledge and technology related to the diagnostic object. Alternatively, it is possible to identify the abnormal phenomenon that is the abnormal pattern with high accuracy in a short time without requiring professional knowledge and technology related to the diagnostic object. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a block diagram showing the functional structure of the abnormality diagnosis device according to the embodiment.

[0016] Figure 2 This is a flowchart for explaining the diagnosis process of the abnormality cause executed by the second diagnosis unit.

[0017] Figure 3 It is a diagram showing an example of selecting an evaluation area.

[0018] Figure 4 It is a diagram showing an example of two-dimensional rendering.

[0019] Figure 5 This is a flowchart for explaining the diagnosis process of the abnormal mode executed by the third diagnosis unit.

[0020] Figure 6 This is a diagram for explaining an example of a method for determining the value of each element of a variation vector.

[0021] Figure 7 This is a diagram showing an example of a database.

[0022] Figure 8 It is a diagram for explaining an output example of the determination result of the third output unit.

[0023] Fig. 9 This is a block diagram showing a hardware configuration example of the abnormality diagnosis device.

[0024] (Explanation of symbols)

[0025] 1: abnormality diagnosis device; 2: data reading unit; 3: data display unit; 4: first diagnosis unit; 5: second diagnosis unit; 6: third diagnosis unit; 7: first output unit; 8: second output unit; 9: third output unit; 10: model making unit; 11: analytical output value calculation unit; 12: abnormality determination unit; 13: data standardization unit; 14: evaluation area selection unit; 15: feature quantity change calculation unit; 16: abnormality cause determination unit; 17: change quantity vectorization unit; 18: database registration unit; 19: similarity calculation unit; 20: abnormality pattern determination unit; 100: CPU; 101: ROM; 102: RAM; 103: HDD; 104: communication interface; 105: I / O interface; 106: input unit; 107: display unit. DETAILED DESCRIPTION

[0026] Hereinafter, the embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In addition, the same reference numerals are added to the same or corresponding parts in the drawings, and the description thereof will not be repeated in principle.

[0027] [A. Functional structure of abnormality diagnosis device]

[0028] First, the functional structure of the abnormality diagnosis device according to the embodiment will be described.

[0029] Figure 1 The abnormality diagnosis device 1 according to the embodiment is configured to determine whether the operating state of a device or machine (hereinafter simply referred to as "device") is normal by analyzing multivariate time series data consisting of operating data and measurement data obtained from the device or machine to be diagnosed, and output the determination result.

[0030] When the operating state of the diagnostic object equipment is determined to be abnormal in the above determination, the abnormality diagnosis device 1 is further configured to diagnose the measurement item that is the cause of the abnormality. Here, the measurement item refers to an item obtained as operating data or measurement data. In addition, the abnormality diagnosis device 1 is configured to diagnose the abnormal mode in which the measurement item participates.

[0031] The equipment to be diagnosed is, for example, a generator, FA (factory automation) equipment, power distribution equipment, elevators, railway electric equipment, etc. In this embodiment, a configuration in which the abnormality diagnosis device 1 is applied to the state monitoring and abnormality diagnosis of a generator is exemplified.

[0032] Reference Figure 1 The abnormality diagnosis device 1 includes a data reading unit 2 , a data display unit 3 , a first diagnosis unit 4 , a second diagnosis unit 5 , a third diagnosis unit 6 , a first output unit 7 , a second output unit 8 , and a third output unit 9 .

[0033] (A-1) Data reading unit

[0034] The data reading unit 2 reads multivariate time series data obtained from the diagnostic object device. That is, the data reading unit 2 obtains multivariate time series data about multiple measurement items from the diagnostic object device. In the present application specification, "variable" refers to a measurement item representing various physical quantities obtained from the diagnostic object. "Multivariate time series data" refers to time series data representing the temporal changes of multiple interrelated measurement items (multivariate data).

[0035] When a generator is used as an example, the measurement items include: items related to the operation data such as the output, rotation speed, voltage and current of the generator; and items related to the time series data of the measurement data such as temperature and vibration measured by a sensor group installed in the machine or component constituting the generator. In addition, it is preferable to include data related to the environment in which the generator is installed, such as the outside temperature, and the operation data and measurement data of the equipment related to the working state of the generator. In addition, these measurement items are only examples, and the number of items is not limited. The time series data composed of these multiple measurement items is treated as multivariate time series data.

[0036] (A-2) Data display section

[0037] The data display unit 3 displays the multivariate time-series data received from the data reading unit 2 in a graphical form.

[0038] (A-3) First Diagnosis Department

[0039] The first diagnostic unit 4 diagnoses the presence or absence of abnormality in the operating state of the diagnostic target device using the multivariate time series data received from the data reader 2. That is, the first diagnostic unit 4 constitutes an "abnormality diagnostic unit" for diagnosing abnormality in the operating state of the device.

[0040] The first diagnostic unit 4 uses the multivariate time series data to calculate an analytical output value. The first diagnostic unit 4 determines whether the operating state of the diagnostic target device is abnormal based on the calculated analytical output value. As such an abnormality determination method, well-known methods such as the Mahometan-Taguchi (MT) method, a one-class support vector machine, a nearest neighbor method, a regression analysis, and a partial space method can be used.

[0041] Specifically, the first diagnostic unit 4 includes a model creation unit 10 , an analysis output value calculation unit 11 , and an abnormality determination unit 12 .

[0042] The model making unit 10 uses the multivariate time series data as learning data and makes a learned model used in abnormal diagnosis according to the selected abnormality determination method. The learned model can also be called a normal model. The model making unit 10 can make a learned model using data obtained by performing arbitrary calculation processing (hereinafter also referred to as "preprocessing") on the multivariate time series data as needed. The arbitrary calculation processing includes, for example, the processing of calculating the differential value or moving average value of the data of each measurement item, or the processing of calculating the sum, difference or average of the data of at least two or more measurement items.

[0043] The analytical output value calculation unit 11 calculates the analytical output value according to the selected abnormality determination method using the prepared learning model and the multivariate time series data of the diagnostic object. The abnormality determination unit 12 determines whether the operating state of the diagnostic object equipment is abnormal by comparing the analytical output value calculated by the analytical output value calculation unit 11 with a preset threshold value for abnormality determination.

[0044] For example, when the MT method is used in the abnormality determination method, the model making unit 10 makes a unit space based on the multivariate time series data or preprocessed data during normal operation. The unit space corresponds to an embodiment of the "learned model" and becomes a reference for determining the operating state of the diagnostic target device.

[0045] The analysis output value calculation unit 11 calculates the Mahalanobis distance by the MT method using the created unit space and the multivariate time series data of the diagnostic object read by the data reading unit 2 and preprocessed as needed. The Mahalanobis distance can be calculated by a known method.

[0046] The abnormality determination unit 12 determines whether there is an abnormality in the operating state of the diagnostic target device by comparing the calculated Mahalanobis distance with a preset threshold value. When the abnormality determination unit 12 determines that there is no abnormality, the abnormality diagnosis device 1 does not proceed to the diagnosis of the abnormality cause and abnormality pattern described later, but continues to use the abnormality determination unit 12 to determine whether there is an abnormality. On the other hand, when it is determined that there is an abnormality, the abnormality diagnosis device 1 performs the diagnosis of the abnormality cause and abnormality pattern.

