Condition monitoring system, diagnostic model creation device, and condition monitoring method

Abstract

To more accurately detect abnormality of rotary equipment.SOLUTION: A condition monitoring system (100) includes a vibration sensor (50) for detecting a vibration value of a rotary element in order to monitor a state of the rotary element constituting a wind power generator (101), a power meter (60) for measuring output of the rotary element, and a diagnosis device (103) for diagnosing a state of the rotary element on the basis of a degree of deviation between a reference vibration value which is a vibration value in correspondence with a reference output acquired from the power meter (60) in a state in which the rotary element is normally operated and a temporal variation value which is a vibration value in correspondence with temporal output acquired from the power meter (60) in a state in which the rotary element is continuously operated.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a condition monitoring system for monitoring the condition of a rotating device. [Background technology]

[0002] 2. Description of the Related Art Conventionally, whether or not an abnormality has occurred in a rotating element such as a rotating device in an apparatus has been determined based on the vibration value of the rotating element.

[0003] For example, Patent Document 1 discloses that the relationship between the normal rotation speed and vibration value of a rotating device is approximated by a higher-order function such as a quadratic or higher function, and the vibration value at any rotation speed is corrected to the vibration value at a reference rotation speed, and an abnormality is determined based on this corrected value. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 7-218333 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the inventors of the present application have found that the method disclosed in Patent Document 1, which determines whether an abnormality has occurred in rotating equipment based on the relationship between rotation speed and vibration value, does not allow for proper determination. In particular, the vibration value of the rotating elements of a wind turbine generator is important information for determining damage to the rotating elements, so greater accuracy is required.

[0006] An object of one aspect of the present invention is to more accurately detect abnormalities in rotating equipment. [Means for solving the problem]

[0007] In order to solve the above problems, a condition monitoring system according to one embodiment of the present invention is a condition monitoring system that monitors the condition of rotating elements that constitute a wind power generation device, and includes a detector that detects the vibration value of the rotating element, a measuring instrument that measures the output of the rotating element, and a diagnostic device that diagnoses the condition of the rotating element based on the degree of deviation between a reference vibration value, which is the vibration value corresponding to the reference output obtained from the measuring instrument when the rotating element is operating normally, and a time-dependent vibration value, which is the vibration value corresponding to the time-dependent output obtained from the measuring instrument when the rotating element is operating continuously.

[0008] In order to solve the above problems, a condition monitoring method according to one embodiment of the present invention is a condition monitoring method for monitoring the condition of a rotating element of a wind power generation plant which includes a rotating element, a detector for detecting the vibration value of the rotating element, and a measuring instrument for measuring the output of the rotating element, and which diagnoses the condition of the rotating element based on the degree of deviation between a reference vibration value, which is the vibration value corresponding to the reference output obtained from the measuring instrument when the rotating element is operating normally, and a time-dependent vibration value, which is the vibration value corresponding to the time-dependent output obtained from the measuring instrument when the rotating element is operating continuously. [Effects of the Invention]

[0009] According to one aspect of the present invention, abnormalities in rotating equipment can be detected more accurately. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a partial cross-sectional view showing the configuration of a wind turbine generator that is a target of monitoring by a status monitoring system according to first to fourth embodiments of the present invention. [Figure 2] 1 is a block diagram showing the configuration of a status monitoring system according to a first embodiment of the present invention. [Figure 3] 10 is a graph showing a distribution of vibration values ​​of a rotating element of the wind turbine generator with respect to the output of the wind turbine generator. [Figure 4] 10 is a graph showing the output relative to the rotation speed of the rotating element. [Figure 5] 10 is a graph plotting the change in the vibration value over time for all outputs. [Figure 6] 1 is a graph showing the change in vibration value over time plotted for outputs obtained below a predetermined low output value. [Figure 7] FIG. 4 is a block diagram showing the configuration of a status monitoring system according to a second embodiment of the present invention. [Figure 8] FIG. 8 is a diagram schematically illustrating a diagnostic model used in the condition monitoring system shown in FIG. 7. [Figure 9] FIG. 10 is a block diagram showing the configuration of a status monitoring system according to a third embodiment of the present invention. [Figure 10] FIG. 10 is a diagram showing an autoencoder used as a diagnostic model in the condition monitoring system of FIG. 9. [Figure 11] FIG. 11 is a diagram showing a change over time in reconstruction error that occurs in a state monitoring system according to a modified example of the third embodiment. [Figure 12] FIG. 12 is a diagram showing the results of averaging the reconstruction errors shown in FIG. 11 by day. [Figure 13] FIG. 10 is a block diagram showing the configuration of a status monitoring system according to a fourth embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present invention will be described in detail.

[0012] [Wind power generation equipment] Prior to describing each embodiment, a wind turbine generator that is the target of monitoring by the status monitoring system of each embodiment will be described. Fig. 1 is a partial cross-sectional view showing a wind turbine generator 101. Fig. 3 is a graph showing the distribution of vibration values ​​of the rotating elements of the wind turbine generator 101 relative to the output of the wind turbine generator 101. Fig. 4 is a graph showing the output relative to the rotation speed of the rotating elements.

[0013] As shown in Fig. 1, the wind turbine generator 101 includes a nacelle 10, a tower 20, a hub 30, and a plurality of blades 40. Fig. 1 shows the configuration of the wind turbine generator 101 as viewed from above.

[0014] The nacelle 10 is a housing that houses rotating elements such as a main bearing 11, a main shaft 12, a gearbox 13, a power transmission shaft 14, and a generator 15, as well as other equipment. The nacelle 10 is rotatably mounted on a tower 20. The tower 20 is installed on the ground. When the wind turbine generator 101 is installed offshore, the tower 20 is installed on a foundation formed on the seabed or on a floating body floating on the sea.