[0047] (A-4) 1st output section

[0048] The first output unit 7 outputs the determination result of the abnormality determination unit 12 of the first diagnostic unit 4. Specifically, the first output unit 7 includes, for example, a display unit, and displays the determination result on the display unit. Furthermore, the first output unit 7 may be configured to send the determination result to an external device using a communication unit (not shown).

[0049] (A-5) Second diagnosis department

[0050] The second diagnostic unit 5 determines which measurement item among the plurality of measurement items (variables) included in the multivariate time series data is the cause (main cause) of the abnormality. That is, the second diagnostic unit 5 constitutes an "abnormality cause diagnostic unit" for diagnosing the measurement item that is the cause of the abnormality.

[0051] The second diagnostic unit 5 uses multivariate time series data of a first interval (time range) which is a normal state before the moment when the abnormality determination unit 12 of the first diagnostic unit 4 determines that an abnormality has occurred (hereinafter also referred to as the "abnormality occurrence moment"), and multivariate time series data of a second interval which is an abnormal state after the abnormality occurrence moment, to diagnose the cause of the abnormality.

[0052] In addition, the multivariate time series data used by the second diagnostic unit 5 in diagnosis may be either the multivariate time series data received from the data reading unit 2 or the data obtained by preprocessing the multivariate time series data. That is, the "multivariate time series data" described below is a concept that includes the multivariate time series data and the data obtained by preprocessing the multivariate time series data.

[0053] The second diagnostic unit 5 extracts the feature quantity of the first interval from the multivariate time series data of the first interval as the normal state before the abnormality occurs by at least one feature quantity extraction method for each measurement item. The second diagnostic unit 5 extracts the feature quantity of the second interval from the multivariate time series data of the second interval as the abnormal state after the abnormality occurs by at least one feature quantity extraction method for each measurement item. In addition, the second diagnostic unit 5 calculates the difference of the feature quantity of the interval before and after the abnormality occurs, that is, the change amount of the feature quantity, for each measurement item. The second diagnostic unit 5 determines the measurement item whose change amount of the feature quantity among the change amounts of the multiple feature quantities corresponding to the multiple measurement items is greater than a preset threshold value as the cause of the abnormality.

[0054] Specifically, the second diagnostic unit 5 includes a data normalization unit 13 , an evaluation region selection unit 14 , a feature quantity change calculation unit 15 , and an abnormality cause identification unit 16 . Figure 2 1 is a flowchart for explaining the abnormality cause diagnosis process executed by the second diagnosis unit 5. When the first diagnosis unit 4 (abnormality determination unit 12) determines that the diagnosis target device has an abnormality, the diagnosis process is executed. Figure 2 Flowchart shown.

[0055] After the diagnosis process of the abnormality cause is started, the data normalization unit 13 first normalizes the time series data of each measurement item with respect to the multivariate time series data through step S10. This is to evaluate the change amount of the characteristic quantity before and after the abnormality occurs in each measurement item with the same reference. The normalization method determined using the multivariate time series data of the first interval before the abnormality occurs can be applied to the multivariate time series data of the second interval after the abnormality occurs.

[0056] As a first example of data normalization, the data normalization unit 13 can convert data into dimensionless numbers by setting the average to 0 and the variance to 1 for each measurement item.

[0057] Alternatively, as a second example, the data normalization unit 13 can convert the data into dimensionless numbers by setting the maximum value to 1 and the minimum value to 0 for each measurement item.

[0058] Alternatively, as a third example, the data normalization unit 13 can convert the data into dimensionless numbers by determining a standardized reference value based on the learned model used for abnormality determination in the first diagnostic unit 4. For example, when the learned model is a unit space in the MT method, the maximum value and minimum value of the unit space data of each measurement item and an arbitrarily determined determination value can be used as a reference to convert the data into dimensionless numbers.

[0059] Specifically, the measured values ​​within the range of the minimum value and the maximum value of the unit space data are converted to "0". In addition, the upper limit judgment value determined for each of the measurement items is converted to "1", and the lower limit judgment value is converted to "-1". Furthermore, the values ​​from -1 to 0, the values ​​from 0 to 1, and the values ​​above 1 or below -1 are converted by linear transformation or the like.

[0060] For example, a linear transformation can be performed using a linear equation (y=ax+b; x is the value after transformation, y is the value before transformation). In a range greater than 0, a linear equation can be prepared by setting the slope a={y(upper limit judgment value)-y(unit space data maximum value)} / (1-0) and the intercept b=y(unit space data maximum value). In a range less than 0, a linear equation can be prepared by setting the slope a={y(unit space data minimum value)-y(lower limit judgment value)} / {0-(-1)} and the intercept b=y(unit space data minimum value).

[0061] Furthermore, along with the updating of the unit space (or the learned model), the criterion (the maximum value and the minimum value of the unit space data in the case of the MT method) and the transformation formula used in the transformation may be updated.

[0062] exist Figure 2 In step S11, the evaluation area selection unit 14 determines an interval (first interval) to be evaluated as "before the abnormality occurs (i.e., normal state)" and an interval (second interval) to be evaluated as "after the abnormality occurs (i.e., abnormal state)" in order to compare the measurement items before and after the abnormality occurs. Figure 3 FIG. 1 is a diagram showing an example of selecting an evaluation area. Figure 3 , a time series change of the analysis output value calculated by the first diagnostic unit 4 is shown.

[0063] like Figure 3 As shown, the evaluation area selection unit 14 defines the interval in which the analysis output value presents a value smaller than the abnormality determination threshold value (as a normal state) as the interval "before the abnormality occurs" (the first interval), and defines the interval in which the analysis output value presents a value greater than the abnormality determination threshold value (as an abnormal state) as the interval "after the abnormality occurs" (the second interval). In addition, the number of data (time range) of the interval before the abnormality occurs and the interval after the abnormality occurs can be arbitrarily set. These two intervals may not necessarily be continuous.

[0064] In step S12, the feature quantity change calculation unit 15 extracts the feature quantity of the first interval from the multivariate time series data of the first interval before the abnormality occurs or the data after the normalization process thereof, for each measurement item, by at least one feature quantity extraction method. The feature quantity change calculation unit 15 extracts the feature quantity of the second interval from the multivariate time series data of the second interval after the abnormality occurs or the data after the normalization process thereof, for each measurement item, by at least one feature quantity extraction method. Furthermore, the feature quantity change calculation unit 15 calculates the difference between the feature quantity in the first interval before the abnormality occurs and the feature quantity in the second interval after the abnormality occurs, obtained by the same feature quantity extraction method, thereby obtaining the change amount of the feature quantity. In addition, the number of feature quantities of each measurement item obtained from each interval may be one or more.

[0065] Specifically, initially, the feature quantity change calculation unit 15 calculates at least one feature quantity for each measurement item based on the multivariate time series data of each interval. That is, the feature quantity change calculation unit 15 extracts at least one feature quantity for each measurement item from the multivariate time series data of the first interval before the abnormality occurs, which is standardized by the data normalization unit 13. In addition, the feature quantity change calculation unit 15 extracts at least one feature quantity for each measurement item from the multivariate time series data of the second interval after the abnormality occurs, which is standardized by the data normalization unit 13.

[0066] In addition, any method can be used to extract the feature quantity. For example, statistics such as the mean, variance, standard deviation, median, mode, maximum value, minimum value, skewness, and kurtosis can be obtained. However, the type of feature quantity extracted for each measurement item is the same.

[0067] Next, the feature quantity change calculation unit 15 calculates the difference between the feature quantity of the first interval before the abnormality occurs and the second interval after the abnormality occurs, which are obtained by the same feature quantity extraction method, for each feature quantity of at least one feature quantity for each measurement item. The calculated difference in feature quantity is equivalent to the change in feature quantity after the abnormality occurs relative to the change in feature quantity before the abnormality occurs.

[0068] In step S13, the abnormality cause determination unit 16 determines the measurement item that is the cause of the abnormality based on the change amount of the feature amount calculated by the feature amount change calculation unit 15. Specifically, the abnormality cause determination unit 16 determines that the data of all the measurement items of the change amount of the feature amount among the at least one feature amount that has been obtained is close to 0 does not change before and after the abnormality occurs, and therefore infers that it is not the cause of the abnormality. In contrast, for the measurement item of the change amount of any feature amount among the change amount of the at least one feature amount that exceeds the preset threshold range, the abnormality cause determination unit 16 determines the measurement item as the cause of the abnormality.

[0069] (A-6) Second output section

[0070] In step S14, the second output unit 8 outputs the determination result in the abnormality cause determination unit 16 of the second diagnosis unit 5. The second output unit 8, for example, has a display unit, and displays the determination result on the display unit. In addition, the second output unit 8 can also be configured to send the determination result to an external device using a communication unit not shown in the figure. The second output unit 8 displays the change amount of at least one characteristic quantity calculated by the characteristic quantity change calculation unit 15 for each measurement item. For example, in the case of displaying the result about the change amount of one characteristic quantity, the change amount of the characteristic quantity of each measurement item can be displayed using a bar graph or the like. In addition, in the case of displaying the result about the change amount of multiple characteristic quantities, a multidimensional graph can be used. The axes of the multidimensional graph are used as the change amount of the characteristic quantity, and the value of the change amount of the characteristic quantity of each measurement item is plotted.

[0071] Figure 4 is a diagram showing an example of displaying the results using two-dimensional rendering. Figure 4 In the two-dimensional plot shown, a two-dimensional plane is defined with the change amount of a certain feature quantity X (e.g., average) as the horizontal axis and the change amount of another feature quantity Y (e.g., standard deviation) as the vertical axis. On this two-dimensional plane, the combined value of the change amount of the feature quantity X and the change amount of the feature quantity Y is plotted for each measurement item. In addition, the change amounts of the feature quantities X and Y are both dimensionless numbers. Figure 4 In the example, for a total of six types of measurement items, namely, the voltage, current, winding temperature, gas temperature, bearing metal temperature, and shaft vibration of the generator, a combination of the average change and the standard deviation change of each measurement item is plotted.

[0072] like Figure 4As shown, the two-dimensional plane is divided into two regions. The first region RGN1 is a region including the origin where the changes in the feature quantities X and Y are both 0, and is surrounded by the threshold range of the change in the feature quantity X (-TH1 to +TH1) and the threshold range of the change in the feature quantity Y (-TH2 to +TH2). In addition, "+TH1" represents the upper limit value of the threshold range of the change in the feature quantity X (average), and "-TH1" represents the lower limit value. "+TH2" represents the upper limit value of the threshold range of the change in the feature quantity Y (standard deviation), and "-TH2" represents the lower limit value. The second region RGN2 is a region outside the first region RGN1, and is a region surrounding the first region RGN1.

[0073] The first region RGN1 is a region indicating that the changes in the feature quantity X and the feature quantity Y are both within the threshold range. The second region RGN2 is a region indicating that the changes in at least one of the feature quantity X and the feature quantity Y exceed the threshold range. That is, the first region RGN1 is a region indicating that the measurement item is normal, and the second region RGN2 is a region indicating that the measurement item is abnormal.

[0074] exist Figure 4 In the example of , the plotted values ​​of the four measurement items, namely, voltage, current, bearing metal temperature, and shaft vibration, are located in the first region RG1. On the other hand, the plotted values ​​of the two measurement items, namely, gas temperature and winding temperature, are located in the second region RGN2. Therefore, the abnormality cause determination unit 16 of the second diagnostic unit 5 can determine the gas temperature and winding temperature as the abnormality causes.

[0075] like Figure 4 As shown, by plotting the change amount of at least one characteristic quantity for each of the plurality of measurement items, the user can visually understand the change amount of the characteristic quantity of each measurement item. Specifically, the user can infer the cause of the abnormality based on the position where the change amount of the characteristic quantity of each measurement item is plotted. In addition, the change amount of the characteristic quantity of the measurement item when the abnormality occurs can be easily grasped based on the distance between the plotted position of the measurement item inferred as the cause of the abnormality and the origin.

[0076] (A-7) Diagnosis Section 3

[0077] The third diagnostic unit 6 diagnoses an abnormal pattern involving a measurement item that is a cause of abnormality. That is, the third diagnostic unit 6 constitutes an "abnormal pattern diagnostic unit" for diagnosing an abnormal pattern that exhibits an abnormal phenomenon.

[0078] In this specification, an abnormal mode refers to an abnormal phenomenon that occurs in the device to be diagnosed. For example, in the case of a generator, the abnormal mode may include insulation degradation of the stator winding, interlayer short circuit of the rotor winding, wear of the bearing, etc. In addition, in the abnormal mode, there are both abnormal modes involving a single measurement item and abnormal modes involving multiple measurement items. By determining the abnormal mode by the third diagnostic unit 6, it is possible to determine whether the operation of the device to be diagnosed should be stopped after the abnormality occurs. Alternatively, the part that needs to be repaired or inspected can be determined based on the determined abnormal mode. In this way, effective treatment methods for abnormal phenomena can be studied based on the abnormal mode.

[0079] The third diagnostic unit 6 has a database. In the database, each of the multiple abnormal phenomena pre-conceived and the change vector corresponding to the abnormal phenomenon are associated and registered. In this specification, a "change vector" is a vector representing the change of equipment and machinery from a normal state to an abnormal state, and is composed of elements obtained based on the change of the characteristic quantity before and after the abnormality occurs for each measurement item. The value of each element can be set to the change of any characteristic quantity before and after the abnormality occurs for each measurement item. Alternatively, the value of each element can be set to a value determined based on the change of any characteristic quantity. Alternatively, the value of each element can be set to a value determined based on a combination of the change of any multiple characteristic quantities.

[0080] When the first diagnostic unit 4 determines the abnormality of the diagnostic object equipment and the second diagnostic unit 5 performs the standardization of the multivariate time series data, the selection of the evaluation area and the calculation of the change amount of the feature quantity, the third diagnostic unit 6 generates a change amount vector based on the change amount of the feature quantity. In addition, the third diagnostic unit 6 calculates the similarity between the generated change amount vector and the change amount vectors corresponding to the plurality of abnormal phenomena registered in the database, and determines the abnormal phenomenon associated with the change amount vector with high similarity as an abnormal pattern.

[0081] Specifically, the third diagnosis unit 6 includes a change amount vectorization unit 17 , a database registration unit 18 , a similarity calculation unit 19 , and an abnormal pattern identification unit 20 . Figure 5 This is a flowchart for explaining the diagnosis process of the abnormality mode executed by the third diagnosis unit 6 .