[0015] The hub 30 supports the blades 40 rotatably on the nacelle 10. The hub 30 also has a pitch variable mechanism that changes the pitch angle of the blades. The blades 40 are blades that catch wind and rotate the hub 30, converting wind power into rotational torque. The blades 40 are arranged on the hub 30 at predetermined intervals.

[0016] The main bearing 11 rotatably supports the main shaft 12. The main bearing 11 is formed of a rolling bearing.

[0017] The main shaft 12 is a rotating shaft directly connected to the hub 30. The main shaft 12 is connected to the input shaft of the gearbox 13, and transmits the rotational torque generated by the blades 40 to the input shaft of the gearbox 13.

[0018] The speed increaser 13 increases the rotational speed of the main shaft 12 and outputs it. The speed increaser 13 is configured by a gear speed increase mechanism including, for example, a planetary gear, an intermediate shaft, a high-speed shaft, etc. Although not shown, inside the speed increaser 13, a plurality of bearings are provided as rotating elements that rotatably support a plurality of shafts.

[0019] The power transmission shaft 14 is connected to the output shaft of the gearbox 13 and is also connected to the input shaft of the generator 15. The power transmission shaft 14 is a rotating shaft that transmits the output rotation of the gearbox 13 to the input shaft of the generator 15.

[0020] The generator 15 converts the rotational energy transmitted by the power transmission shaft 14 into electrical energy. The generator 15 is configured by, for example, an induction generator. Note that the generator 15 also includes a bearing as a rotating element for rotatably supporting a rotor.

[0021] A control device (not shown) is housed as another device in the nacelle 10. The control device integrates and controls various control units, such as a pitch control unit that controls the pitch variable mechanism described above, a yaw control unit that controls the orientation of the nacelle, and a power control unit that controls power.

[0022] [Embodiment 1] Hereinafter, the first embodiment of the present invention will be described in detail.

[0023] FIG. 2 is a block diagram showing the configuration of the condition monitoring system 100 according to the first embodiment.

[0024] 2 is a system that monitors the state of rotating elements that make up a wind turbine generator 101. The state monitoring system 100 includes a vibration sensor 50 (detector), a power meter 60 (measuring instrument), a data logging device 102, a diagnostic device 103, and a server device 105.

[0025] At least one vibration sensor 50 is provided in the wind turbine generator 101. The vibration sensor 50 detects vibration values ​​of the rotating elements of the wind turbine generator 101, namely, the main bearing 11, the main shaft 12, the step-up gear 13, the power transmission shaft 14, and the generator 15. For this reason, the vibration sensors 50 are disposed at various locations on the above-mentioned rotating elements. For example, the vibration sensors 50 are disposed at a position on the main bearing 11 near the main shaft 12, a position on the step-up gear 13 near the main shaft 12, an upper surface of the step-up gear 13, a position on the step-up gear 13 near the power transmission shaft 14, and a position on the generator 15 near the power transmission shaft 14.

[0026] The wattmeter 60 is provided in the wind turbine generator 101. The wattmeter 60 measures the output of the rotating elements, i.e., the output power of the generator 15. The rotational output (rotational energy) of the main bearing 11, main shaft 12, gearbox 13, and power transmission shaft 14 is transmitted to the generator 15 and ultimately converted into electric power (electrical energy) by the generator 15. In other words, the wattmeter 60 is a measuring instrument that measures the output of the rotating elements (power generation output or output power).

[0027] The data logging device 102 is a device that records and accumulates in time series the output obtained from the power meter 60 and the vibration values ​​obtained from the vibration sensor 50. The data logging device 102 samples the output and the vibration values ​​at a high speed at a predetermined sampling frequency.

[0028] The data logging device 102 accumulates a reference output and a time-dependent output as outputs. The reference output is an output obtained from the power meter 60 when the rotating elements are operating normally. The time-dependent output is an output obtained from the power meter 60 when the rotating elements are operating continuously. The data logging device 102 may be provided in the wind turbine generator 101, or may be installed in a location remote from the wind turbine generator 101.

[0029] The data logging device 102 stores the reference vibration value and the time-dependent vibration value as vibration values. The reference vibration value is the vibration value obtained when a reference output is obtained and is an output corresponding to the reference output. The time-dependent vibration value is the vibration value obtained when a time-dependent output is obtained and is an output corresponding to the time-dependent output.

[0030] The diagnostic device 103 is a device that diagnoses the state of the rotating elements in the wind turbine generator 101 based on the degree of deviation between the reference vibration value and the time-dependent vibration value acquired from the data logging device 102. The diagnostic device 103 has a data processing unit 1, a storage unit 2, a diagnostic unit 4, and a communication unit 5 in order to perform the diagnosis.

[0031] The data processing unit 1 sets an upper limit value for the output so as to obtain a reference vibration value for a reference output and a time-varying vibration value for the time-varying output in an output range in which the output is equal to or less than a predetermined value by performing a predetermined process on the time-varying vibration value obtained from the data logging device 102. The reason for setting an upper limit value for the output in this manner is that, since the wind power used in wind power generation is more stable in winter than in summer, abnormalities in the vibration value can be monitored specifically for vibration values ​​during periods such as summer when the wind power generation device 101 is operated at a relatively low output.

[0032] In addition, the data processing unit 1 performs predetermined processing on the time-dependent vibration values ​​acquired from the data logging device 102, and sets a threshold value of the degree of deviation as a criterion for diagnosing that the condition of the rotating element is not normal by the diagnosing unit 4.

[0033] The data processing unit 1 performs the following processes (1) to (9) to set the upper limit value and the threshold value. The data processing unit 1 stores the set threshold value and upper limit value in the storage unit 2.

[0034] (1) A time-series waveform of vibration acceleration is generated based on the vibration values ​​over time acquired from the data logging device 102.

[0035] (2) Using a high-pass filter, the waveform of high frequency components is extracted, excluding the fundamental rotation frequency component from the time series waveform.

[0036] (3) Absolute value processing is performed on the extracted waveform.