[0082] When the diagnosis process of the abnormal mode starts, the change amount vectorization unit 17 generates a change amount vector at first in step S20. The change amount vector can be generated based on the difference between the feature amount in the first interval before the abnormality of each measurement item and the feature amount in the second interval after the abnormality occurs, that is, the change amount of the feature amount calculated by the feature amount change calculation unit 15 of the second diagnosis unit 5.

[0083] Specifically, the variation vectorization unit 17 obtains the value of the element constituting the variation vector based on the variation of the feature quantity of each measurement item calculated by the feature quantity variation calculation unit 15, and generates the variation vector. The value of each element can be set to the variation of any feature quantity before and after the abnormality of each measurement item occurs. Alternatively, the value of each element can be set to a value obtained by using the variation of any feature quantity of each measurement item and following a classification method of n stages (n is an integer) determined in advance. Alternatively, the value of each element can be set to a value obtained by using a combination of the variation of any plurality of feature quantities of each measurement item and following a classification method of m stages (m is an integer) determined in advance.

[0084] First, the case where the change amount of any feature amount calculated by the feature amount change calculation unit 15 is used as an element of the change amount vector is described. In the following description, the following case is taken as an example: time series data of a total of six types of measurement items, namely, voltage, current, winding temperature, gas temperature, bearing metal temperature, and shaft vibration of the generator are obtained, and abnormality diagnosis and abnormal pattern diagnosis are performed. The case where the average change amount is selected as the change amount of the feature amount is described. When the characteristic quantity change calculation unit 15 obtains that the change in the characteristic quantity (average) of voltage is "-0.5", the change in the characteristic quantity (average) of current is "0.5", the change in the characteristic quantity (average) of winding temperature is "2.0", the change in the characteristic quantity (average) of gas temperature is "1.2", the change in the characteristic quantity (average) of bearing metal temperature is "-0.2", and the change in the characteristic quantity (average) of shaft vibration is "0.6", the change quantity vector can be obtained as (voltage, current, winding temperature, gas temperature, bearing metal temperature, shaft vibration) = (-0.5, 0.5, 2.0, 1.2, -0.2, 0.6). Here, as the change in the characteristic quantity, the average is used as an example, but the change in other characteristic quantities can also be used. Alternatively, the change in multiple characteristic quantities can be used to increase the constituent elements of the vector. Alternatively, the number of change in multiple characteristic quantities can be reduced by principal component analysis, etc.

[0085] Next, the case where the value obtained by using the change amount of any feature amount calculated by the feature amount change calculation unit 15 and the classification method of n stages determined in advance is used as the element of the vector is described. For example, it can be determined to be a value classified into n stages based on a preset threshold value. When the threshold is set to -0.8 and 0.8, the change amount of the feature amount is converted into a discrete value classified into three in such a way that it becomes "-1" when the change amount of the feature amount is less than -0.8, becomes "0" when the change amount of the feature amount is greater than -0.8 and less than 0.8, and becomes "1" when the change amount of the feature amount is greater than 0.8. In this case, the change amount vector can be obtained as (voltage, current, winding temperature, gas temperature, bearing metal temperature, shaft vibration) = (0, 0, 1, 1, 0, 0). The number of classifications n, the threshold value that becomes the basis for classification, and the converted value can be arbitrarily determined. Here, as the change amount of the feature amount, the average is used as an example, but the change amount of other feature amounts can also be used. Alternatively, the variation amounts of a plurality of feature quantities may be used to increase the number of components of the vector. Alternatively, the number of variation amounts of a plurality of feature quantities may be reduced by principal component analysis or the like.

[0086] Next, a case will be described in which a value obtained by using a combination of arbitrary plural feature quantity changes calculated by the feature quantity change calculation unit 15 and following a predetermined m-stage classification method is used as an element of a vector.

[0087] Figure 6 This is a diagram for explaining an example of a method of determining the value of each element of a variation vector when using a combination of variations of an arbitrary plurality of feature quantities and values ​​obtained according to a predetermined m-stage classification method as elements of a vector. Figure 6 is based on Figure 4 The display example of the abnormality cause determination result shown is a diagram created by two-dimensional rendering.

[0088] Specifically, in the two-dimensional plane defined by the variation of the feature quantity X (average) and the variation of the feature quantity Y (standard deviation), the second region RGN2 (corresponding to the region indicating that the measurement item is the cause of the abnormality) is further divided into four regions RGN2_1 to RGN2_4. Region RGN2_1 is a region where the variation of the feature quantity X and the variation of the feature quantity Y are both positive. Region RGN2_2 is a region where the variation of the feature quantity X is positive and the variation of the feature quantity Y is negative. Region RGN2_3 is a region where the variation of the feature quantity X and the variation of the feature quantity Y are both negative. Region RGN2_4 is a region where the variation of the feature quantity X is negative and the variation of the feature quantity Y is positive.

[0089] Arbitrary values ​​are assigned to each of the four regions RGN2_1 to RGN2_4 of the first region RGN1 and the second region RGN2. Figure 6 In the example, "0" is assigned to the first region RGN1, "1" is assigned to the region RGN2_1, "2" is assigned to the region RGN2_2, "3" is assigned to the region RGN2_3, and "4" is assigned to the region RGN2_4 (values ​​indicated by the allocated points in the figure). In this example, the combination of the variation of the two feature quantities, feature quantity X (average) and feature quantity Y (standard deviation), is used to classify into five values. Figure 6 In the example, a classification method is shown in which an area is divided into five and five levels of integers are assigned to each of the five areas, but the number of divisions m of the area and the method of assigning values ​​can be arbitrarily selected. In addition, the change amount of the three feature quantities can be used to classify the area of ​​the three-dimensional space with an arbitrary number of divisions. Regarding the assignment of values, it is also possible to assign meaningless values ​​to the numerical values ​​and their order as in dummy variables.

[0090] like Figure 6 As shown in FIG. 1 , the six measurement items drawn on the two-dimensional plane are converted into values ​​assigned to each area where the drawing exists. Figure 6 In the example of , since the plotted value is in the first region RGN1, the four measurement items of voltage, current, bearing metal temperature, and shaft vibration are converted to "0". On the other hand, since the plotted value is in the region RGN2_1, the two measurement items of gas temperature and winding temperature are converted to "1". In this example, the change vector (voltage, current, winding temperature, gas temperature, bearing metal temperature, shaft vibration) = (0, 0, 1, 1, 0, 0).

[0091] In this way, the variation vectorization unit 17 generates, for each measurement item, a variation vector having as its element the variation of the feature quantity or a value obtained by conversion based on the variation of the feature quantity and a predetermined classification method.

[0092] exist Figure 5 In step S21, the database registration unit 18 creates a database used for determining the abnormality pattern. The database may be implemented in the abnormality diagnosis device 1 or by an external information processing device (eg, a server) that is communicatively connected to the abnormality diagnosis device 1.

[0093] The database registering unit 18 registers each of the plurality of abnormal phenomena in the database by associating it with a change vector corresponding to the abnormal phenomenon. Specifically, the database registering unit 18 can generate a change vector using multivariate time series data obtained before and after a certain abnormal phenomenon occurs, and register the generated change vector in the database by associating it with the abnormal phenomenon.