[0037] (4) An envelope (for example, an envelope as shown in FIG. 5, which will be described later) is created by connecting the peak intervals of the waveform that has been subjected to absolute value processing.

[0038] (5) Calculate the effective value of the envelope for each frequency band.

[0039] By obtaining an effective value by filtering in a frequency range corresponding to the rotating elements (shafts, gears, bearings) at the location where the vibration sensor 50 is installed, it becomes possible to grasp the characteristics of each rotating element. Note that the effective value may be the effective value of the vibration velocity other than the effective value of the vibration acceleration.

[0040] The effective value is obtained in a frequency band that includes abnormalities in only the bearing or gear corresponding to the mounting portion of the vibration sensor 50, or in all of these.

[0041] (6) Create a graph with the output on the horizontal axis and the effective value on the vertical axis.

[0042] For one vibration sensor 50, graphs such as those shown in FIG. 3 are created for at least two or more frequencies (for only the bearings or only the gears, or even for frequency bands including both).

[0043] (7) From the graph above, an envelope E in FIG. 3 having peaks and valleys of the effective value is created.

[0044] (8) Set a threshold based on the envelope E.

[0045] (9) Set the upper limit of the output based on the characteristics of the graph above.

[0046] Regarding the threshold, for example, as shown in Fig. 3, it is assumed that normal vibration values ​​are obtained in range R1, and abnormal vibration values ​​are obtained in range R2, which has a minimum value of the maximum value of range R1 and a maximum value twice the minimum value. In this case, the data processing unit 1 sets "2" as the threshold for determining that there is a warning, as the value of the ratio between the dimensionless temporal vibration value V1 and the dimensionless reference vibration value V2, as shown in the following equation. The data processing unit 1 also sets "4" as the threshold for determining that there is a danger, as shown in the following equation.

[0047] V1 / V2≧2…Caution V1 / V2≧4…Danger Furthermore, the threshold value is not limited to the threshold value expressed by the above ratio, and may be expressed by transforming the above ratio relationship into the difference (absolute value of the difference) between the temporal vibration value V1 and the reference vibration value V2. Specifically, as shown in the following formula, if the difference is equal to or greater than the reference vibration value (one time the reference vibration value), it is determined to be a warning, and if the difference is equal to or greater than the reference vibration value (three times the reference vibration value), it is determined to be a danger. The data processing unit 1 sets the multiples of the reference vibration value, "1" and "3", as the threshold values ​​for determining "caution" and "danger", respectively.

[0048] V1-V2≧V2…Caution V1-V2≧3×V2…Danger The threshold value is ultimately determined taking into consideration the operational policy regarding the degree to which detection misses and false positives are tolerated, and information regarding the degree to which deviation from normal data must be large before it can be determined that damage requiring repair or other measures has occurred, etc. This also applies to the threshold values ​​in the following embodiments.

[0049] Furthermore, the data processing unit 1 applies a band-pass filter or the like to acquire vibration values ​​for each frequency band and convert the vibration values ​​so that they more easily reflect abnormalities. For example, the data processing unit 1 may convert the vibration values ​​so as to emphasize vibrations in the several kHz band. Frequencies in the several kHz band correspond to the natural vibration frequency when an abnormality occurs in the bearings of the gearbox 13. This makes it easier to observe the vibration values ​​when an abnormality occurs in the bearings of the gearbox 13.

[0050] Furthermore, the data processing unit 1 uses the upper limit value to remove unnecessary vibration values ​​such as those near peaks on the high output side and extract vibration values ​​on the low output side. In other words, the data processing unit 1 obtains a reference vibration value for a reference output and a time-dependent vibration value for a time-dependent output in an output range in which the output is equal to or less than a predetermined value (upper limit value).

[0051] Furthermore, the data processing unit 1 may acquire a reference vibration value for a reference output and a time-dependent vibration value for a time-dependent output in a range where the rotation speed and output of the rotating element do not have a linear relationship.

[0052] Generally, the wind turbine generator 101 adjusts its output according to its rotation speed. Specifically, when the wind speed exceeds a certain level, the angle (pitch angle) of the blades 40 is changed to rotate at a certain constant rotation speed. For this reason, as shown in Fig. 4, in the wind turbine generator 101, output control causes the output to change even at the same rotation speed (for example, 0.69 and 1.0).

[0053] As described above, the rotation speed and the output are not in a linear relationship over the entire range of the rotation speed. Therefore, unlike general diagnostic devices that perform diagnosis using a linear relationship between the rotation speed and the vibration value, the diagnostic device 103 performs diagnosis based on the above-mentioned vibration value over time acquired by the data processing unit 1.

[0054] Furthermore, the data processing unit 1 creates a table that associates reference outputs with reference vibration values ​​in order to obtain a reference vibration value corresponding to a reference output having the same value as the time-dependent output when diagnosing the state of the rotating element. When a time-dependent output is input, the table outputs a reference vibration value corresponding to a reference output having the same value as the time-dependent output.

[0055] The storage unit 2 stores the thresholds and upper limit values ​​set by the data processing unit 1. The storage unit 2 also stores the table created by the data processing unit 1.

[0056] The output unit 3 calculates the degree of deviation using the vibration value on the low output side extracted by the data processing unit 1. Specifically, the output unit 3 outputs the ratio of the time-dependent vibration value to the reference vibration value, the ratio of the dimensionless time-dependent vibration value to the dimensionless reference vibration value, or the difference between the time-dependent vibration value and the reference vibration value as the degree of deviation.

[0057] The diagnosing unit 4 diagnoses that the state of the rotating element is not normal when the degree of deviation calculated by the output unit 3 is equal to or greater than the threshold value stored in the memory unit 2. Furthermore, the diagnosing unit 4 diagnoses that the state of the rotating element is normal when the degree of deviation is less than the threshold value.

[0058] The communication unit 5 communicates with the data logging device 102 and the server device 105. Specifically, the communication unit 5 outputs the output and vibration values ​​received from the data logging device 102 to the data processing unit 1, and transmits the diagnosis result by the diagnosis unit 4 to the server device 105.

[0059] The server device 105 accumulates the diagnostic results of the diagnostic unit 4 and analyzes the diagnostic results as necessary. Based on the analysis results, a maintenance plan for the rotating elements is formulated.

[0060] Diagnosis of the condition of a rotating element using the condition monitoring system 100 configured as described above will now be described. Fig. 5 is a graph plotting the change in vibration value over time for all outputs. Fig. 6 is a graph plotting the change in vibration value over time for outputs obtained below a predetermined low output value.

[0061] First, the data processing unit 1 of the diagnostic device 103 acquires the reference output and reference vibration value from the data logging device 102 as the mechanical vibration characteristics of the wind turbine generator 101 during a period when the wind turbine generator 101 is in a good operating condition. Based on the acquired reference output and reference vibration value, the data processing unit 1 sets a threshold value and an upper limit value, and creates a table.

[0062] The above-mentioned period is a period during which no failure occurs in the rotating elements of the wind turbine generator 101, and is, for example, a predetermined period (for example, 1 to 3 months) after the wind turbine generator 101 is installed. Depending on the situation, the above-mentioned period may also include a period during which the rotating elements are in an initial failure state.

[0063] During the operational phase, in which the status of the rotating elements is monitored, the data processing unit 1 extracts time-varying outputs below an upper limit value from the time-varying outputs acquired from the data logging device 102. Without limiting vibration values ​​by an upper limit value, as shown in FIG. 5, unnecessary vibration values, such as those on the high-output side, are included in the vibration values ​​across the entire output range. In contrast, if the upper limit value is set to, for example, 37.5% of the rated output, the unnecessary vibration values ​​are removed during period T1, as shown in FIG. 6, and an increase in vibration values ​​that is thought to be due to the progression of damage to the rotating elements can be detected during period T2. Thus, in periods T1 and T2, in a rotating element in which vibration peaks occur at multiple points relative to the output, such as in the drive system of a wind turbine generator 101, in order to detect abnormal vibration values ​​exceeding each of the vibration peaks, the vibration values ​​are plotted within a range of output values ​​in which each vibration peak can be distinguished.

[0064] The data processing unit 1 refers to the table in the memory unit 2 to obtain a reference vibration value corresponding to a reference output having the same value as the extracted time-varying output that is equal to or less than the upper limit value, and provides the reference vibration value to the output unit 3 together with the time-varying vibration value corresponding to the extracted time-varying output.

[0065] The output unit 3 calculates the degree of deviation based on the time-dependent vibration value and the reference vibration value from the data processing unit 1. The output unit 3 provides the calculated degree of deviation to the diagnosis unit 4.

[0066] The diagnosing unit 4 compares the degree of deviation from the output unit 3 with the threshold value stored in the memory unit 2. When the degree of deviation is less than the threshold value, the diagnosing unit 4 diagnoses that the state of the rotating element is normal. When the degree of deviation is equal to or greater than the threshold value, the diagnosing unit 4 diagnoses that the state of the rotating element is abnormal. For example, when the output unit 3 calculates the degree of deviation as the ratio between the dimensionless temporal vibration value and the dimensionless reference vibration value, the diagnosing unit 4 diagnoses that the rotating element is in a state requiring attention if the ratio is equal to or greater than the threshold value "2" described above. In the above case, the diagnosing unit 4 diagnoses that the rotating element is in a dangerous state if the ratio is equal to or greater than the threshold value "2" described above.

[0067] The communication unit 5 transmits the diagnosis result from the diagnosing unit 4 to the server device 105. The server device 105 accumulates the received diagnosis results and analyzes the accumulated diagnosis results as necessary. Furthermore, if the server device 105 obtains a diagnosis result indicating that an abnormality has occurred in the state of the rotating elements, it notifies the management center that manages the wind turbine generator 101.

[0068] As described above, the condition monitoring system 100 according to this embodiment includes the vibration sensor 50, the power meter 60, and the diagnostic device 103.

[0069] Furthermore, the condition monitoring method using the condition monitoring system 100 is a condition monitoring method that monitors the condition of a rotating element of a wind turbine generator 101 that is equipped with a vibration sensor 50 and a power meter 60. The condition monitoring method diagnoses the condition of the rotating element based on the degree of deviation between a reference vibration value, which is a vibration value corresponding to a reference output acquired from the power meter 60, and a time-dependent vibration value, which is a vibration value corresponding to a time-dependent output acquired from the power meter 60.

[0070] With the above configuration, it is possible to diagnose whether or not an abnormality has occurred in the state of the rotating element based on the degree of deviation. Furthermore, the output has less variation in vibration value than the rotation speed, and is more likely to be distributed along a curve. Specifically, as shown in FIG. 4, the output changes at the same rotation speed in a region of rotation speeds around 0.69 of the rated rotation speed (for example, 0.67 to 0.71 of the rated rotation speed). In contrast, the output changes even if the rotation speed is constant as long as it is 0 or higher. Therefore, the vibration value can be monitored and evaluated uniquely with respect to the output. This allows the vibration value to be correctly evaluated. Furthermore, when configuring a diagnostic device using machine learning, as in the second to fourth embodiments described below, the accuracy of diagnosis can be improved.

[0071] The diagnostic device 103 acquires a reference vibration value for a reference output and a time-dependent vibration value for a time-dependent output in a range where the rotation speed of the rotating element and the output are not in a linear relationship. This makes it possible to more accurately represent the distribution of vibration values ​​in a curve, particularly in a range where the rotation speed is constant relative to the output. This allows for more accurate evaluation of vibration values.

[0072] The diagnostic device 103 acquires a reference vibration value for a reference output in an output range where the output is equal to or less than a predetermined value, and a time-dependent vibration value for a time-dependent output. This makes it possible to monitor abnormal vibration values, focusing on vibration values ​​during periods such as summer when the wind power generation plant 101 operates at a relatively low output. Wind power used in wind power generation is more stable in winter than in summer. Therefore, power generation can be focused on winter, and maintenance of rotating elements can be performed in summer.