[0094] Alternatively, the database registration unit 18 also estimates the values ​​of each element of the variation vector to generate a variation vector for an abnormal phenomenon that may occur as a phenomenon but the multivariate time series data before and after the abnormality occurs is not obtained, so that the abnormal phenomenon and the variation vector can be associated and pre-registered in the database. The characteristics and states of the equipment are various for each equipment, so it is difficult to accurately estimate the time series data of each measurement item under the abnormal state. However, it is relatively easy to estimate whether the change of the characteristic quantity (average or standard deviation, etc.) of each measurement item in a certain abnormal phenomenon from the normal state to the abnormal state is no change, positive change, or negative change based on the physical model or past knowledge (that is, estimate the values ​​of each element of the variation vector based on the physical model or past knowledge). Therefore, abnormal phenomena that do not obtain the multivariate time series data before and after the abnormality occurs can also be registered in the database.

[0095] An example of a case where a variation vector of an abnormal phenomenon is estimated based on a physical model or past knowledge and registered in a database is described. For example, it is known that in the case of an abnormality of interlayer short circuit in the rotor coil of a generator, the excitation current and shaft vibration increase and the resistance decreases (that is, the winding temperature decreases). Therefore, the average values ​​of at least two measurement items, the excitation current and shaft vibration, increase, and a variation vector indicating a change in the average value of the winding temperature decreases can be generated and associated with the interlayer short circuit of the rotor coil and registered in the database.

[0096] For other abnormal phenomena, the measurement items and their characteristic quantities and the changes in the characteristic quantities before and after the occurrence of a certain abnormal phenomenon can be inferred based on physical models or past knowledge, so that the change vector can be inferred and the abnormal phenomenon can be associated with the corresponding change vector and registered in the database.

[0097] As another example of a method for generating a variation vector and registering a database for an abnormal phenomenon in which measured multivariate time series data does not exist (has not been obtained), there is the following method: using a variation vector and a database generated based on multivariate time series data actually obtained before and after the occurrence of an abnormality in similar equipment.

[0098] When the generator is used as an example for explanation, the variation vector of abnormal phenomenon a is obtained based on the multivariate time series data before and after the occurrence of abnormal phenomenon a of generator A, and the variation vector of abnormal phenomenon b is obtained based on the multivariate time series data before and after the occurrence of abnormal phenomenon b of generator B. These abnormal phenomena a and b and their corresponding variation vectors can be registered in a database for diagnosing abnormal patterns of other generators C. That is, if they are similar devices, the variation vectors and their abnormal phenomena can be registered in the same database, and the abnormal pattern diagnosis of similar devices can be performed using the database.

[0099] Furthermore, when the variation vectors generated from multivariate time series data of similar but different devices are registered in the same database, the multivariate time series data can be standardized according to different reference values ​​for each device when generating the variation vectors. In addition, with regard to the generation of the variation vectors, it is preferred that the elements (measurement items) constituting the variation vectors and the method for obtaining the values ​​of the elements (the variation of the feature quantity used and the classification method) be the same.

[0100] The absolute values ​​of various measurement items such as voltage, current, and winding temperature of a generator vary depending on the device or the environment in which it is installed. Therefore, in a database that associates the abnormal state itself (e.g., vectorization) with the abnormal phenomenon instead of the change vector indicating the change from normal to abnormal, it is impossible to use the database used in a certain device for the diagnosis of other generators. However, by registering the change vector as a change of feature quantity from normal to abnormal in the database, the device-specific information is excluded, so the change vector and database generated from multivariate time series data obtained from similar devices can be used.

[0101] In this way, by using a method of estimating a variation vector based on a physical model or past knowledge, and a method of generating a variation vector using multivariate time series data before and after an abnormality occurs in a similar device, it is possible to perform diagnosis on devices and machines that rarely experience abnormalities using a database that has a sufficient number of registered abnormality patterns.

[0102] Figure 7 is a diagram showing an example of a database. Figure 7 As shown, in the database, change vectors are registered for each of the plurality of abnormal phenomena a to f. Figure 7 The change vector shown is obtained by Figure 6 The variation vector shown in the example is generated in the same way.

[0103] For example, for abnormal phenomenon a, a change vector (voltage, current, winding temperature, gas temperature, bearing metal temperature, shaft vibration) = (1, 1, 1, 1, 0, 4) is registered. For abnormal phenomenon b, a change vector (voltage, current, winding temperature, gas temperature, bearing metal temperature, shaft vibration) = (0, 0, 1, 1, 0, 0) is registered.

[0104] exist Figure 5 In step S22, the similarity calculation unit 19 compares the similarity values ​​registered in the database ( Figure 7 ) and the unknown abnormal phenomenon generated in step S20, and calculate the similarity of the two change amount vectors. The similarity can be calculated using any known method based on the standard of the elements of the generated change amount vector, such as cosine similarity, Pearson's correlation coefficient, Euclidean distance, Spearman's rank correlation coefficient, Cramer's association coefficient, etc.

[0105] In step S23, the abnormal pattern determination unit 20 determines the abnormal pattern based on the similarity calculated by the similarity calculation unit 19. Specifically, the abnormal pattern determination unit 20 can determine the abnormal phenomenon corresponding to the largest similarity among multiple similarities corresponding to multiple abnormal phenomena registered in the database as the abnormal pattern. Alternatively, the abnormal pattern determination unit 20 can determine the abnormal phenomenon corresponding to the similarity exceeding a preset threshold among multiple similarities as the abnormal pattern. Therefore, when there are more than two similarities exceeding the threshold, more than two abnormal phenomena are determined as candidates for the abnormal pattern.

[0106] (A-8) Output section 3

[0107] In step S24, the third output unit 9 outputs the determination result of the abnormality pattern determination unit 20 of the third diagnosis unit 6. The third output unit 9 has, for example, a display unit, and displays the determination result on the display unit. In addition, the third output unit 9 may be configured to send the determination result to an external device using a communication unit (not shown).

[0108] Figure 8 FIG. 2 is a diagram for explaining an output example of the determination result outputted by the third output unit 9. Figure 8 In the example, the change vector in the unknown abnormal phenomenon is shown relative to the change vector registered in the database (refer to Figure 7 )'s multiple abnormal phenomena a~f...'s similarity of each change vector. Figure 8The table shown may be configured so that a plurality of abnormal phenomena a to f, . . . are arranged in descending order of similarity.

[0109] In addition, when the results of abnormal pattern diagnosis are implemented and the similarity is reduced relative to any abnormal phenomenon registered in the database, it is highly likely that the phenomenon is other than the abnormal phenomenon registered in the database, and the phenomenon and the change vector can be newly registered in the database. In addition, regarding the phenomenon registered in the database, not only abnormal phenomena, but also change vectors can be generated based on multivariate time series data obtained from the equipment before and after any event such as maintenance or inspection, and registered in the database in association with the event name. In this way, by appropriately adding and correcting the database, the database can be further enriched.

[0110] As described above, when the first diagnostic unit 4 determines that the diagnostic target device has an abnormality, the second diagnostic unit 5 and the third diagnostic unit 6 diagnose the cause of the abnormality and the abnormality pattern, respectively.

[0111] [B. Hardware Configuration Example of Abnormality Diagnosis Device]

[0112] Next, a hardware configuration example of the abnormality diagnosis device 1 according to the present embodiment will be described.