[0073] The output unit 3 calculates the degree of deviation as the ratio of the time-dependent vibration value to the reference vibration value, the ratio of the dimensionless time-dependent vibration value to the dimensionless reference vibration value, or the difference between the time-dependent vibration value and the reference vibration value. Furthermore, the diagnosis unit 4 diagnoses that the state of the rotating element is abnormal when the degree of deviation is equal to or greater than a predetermined threshold. Thus, by appropriately setting the threshold according to the degree to which the state of the rotating element is abnormal, for example, a state that is not abnormal but requires attention, or a state that is abnormal and dangerous, the state of the rotating element can be properly diagnosed.

[0074] In the condition monitoring system 100, the condition of the rotating elements is diagnosed based on the vibration value relative to the output, but a comprehensive diagnosis may be performed in combination with a diagnosis based on the vibration value relative to the rotation speed of the rotating elements. For example, depending on the relationship between the rotation speed and the output, data that can be determined to be abnormal based on the distribution of the rotation speed and the vibration value may be erroneously determined to be normal based on the distribution of the output and the vibration value. Therefore, by finally determining data that is determined to be abnormal based on either one of the two determination results as abnormal, it is possible to prevent missed detection of abnormalities.

[0075] Alternatively, a variable p representing the operating state of the wind turbine generator 101 may be created by principal component analysis (PCA) based on the relationship between the output and rotational speed, and the state of the rotating element may be diagnosed (anomaly detection) using the relationship between the variable p and the vibration value. In this diagnosis method, first, PCA is performed on two variables, the output value and the rotational speed, and the first principal component is extracted and set as the variable p. This allows the state of the wind turbine generator 101, which is reflected in both the output value and the rotational speed, to be represented without omission by the variable p. Next, similar to a typical anomaly detection method using machine learning, the distribution of the variable p and the vibration value using normal data is learned, and the degree of anomaly of unknown data is calculated based on the learned content.

[0076] [Embodiment 2] A second embodiment of the present invention will be described below. For ease of explanation, components having the same functions as those described in the first embodiment will be denoted by the same reference numerals, and their description will not be repeated. Fig. 7 is a block diagram showing the configuration of a status monitoring system 100A according to the second embodiment. Fig. 8 is a diagram showing a schematic diagram of a diagnostic model M1 used in the status monitoring system 100A.

[0077] 7 is a system that monitors the state of rotating elements that make up a wind turbine generator 101. Like the state monitoring system 100 of the first embodiment described above, the state monitoring system 100A includes a vibration sensor 50, a power meter 60, a data logging device 102, and a server device 105. The state monitoring system 100A also includes a diagnostic device 103A and a diagnostic model creation device 104A.

[0078] The diagnostic model creation device 104A is a device that creates a diagnostic model M1 used in the diagnostic device 103A in the condition monitoring system 100A. The diagnostic model M1 is a learning model that sets the kth (k is any natural number) closest normal value to input data as a reference value from a normal value group that includes only previously prepared reference vibration values ​​and reference outputs corresponding to the reference vibration values ​​as normal values, and outputs the distance between the input data and the reference value. The diagnostic model M1 is based on the so-called KNN (K-Nearest Neighbor).

[0079] The diagnostic model creation device 104A includes a data processing unit 6, a setting unit 7A, a model creation unit 8A, and a storage unit 9A.

[0080] The data processing unit 6, like the data processing unit 1 included in the condition monitoring system 100, extracts low-output vibration values ​​from the vibration values ​​acquired from the data logging device 102. Also, like the data processing unit 1, the data processing unit 6 applies a band-pass filter or the like to acquire vibration values ​​for each frequency band, and converts the vibration values ​​so that they more easily reflect abnormalities.

[0081] The setting unit 7A sets a predetermined quantile in the normal value group as the threshold value, and stores the set threshold value in the storage unit 9A.

[0082] The model creation unit 8A generates a diagnostic model M1. Specifically, the model creation unit 8A acquires a reference vibration value, which is a normal value, and a reference output corresponding to the reference vibration value to form a normal value group. When a time-varying vibration value and a time-varying output corresponding to the time-varying vibration value are input as input data, the model creation unit 8A sets the kth (k is an arbitrary natural number) closest normal value to the input data as a reference value and creates the diagnostic model M1 so as to calculate the distance between the input data and the reference value.

[0083] Diagnostic device 103A is a device that diagnoses the state of rotating elements in wind turbine generator 101 based on the degree of deviation between reference vibration values ​​and time-dependent vibration values ​​acquired from data logging device 102. In order to perform diagnosis, diagnostic device 103A has data processing unit 1 and communication unit 5 that diagnostic device 103 described above has, as well as a memory unit 2A, an output unit 3A, and a diagnostic unit 4A. Unlike data processing unit 1 in diagnostic device 103, data processing unit 1 in diagnostic device 103A does not set a threshold value.

[0084] The storage unit 2A stores the threshold set by the setting unit 7A and the diagnostic model M1 trained by the diagnostic model generation device 104A.

[0085] The output unit 3A inputs the temporal vibration value and the temporal output corresponding to the temporal vibration value as input data to the diagnostic model M1 in the storage unit 2A, and outputs the distance from the reference value output from the diagnostic model M1 as the degree of deviation.

[0086] The diagnostic unit 4A diagnoses that the state of the rotating element is abnormal when the deviation degree output by the output unit 3A is equal to or greater than a predetermined threshold value, and diagnoses that the state of the rotating element is normal when the deviation degree is less than the threshold value.

[0087] Diagnosis of the state of the rotating element by the state monitoring system 100A configured as above will be described.

[0088] First, in the diagnostic model creation device 104A, the diagnostic model M1 is created and thresholds are set during the period when the operating state of the wind turbine generator 101 is good as described above.