[0113] Fig. 9 It is shown Figure 1 A block diagram of an example of a hardware configuration of an abnormality diagnosis device 1. Fig. 9 The abnormality diagnosis device 1 includes a CPU (Central Processing Unit) 100 and a memory for storing programs and data. The CPU 100 operates according to the program, thereby realizing Figure 1 The functional structure shown.

[0114] The memory includes a ROM (Read Only Memory) 101, a RAM (Random Access Memory) 102, and a HDD (Hard Disk Drive) 103. The ROM 101 can store programs executed by the CPU 100. The RAM 102 can temporarily store data used in the execution of the program in the CPU 100, and can function as a temporary data storage device used as a work area. The HDD 103 is a non-volatile storage device that can store read data read by the data reading unit 2, determination results determined by the first diagnostic unit 4, the second diagnostic unit 5, and the third diagnostic unit 6, etc. In addition to or instead of the HDD, a semiconductor storage device such as a flash memory can be used.

[0115] The abnormality diagnosis device 1 further includes a communication interface (I / F) 104, an I / O (Input / Output) interface 105, an input unit 106, and a display unit 107. The communication interface 104 is an interface for the abnormality diagnosis device 1 to communicate with an external device including a diagnostic target device. The communication interface 104 corresponds to an embodiment of the "data reading unit 2".

[0116] The I / O interface 105 is an interface for inputting to the abnormality diagnosis device 1 or outputting from the abnormality diagnosis device 1. Fig. 9 As shown, the I / O interface 105 is connected to the input unit 106 and the display unit 107 .

[0117] The input unit 106 receives input including instructions from the user to the abnormality diagnosis device 1. The input unit 106 includes a keyboard, a mouse, and a touch panel integrally formed with the display screen of the display unit, and receives settings of parameters used in the first diagnosis unit 4, the second diagnosis unit 5, and the third diagnosis unit 6.

[0118] The display unit 107 corresponds to an embodiment of the "data display unit 3", "first output unit 7", "second output unit 8" and "third output unit 9". The display unit 107 can display multivariate time series data obtained from the diagnostic target device, and the determination results determined by the first diagnostic unit 4, the second diagnostic unit 5 and the third diagnostic unit 6.

[0119] In addition, Fig. 9 An example of the hardware structure of the abnormality diagnosis device 1 is shown in the figure, but other structures may also be used. For example, multiple CPUs may be provided. The first diagnosis unit 4, the second diagnosis unit 5, and the third diagnosis unit 6 may be configured as a dedicated circuit such as an ASIC (Application Specific Integrated Circuit), or may be configured using an FPGA (Field Programmable Gate Array). Alternatively, the first diagnosis unit 4, the second diagnosis unit 5, and the third diagnosis unit 6 may be configured by appropriately combining at least two of the CPU, ASIC, and FPGA.

[0120] [C. Effect]

[0121] As described above, according to the abnormality diagnosis device 1 involved in this embodiment, based on the feature quantity extracted from the multivariate time series data of the first interval before the abnormality occurs, the feature quantity extracted from the multivariate time series data of the second interval after the abnormality occurs, and the difference between the feature quantity of the first interval and the feature quantity of the second interval obtained by the same feature quantity extraction method, that is, the change amount of the feature quantity, the measurement item that becomes the cause of the abnormality is diagnosed. In addition, for each measurement item, a change amount vector is generated with the change amount of the feature quantity, the value obtained by using the change amount of the feature quantity and following the classification method of n stages determined in advance, or the value obtained by using the combination of the change amounts of multiple feature quantities with different feature quantity extraction methods and following the classification method of m stages determined in advance. Based on the generated change amount vector and the database in which various abnormal phenomena are registered, an unknown abnormal pattern is diagnosed. By setting it as such a structure, even if there is no professional knowledge and technology related to the diagnostic object equipment, it is possible to diagnose the cause of the abnormality and the abnormal pattern with high accuracy in a short time.

[0122] Implementation method 2.

[0123] In Embodiment 1, a method for diagnosing the cause of abnormality and then diagnosing the abnormal mode when the first diagnostic unit 4 (abnormality diagnostic unit) determines that the abnormality is present is described. Here, a configuration is described in which, when the first diagnostic unit 4 determines that the abnormality is present, the determination of the measurement item as the cause of the abnormality (abnormality cause diagnosis) is not performed but only the abnormal mode is diagnosed.

[0124] First, as shown in the first embodiment, the first diagnostic unit 4 determines whether it is abnormal, and when it is determined that the state is abnormal, the abnormal mode is diagnosed. At this time, the third diagnostic unit 6 (abnormal mode diagnostic unit) can be a structure in which the second diagnostic unit 5 except the abnormal cause determination unit 16 and each function constituting the third diagnostic unit 6 are combined. Specifically, the third diagnostic unit 6 can be configured to include a data normalization unit 13, an evaluation area selection unit 14, a feature quantity change calculation unit 16, a change quantity vectorization unit 17, a database registration unit 18, a similarity calculation unit 19, and an abnormal mode determination unit 20. The calculation processing in each function can be implemented using the same method as in the first embodiment.

[0125] The embodiments disclosed herein are to be understood as illustrative in all points and not restrictive. The scope of the present disclosure is indicated by the claims rather than the above description, and is intended to include all modifications within the meaning and scope equivalent to the claims.

Claims

1. A method for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis method comprises: A step of acquiring multivariate time series data on a plurality of measurement items from the device; A step of diagnosing an abnormality in the operating state of the equipment using the multivariate time series data; as well as When it is determined in the step of diagnosing the abnormality that the operating state of the device is abnormal, a step of diagnosing the cause of the abnormality is performed, The steps for diagnosing the cause of the abnormality include: A step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality occurs; A step of extracting a feature quantity of the second interval by using the at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality occurs; A step of obtaining a change amount of a feature quantity based on a difference between a feature quantity of the first section and a feature quantity of the second section obtained by the same feature quantity extraction method; and a step of diagnosing a measurement item that is a cause of the abnormality based on the amount of change in the characteristic amount with respect to the plurality of measurement items, In the step of diagnosing the abnormality, a learned model serving as a basis for diagnosis is prepared using the multivariate time series data, and the abnormality of the operating state of the device is diagnosed based on the learned model. In the step of extracting the feature quantity of the first interval, the feature quantity of the first interval is extracted from data obtained by normalizing the multivariate time series data of the first interval according to the learned model. In the step of extracting the feature amount of the second interval, the feature amount of the second interval is extracted from data obtained by normalizing the multivariate time series data of the second interval based on the learned model.

2. The abnormality diagnosis method according to claim 1, wherein: The abnormality diagnosis method further comprises: when it is determined in the step of diagnosing the abnormality that the operating state of the device is abnormal, the step of diagnosing an abnormality pattern that exhibits the abnormal phenomenon; The steps of diagnosing the abnormal pattern include: For each measurement item, a step of generating a change amount vector having as its elements a change amount of the feature amount, a value obtained by using the change amount of the feature amount and following a classification method of n stages determined in advance, or a value obtained by using the change amounts of a plurality of the feature amounts having different feature amount extraction methods and following a classification method of m stages determined in advance, wherein n is an integer and m is an integer; a step of creating a database in which each of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon; and A step of diagnosing an unknown abnormality pattern based on the database and the variation vector generated by the generating step.