[0089] The setting unit 7A sets a predetermined quantile point in the normal value group as the threshold. The setting unit 7A may set the quantile point using a known method. For example, the setting unit 7A sets the quantile point within a range in which 95% or 99% of the normal values ​​are distributed.

[0090] Furthermore, the model creation unit 8A acquires a plurality of pairs of reference output and reference vibration value as normal values ​​from the data logging device 102 via the data processing unit 6, uses these pairs as input data, and forms a normal value group using the normal values. As shown in FIG. 8, when creating the diagnostic model M1, the model creation unit 8A sets a tentative time-course vibration value and a time-course output corresponding to the time-course vibration value as input data D to be diagnosed. The model creation unit 8A creates the diagnostic model M1, which sets the normal value that is the kth closest to the input data D from the normal value group as a reference value Vr and outputs the distance L between the input data and the reference value Vr. The created diagnostic model M1 is downloaded to the diagnostic device 103A and stored in the storage unit 2A, thereby being implemented in the diagnostic device 103A.

[0091] In Figure 8, input data D located close to the normal value group is represented by "D1," and input data D located far from the normal value group is represented by "D2." The reference value Vr for input data D1 is represented by "Vr1," and the reference value Vr for input data D2 is represented by "Vr2." The distance L for input data D1 is represented by "L1," and the distance L for input data D2 is represented by "L2." Figure 8 also shows the case where k=5.

[0092] During the operational phase of monitoring the state of the rotating elements, the data processing unit 1A extracts the time-dependent outputs below an upper limit value from the time-dependent outputs obtained from the data logging device 102, and provides the time-dependent outputs together with the time-dependent vibration values ​​corresponding to the time-dependent outputs to the output unit 3A.

[0093] The output unit 3A inputs the time-dependent output and time-dependent vibration values ​​as input data to the diagnostic model M1 in the storage unit 2A. The diagnostic model M1 sets the kth closest normal value in the normal value group to the input data as a reference value, and outputs the distance between the reference value and the input data. The output unit 3A outputs the distance as a deviation degree.

[0094] 8, when k is set to 5, if input data D1 located close to the normal value group is input, the diagnostic model M1 sets the normal value in the normal value group that is fifth closest to the input data D1 as the reference value Vr1 and outputs the distance L1 between the reference value Vr1 and the input data D1. On the other hand, if input data D2 located far from the normal value group is input, the diagnostic model M1 sets the normal value in the normal value group that is fifth closest to the input data D2 as the reference value Vr2 and outputs the distance L2 between the reference value Vr2 and the input data D2.

[0095] The diagnostic unit 4A compares the degree of deviation from the output unit 3A with the threshold value in the memory unit 2A. When the degree of deviation is less than the threshold value, the diagnostic unit 4A diagnoses that the state of the rotation element is normal. When the degree of deviation is equal to or greater than the threshold value, the diagnostic unit 4A diagnoses that the state of the rotation element is abnormal. In the above example, if the diagnostic unit 4A determines that the distance L1 is less than the threshold value, it diagnoses the input data D1 as normal data, and if the distance L2 is equal to or greater than the threshold value, it diagnoses the input data D2 as abnormal data.

[0096] As described above, the condition monitoring system 100A according to this embodiment includes the vibration sensor 50, the power meter 60, and the diagnostic device 103A.

[0097] According to the above configuration, the state of the rotating element can be diagnosed using a simple algorithm.

[0098] The condition monitoring system 100A also includes a diagnostic model creation device 104A. This allows a normal value group to be prepared in advance based on normal input data. Therefore, a simple diagnostic model can be created based on the normal value group.

[0099] [Embodiment 3] A third embodiment of the present invention will be described below. For ease of explanation, components having the same functions as those described in the first and second embodiments will be denoted by the same reference numerals, and their description will not be repeated. Fig. 9 is a block diagram showing the configuration of a condition monitoring system 100B according to the third embodiment. Fig. 10 is a diagram showing an autoencoder used as a diagnostic model in the condition monitoring system 100B.

[0100] 9 is a system that monitors the state of rotating elements that make up a wind turbine generator 101. Like the state monitoring system 100 of the first embodiment described above, the state monitoring system 100B includes a vibration sensor 50, a power meter 60, a data logging device 102, and a server device 105. The state monitoring system 100B also includes a diagnostic device 103B and a diagnostic model creation device 104B.

[0101] The diagnostic model creation device 104B is a device that creates a diagnostic model M2 used in the diagnostic device 103B in the condition monitoring system 100B. The diagnostic model M2 is a learning model that extracts features by compressing input data and restores the features as output data. The diagnostic model M2 is configured by a so-called autoencoder as shown in FIG. 10. The autoencoder uses an encoder (input layer) to reduce the dimensions of the input data and convert it into latent variables, which are compressed features, and a decoder (output layer) to restore (reconstruct) the latent variables to the input data and output it.

[0102] The diagnostic model creation device 104B includes a data processing unit 6, a setting unit 7B, a model creation unit 8B, and a storage unit 9B.

[0103] The model creation unit 8B creates a diagnostic model M2. Specifically, the model creation unit 8B creates a diagnostic model M2 in which the weights of the encoder and the decoder are adjusted based on normal input data so that the difference (reconstruction error) between the input data and the output data is minimized. As the normal input data, a reference vibration value and a reference output corresponding to the reference vibration value are used.

[0104] The setting unit 7B sets the threshold value based on the minimum difference obtained from the model creating unit 8B. The setting unit 7B stores the set threshold value in the storage unit 9B.

[0105] The diagnostic device 103B is a device that diagnoses the state of the rotating elements in the wind turbine generator 101 based on the degree of deviation between the reference vibration value and the time-dependent vibration value acquired from the data logging device 102. In order to perform diagnosis, the diagnostic device 103B has the data processing unit 1 and communication unit 5 that the diagnostic device 103A described above has, as well as a memory unit 2B, an output unit 3B, and a diagnostic unit 4B.