3. A method for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis method comprises: A step of acquiring multivariate time series data on a plurality of measurement items from the device; A step of diagnosing an abnormality in the operating state of the equipment using the multivariate time series data; as well as When it is determined in the step of diagnosing the abnormality that the operating state of the device is abnormal, a step of diagnosing the cause of the abnormality is performed, The steps for diagnosing the cause of the abnormality include: A step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality occurs; A step of extracting a feature quantity of the second interval by using the at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality occurs; A step of obtaining a change amount of a feature quantity based on a difference between a feature quantity of the first section and a feature quantity of the second section obtained by the same feature quantity extraction method; and a step of diagnosing a measurement item that is a cause of the abnormality based on the amount of change in the characteristic amount with respect to the plurality of measurement items, The abnormality diagnosis method further comprises: when it is determined in the step of diagnosing the abnormality that the operating state of the device is abnormal, the step of diagnosing an abnormality pattern that exhibits the abnormal phenomenon; The steps of diagnosing the abnormal pattern include: For each measurement item, a step of generating a change amount vector having as its elements a change amount of the feature amount, a value obtained by using the change amount of the feature amount and following a classification method of n stages determined in advance, or a value obtained by using the change amounts of a plurality of the feature amounts having different feature amount extraction methods and following a classification method of m stages determined in advance, wherein n is an integer and m is an integer; a step of creating a database in which each of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon; and A step of diagnosing an unknown abnormality pattern based on the database and the variation vector generated by the generating step.

4. A method for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis method comprises: A step of acquiring multivariate time series data on a plurality of measurement items from the device; A step of diagnosing an abnormality in the operating state of the equipment using the multivariate time series data; as well as When the abnormality in the operation state of the device is determined in the step of diagnosing the abnormality, a step of diagnosing an abnormality pattern showing the abnormal phenomenon is performed, The steps of diagnosing the abnormal pattern include: A step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality determination; A step of extracting a feature quantity of the second interval by using the at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality determination; A step of obtaining a change amount of a feature quantity based on a difference between a feature quantity of the first interval and a feature quantity of the second interval obtained by the same feature quantity extraction method; For each measurement item, a step of generating a change amount vector having as its elements a change amount of the feature amount, a value obtained by using the change amount of the feature amount and following a classification method of n stages determined in advance, or a value obtained by using the change amounts of a plurality of the feature amounts having different feature amount extraction methods and following a classification method of m stages determined in advance, wherein n is an integer and m is an integer; a step of creating a database in which each of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon; and A step of diagnosing an unknown abnormality pattern based on the database and the variation vector generated by the generating step.

5. The abnormality diagnosis method according to claim 4, wherein: In the step of diagnosing the abnormality, a learned model serving as a basis for diagnosis is prepared using the multivariate time series data, and the abnormality of the operating state of the device is diagnosed based on the learned model. In the step of extracting the feature quantity of the first interval, the feature quantity of the first interval is extracted from data obtained by normalizing the multivariate time series data of the first interval according to the learned model. In the step of extracting the feature amount of the second interval, the feature amount of the second interval is extracted from data obtained by normalizing the multivariate time series data of the second interval based on the learned model.

6. The abnormality diagnosis method according to any one of claims 2 to 5, wherein: In the step of diagnosing the unknown abnormality pattern, the unknown abnormality pattern is diagnosed based on the similarity between the variation vectors of each of the plurality of abnormal phenomena registered in the database and the variation vector generated in the generating step.

7. The abnormality diagnosis method according to any one of claims 2 to 5, wherein: In the step of creating the database, As a method for generating the variation vector registered in the database, at least one of a method for generating the variation vector using the multivariate time series data before and after the abnormality occurs obtained from the device, a method for generating the variation vector by inferring the variation of the characteristic quantity based on a physical model of the abnormal phenomenon or past knowledge, and a method for generating the variation vector using the multivariate time series data before and after the abnormality occurs obtained from a device similar to the device is used. The generated variation vector is associated with the corresponding abnormal phenomenon and registered in the database.

8. The abnormality diagnosis method according to any one of claims 1 to 5, wherein: The at least one feature quantity extraction method is a process of obtaining at least one of a mean, a variance, a standard deviation, a median, a mode, a maximum value, a minimum value, a skewness, and a kurtosis.

9. An abnormality diagnosis device for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis device comprises: a data reading unit for acquiring multivariate time series data on a plurality of measurement items from the device; an abnormality diagnosis unit that uses the multivariate time series data to diagnose an abnormality in the operating state of the device; as well as an abnormality cause diagnosis unit for diagnosing the cause of the abnormality when the abnormality diagnosis unit determines that the operating state of the device is abnormal; In the abnormality cause diagnosis unit, extracting a feature quantity of the first interval from the multivariate time series data of the first interval before the abnormality occurs by at least one feature quantity extraction method for each measurement item, extracting a feature quantity of the second interval from the multivariate time series data of the second interval after the abnormality occurs, using the at least one feature quantity extraction method for each measurement item, The amount of change in the feature quantity is obtained based on the difference between the feature quantity of the first section and the feature quantity of the second section obtained by the same feature quantity extraction method, diagnosing a measurement item that is a cause of the abnormality based on the amount of change in the characteristic amount with respect to the plurality of measurement items, The abnormality diagnosis unit uses the multivariate time series data to create a learned model that serves as a basis for diagnosis, and diagnoses abnormality in the operating state of the equipment based on the learned model. The abnormal cause diagnosis unit extracts the feature value of the first interval from the data obtained by standardizing the multivariate time series data of the first interval according to the learned model, and extracts the feature value of the second interval from the data obtained by standardizing the multivariate time series data of the second interval according to the learned model.

10. The abnormality diagnosis device according to claim 9, wherein: The abnormality diagnosis device further comprises an abnormality pattern diagnosis unit, which, when the abnormality diagnosis unit diagnoses the abnormality of the device, diagnoses an abnormality pattern that exhibits the abnormal phenomenon. In the abnormal pattern diagnosis unit, For each measurement item, a variation vector is generated, which has as its elements the variation of the feature quantity, the value obtained by using the variation of the feature quantity and following a classification method of n stages determined in advance, or the value obtained by using the variation of a plurality of feature quantities having different feature quantity extraction methods and following a classification method of m stages determined in advance, wherein n is an integer, m is an integer, A database is created in which each abnormal phenomenon of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon. An unknown abnormal pattern is diagnosed based on the database and the variation vector generated by the generating step.

11. An abnormality diagnosis device for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis device comprises: a data reading unit for acquiring multivariate time series data on a plurality of measurement items from the device; an abnormality diagnosis unit that uses the multivariate time series data to diagnose an abnormality in the operating state of the device; as well as an abnormality cause diagnosis unit for diagnosing the cause of the abnormality when the abnormality diagnosis unit determines that the operating state of the device is abnormal; In the abnormality cause diagnosis unit, extracting a feature quantity of the first interval from the multivariate time series data of the first interval before the abnormality occurs by at least one feature quantity extraction method for each measurement item, extracting a feature quantity of the second interval from the multivariate time series data of the second interval after the abnormality occurs, using the at least one feature quantity extraction method for each measurement item, The amount of change in the feature quantity is obtained based on the difference between the feature quantity of the first section and the feature quantity of the second section obtained by the same feature quantity extraction method, diagnosing the measurement item that is the cause of the abnormality based on the amount of change in the characteristic amount with respect to the plurality of measurement items, The abnormality diagnosis device further comprises an abnormality pattern diagnosis unit, which, when the abnormality diagnosis unit diagnoses the abnormality of the device, diagnoses an abnormality pattern that exhibits the abnormal phenomenon. In the abnormal pattern diagnosis unit, For each measurement item, a variation vector is generated, which has as its elements the variation of the feature quantity, the value obtained by using the variation of the feature quantity and following a classification method of n stages determined in advance, or the value obtained by using the variation of a plurality of feature quantities having different feature quantity extraction methods and following a classification method of m stages determined in advance, wherein n is an integer, m is an integer, A database is created in which each abnormal phenomenon of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon. An unknown abnormal pattern is diagnosed based on the database and the variation vector generated by the generating step.