[0106] The storage unit 2B stores the thresholds set by the diagnostic model generation device 104B, and also stores the diagnostic model M2 generated by the diagnostic model generation device 104B.

[0107] The output unit 3B inputs the time-dependent vibration value and the time-dependent output corresponding to the time-dependent vibration value as input data to the diagnostic model M2 in the memory unit 2B, and outputs the difference between the input data and the output data output from the diagnostic model M2 as the degree of deviation.

[0108] The diagnostic unit 4B diagnoses that the state of the rotating element is abnormal when the deviation degree output from the output unit 3B is equal to or greater than a predetermined threshold value, and diagnoses that the state of the rotating element is normal when the deviation degree is less than the threshold value.

[0109] Diagnosis of the state of the rotating element by the state monitoring system 100B configured as above will be described.

[0110] First, in the diagnostic model creation device 104B, the diagnostic model M2 is created and thresholds are set during the period when the operating state of the wind turbine generator 101 is good as described above.

[0111] The model creation unit 8B creates a diagnostic model M2 by adjusting the first weight for compressing the input data and the second weight for restoring the above features to the output data so that the difference between the input data and the output data is minimized. As input data, a normal reference vibration value and a reference output corresponding to the normal reference vibration value are used. The first weight is the encoder weight, and the second weight is the decoder weight.

[0112] The setting unit 7B also sets a threshold based on the minimum difference. For example, the setting unit 7B sets a value slightly larger than the difference (a value obtained by adding a margin to the difference) as the threshold. Alternatively, the setting unit 7B may set a predetermined quantile (95%, 99%, etc.) of normal data used as input data when creating the diagnostic model M2 as the threshold.

[0113] On the other hand, in the diagnostic device 103B, the output unit 3B inputs, as input data, the temporal vibration values ​​and the temporal outputs corresponding to the temporal vibration values, which are acquired from the data logging device 102 via the data processing unit 1, to the diagnostic model M2 in the storage unit 2B. The output unit 3B acquires the difference between the input data and the output data output from the diagnostic model M2, and outputs it as a deviation degree.

[0114] The diagnostic unit 4B compares the degree of deviation from the output unit 3B with a threshold value stored in the memory unit 2B. When the degree of deviation is less than the threshold value, the diagnostic unit 4B diagnoses that the state of the rotating element is normal. When the degree of deviation is equal to or greater than the threshold value, the diagnostic unit 4B diagnoses that the state of the rotating element is abnormal.

[0115] The diagnostic model M2, which has been trained only with normal input data, restores the compressed features when normal input data is input during diagnosis, and by correctly restoring the features, outputs output data that is close to the normal input data. In this case, the deviation degree output from the output unit 3B is less than the threshold, so the state of the rotation element is diagnosed as normal.

[0116] On the other hand, when abnormal input data is input to the diagnostic model M2 during diagnosis, the compressed features cannot be correctly restored, resulting in a large difference between the input data and the output data. In this case, if the degree of deviation output from the output unit 3B is equal to or greater than a threshold, the state of the rotation element is diagnosed as abnormal.

[0117] As described above, the condition monitoring system 100B according to this embodiment includes the vibration sensor 50, the power meter 60, and the diagnostic device 103B.

[0118] According to the above configuration, the diagnostic model M2 is trained to minimize the difference between the input and output data. Therefore, when abnormal input data is input, the difference between the input and output data, i.e., the degree of deviation, increases and exceeds the threshold, thereby improving the accuracy of the diagnosis.

[0119] The condition monitoring system 100B also includes a diagnostic model creation device 104B, which learns the first and second weights so as to minimize the difference between the input and output data, thereby enabling the creation of a highly accurate diagnostic model.

[0120] Next, a modified example of this embodiment will be described. Fig. 11 is a diagram showing changes over time in the reconstruction error that occurred in the status monitoring system according to this modified example. Fig. 12 is a diagram showing the results of averaging the reconstruction error shown in Fig. 11 by day.

[0121] In a condition monitoring system 100B according to this modification, a diagnostic model creation device 104B uses CNN-VAE as a diagnostic model M2.

[0122] CNN-VAE is an autoencoder that uses a convolutional neural network for the encoder and decoder. A convolutional neural network is a neural network model that adds a convolutional layer that captures the features of the input data and a pooling layer that reduces bias toward those features between the input and output layers. A variational autoencoder (VAE) is a model that introduces a probability distribution (mean μ and variance σ of an n-dimensional Gaussian distribution) into the latent variables of the autoencoder.

[0123] The model creation unit 8B of the diagnostic model creation device 104B creates a diagnostic model M2 that has learned from data of various driving conditions in a normal state. By using data of various driving conditions in the learning, the diagnostic model M2 can capture state changes while taking into account the diversity of normal data that occurs due to changes in driving conditions.

[0124] In the condition monitoring system 100B, the diagnostic device 103B restores the acquired input data using the trained diagnostic model M2. At this time, if the input data contains features that are not included in the trained normal data, the diagnostic model M2 cannot accurately restore the input data. Therefore, the diagnostic device 103B averages the absolute values ​​of the differences between the input data and the restored output data, and uses this average value as the difference between the input data and the output data to diagnose the condition of the rotating element. In addition, as a post-processing step, the variance of the difference can be reduced by averaging the difference (reconstruction error) for each day.

[0125] The input data and the reconstruction error of the input data were compared and verified for the rotating element states at the time of diagnosis: normal, predictive, and abnormal. A predictive state is a transition state from normal to abnormal; it is not an abnormality, but is an unusual state with a high possibility of progressing to an abnormality. When diagnosed as normal, the input data was largely restored. However, when diagnosed as predictive, the periodic increase seen in vibration values ​​in the several kHz range caused by bearing damage was not restored to the output data. Furthermore, when diagnosed as abnormal, the overall increase in vibration values ​​was not restored to the output data either.