12. An abnormality diagnosis device for diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis device comprises: a data reading unit for acquiring multivariate time series data on a plurality of measurement items from the device; an abnormality diagnosis unit that uses the multivariate time series data to diagnose an abnormality in the operating state of the device; as well as an abnormality pattern diagnosis unit that, when the abnormality diagnosis unit determines that the operating state of the device is abnormal, diagnoses an abnormality pattern that exhibits the abnormal phenomenon; In the abnormal pattern diagnosis unit, extracting a feature quantity of the first interval from the multivariate time series data of the first interval before the abnormality occurs by at least one feature quantity extraction method for each measurement item, extracting a feature quantity of the second interval from the multivariate time series data of the second interval after the abnormality occurs, using the at least one feature quantity extraction method for each measurement item, The amount of change in the feature quantity is obtained based on the difference between the feature quantity of the first section and the feature quantity of the second section obtained by the same feature quantity extraction method, For each measurement item, a variation vector is generated, which has as its elements the variation of the feature quantity, the value obtained by using the variation of the feature quantity and following a classification method of n stages determined in advance, or the value obtained by using the variation of a plurality of feature quantities having different feature quantity extraction methods and following a classification method of m stages determined in advance, wherein n is an integer, m is an integer, A database is created in which each abnormal phenomenon of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon. An unknown abnormal pattern is diagnosed based on the database and the variation vector generated by the generating step.

13. The abnormality diagnosis device according to claim 12, wherein: The abnormality diagnosis unit uses the multivariate time series data to create a learned model that serves as a basis for diagnosis, and diagnoses abnormality in the operating state of the equipment based on the learned model. The abnormal pattern diagnosis unit extracts the feature value of the first interval from the data obtained by standardizing the multivariate time series data of the first interval according to the learned model, and extracts the feature value of the second interval from the data obtained by standardizing the multivariate time series data of the second interval according to the learned model.

14. The abnormality diagnosis device according to any one of claims 10 to 13, wherein: The abnormal pattern diagnosis unit diagnoses the unknown abnormal pattern based on a degree of similarity between the variation vectors of each of the plurality of abnormal phenomena registered in the database and the generated variation vector.

15. The abnormality diagnosis device according to any one of claims 10 to 13, wherein: In the abnormal pattern diagnosis unit, As a method for generating the variation vector registered in the database, at least one of a method for generating the variation vector using the multivariate time series data before and after the abnormality occurs obtained from the device, a method for generating the variation vector by inferring the variation of the characteristic quantity based on a physical model of the abnormal phenomenon or past knowledge, and a method for generating the variation vector using the multivariate time series data before and after the abnormality occurs obtained from a device similar to the device is used. The generated variation vector is associated with the corresponding abnormal phenomenon and registered in the database.

16. The abnormality diagnosis device according to any one of claims 9 to 13, wherein: The at least one feature quantity extraction method is a process of obtaining at least one of a mean, a variance, a standard deviation, a median, a mode, a maximum value, a minimum value, a skewness, and a kurtosis.

17. A computer-readable storage medium having recorded thereon an abnormality diagnosis program for causing a computer to execute a process of diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis program causes the computer to execute: A step of acquiring multivariate time series data on a plurality of measurement items from the device; A step of diagnosing an abnormality in the operating state of the equipment using the multivariate time series data; as well as When it is determined in the step of diagnosing the abnormality that the operating state of the device is abnormal, a step of diagnosing the cause of the abnormality is performed, The steps for diagnosing the cause of the abnormality include: A step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality occurs; A step of extracting a feature quantity of the second interval by using the at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality occurs; A step of obtaining a change amount of a feature quantity based on a difference between a feature quantity of the first section and a feature quantity of the second section obtained by the same feature quantity extraction method; and a step of diagnosing a measurement item that is a cause of the abnormality based on the amount of change in the characteristic amount with respect to the plurality of measurement items, In the step of diagnosing the abnormality, a learned model serving as a basis for diagnosis is prepared using the multivariate time series data, and the abnormality of the operating state of the device is diagnosed based on the learned model. In the step of extracting the feature quantity of the first interval, the feature quantity of the first interval is extracted from data obtained by normalizing the multivariate time series data of the first interval according to the learned model. In the step of extracting the feature amount of the second interval, the feature amount of the second interval is extracted from data obtained by normalizing the multivariate time series data of the second interval based on the learned model.

18. The computer-readable storage medium of claim 17, wherein: The abnormality diagnosis program causes the computer to further execute the steps of: diagnosing an abnormality pattern that exhibits a phenomenon of the abnormality when an abnormality in the operating state of the device is determined in the step of diagnosing the abnormality; The steps of diagnosing the abnormal pattern include: For each measurement item, a step of generating a change amount vector having as its elements a change amount of the feature amount, a value obtained by using the change amount of the feature amount and following a classification method of n stages determined in advance, or a value obtained by using the change amounts of a plurality of the feature amounts having different feature amount extraction methods and following a classification method of m stages determined in advance, wherein n is an integer and m is an integer; a step of creating a database in which each of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon; and A step of diagnosing an unknown abnormality pattern based on the database and the variation vector generated by the generating step.

19. A computer-readable storage medium having recorded thereon an abnormality diagnosis program for causing a computer to execute a process of diagnosing an abnormality of a device to be diagnosed, wherein: The abnormality diagnosis program causes the computer to execute: A step of acquiring multivariate time series data on a plurality of measurement items from the device; A step of diagnosing an abnormality in the operating state of the equipment using the multivariate time series data; as well as When the abnormality in the operation state of the device is determined in the step of diagnosing the abnormality, a step of diagnosing an abnormality pattern showing the abnormal phenomenon is performed, The steps of diagnosing the abnormal pattern include: A step of extracting a feature quantity of the first interval by at least one feature quantity extraction method for each measurement item from the multivariate time series data of the first interval before the abnormality determination; A step of extracting a feature quantity of the second interval by using the at least one feature quantity extraction method for each measurement item from the multivariate time series data of the second interval after the abnormality determination; A step of obtaining a change amount of a feature quantity based on a difference between a feature quantity of the first interval and a feature quantity of the second interval obtained by the same feature quantity extraction method; For each measurement item, a step of generating a change amount vector having as its elements a change amount of the feature amount, a value obtained by using the change amount of the feature amount and following a classification method of n stages determined in advance, or a value obtained by using the change amounts of a plurality of the feature amounts having different feature amount extraction methods and following a classification method of m stages determined in advance, wherein n is an integer and m is an integer; a step of creating a database in which each of a plurality of abnormal phenomena is registered in association with a change amount vector corresponding to the abnormal phenomenon; and A step of diagnosing an unknown abnormality pattern based on the database and the variation vector generated by the generating step.

Images