[0126] We examined the transition of the reconstruction error, which is the average of the absolute value of the difference between the input data and output data during diagnosis. As shown in the upper graph of Figure 11, the reconstruction error first became very large during period P3, when the damage had progressed significantly and an abnormality was diagnosed. Furthermore, as shown in the lower graph of Figure 11, which focuses on normal and predictive data, we confirmed that the change in the reconstruction error was small during period P1, when the data was diagnosed as normal, but gradually increased during period P2, when the data was diagnosed as predictive. However, because the reconstruction error has a large variance, it is difficult to perform abnormality detection using a threshold value as is. Therefore, we averaged the reconstruction error daily as post-processing. As shown in the upper and lower graphs of Figure 12, the averaging process reduced the variance, and we were able to capture clear trend changes for the normal data (period P1) and predictive data (period P2) at a level that allowed for threshold detection.

[0127] [Embodiment 4] A fourth embodiment of the present invention will be described below. For ease of explanation, components having the same functions as those described in the first to third embodiments will be denoted by the same reference numerals, and the description thereof will not be repeated. Fig. 13 is a block diagram showing the configuration of a status monitoring system 100C according to the fourth embodiment.

[0128] 13 is a system that monitors the state of rotating elements that make up a wind turbine generator 101. Like the state monitoring system 100 of the first embodiment described above, the state monitoring system 100C includes a vibration sensor 50, a power meter 60, and a data logging device 102. The state monitoring system 100C also includes a diagnostic device 103C and a diagnostic model creation device 104C.

[0129] The diagnostic model creation device 104C is a device that creates a diagnostic model M3 used in the diagnostic device 103C in the condition monitoring system. The diagnostic model M3 extracts features by compressing reference vibration values ​​and reference outputs corresponding to the reference vibration values ​​as input data, restores the features as output data, and forms a first distribution of the differences between the input data and the output data and the features. The diagnostic model M3 also outputs the distance between the first distribution and a second distribution of the differences and features when a time-varying vibration value and a time-varying output corresponding to the time-varying vibration value are input. The diagnostic model M3 is configured using a so-called DAGMM (Deep Autoencoding Gaussian Mixture Model).

[0130] The DAGMM consists of a compression network, an estimation network, and a GMM. The compression network is composed of an autoencoder, and outputs features compressed by the autoencoder's encoder and the difference (reconstruction error) between the input data input to the autoencoder and the output data output from the autoencoder. The estimation network estimates the probability (degree of membership) of which normal distribution the input data belongs to in the GMM, based on the features from the compression network and the difference. The GMM estimates the parameters (mean and variance) of a Gaussian mixture model based on the features from the compression network and the degree of membership from the estimation network.

[0131] The diagnostic model creation device 104C includes a data processing unit 6, a setting unit 7C, a model creation unit 8C, and a storage unit 9C.

[0132] The model creation unit 8C creates a diagnostic model M3. Specifically, the model creation unit 8C inputs normal input data and forms a distribution (first distribution) between the difference between the input data and output data and the above-mentioned features. A reference vibration value and a reference output corresponding to the reference vibration value are used as the normal input data. In addition, the model creation unit 8C creates the diagnostic model M3 so as to output a distribution (second distribution) between the difference and the above-mentioned features when a time-dependent vibration value and a time-dependent output corresponding to the time-dependent vibration value are input, and the distance from the first distribution.

[0133] The setting unit 7C sets a threshold value based on the first distribution acquired from the model creating unit 8C. The setting unit 7C stores the set threshold value in the storage unit 9C.

[0134] The diagnostic device 103C is a device that diagnoses the state of the rotating elements in the wind turbine generator 101 based on the degree of deviation between the reference vibration value and the time-dependent vibration value acquired from the data logging device 102. In order to perform diagnosis, the diagnostic device 103C has the data processing unit 1 and communication unit 5 that the diagnostic device 103A described above has, as well as a memory unit 2C, an output unit 3C, and a diagnostic unit 4C.

[0135] The storage unit 2C stores the threshold value set by the diagnostic model generation device 104C, and also stores the diagnostic model M3 generated by the diagnostic model generation device 104C.

[0136] The output unit 3C outputs the distance between the first distribution and the second distribution as a degree of deviation to the diagnostic model M3, which is obtained by inputting the time-varying vibration values ​​acquired from the data logging device 102 and the time-varying outputs corresponding to the time-varying vibration values ​​as input data.

[0137] The diagnostic unit 4C diagnoses that the state of the rotating element is abnormal when the deviation degree calculated by the output unit 3C is equal to or greater than a predetermined threshold value, and diagnoses that the state of the rotating element is normal when the deviation degree is less than the threshold value.

[0138] Diagnosis of the state of the rotating element by the state monitoring system 100C configured as above will be described.

[0139] First, in the diagnostic model creation device 104C, a diagnostic model M3 is created and thresholds are set during the period when the operating state of the wind turbine generator 101 is good as described above.

[0140] The model creation unit 8C creates a diagnostic model M3 so as to receive as input data the normal reference vibration value and the reference output corresponding to the normal reference vibration value, and to output the distance between the first distribution and the second distribution.

[0141] During training, the diagnostic model M3 estimates the probability that the input data will be classified into each normal distribution of the GMM in the DAGMM estimation network based on the above differences and the above features as a degree of membership. Also, during training, the diagnostic model M3 obtains a first distribution of the differences and features as GMM parameters based on the above membership, the compressed features from the autoencoder, and the reconstruction error using the GMM of the DAGMM.

[0142] The setting unit 7C sets the threshold based on the first distribution. For example, the setting unit 7C sets a predetermined quantile in the first distribution as the threshold. Specifically, the setting unit 7C sets the quantile in a range in which 95% or 99% of normal values ​​are distributed. The setting unit 7C may also set the threshold based on other known methods.

[0143] On the other hand, in the diagnostic device 103C, the output unit 3C inputs, as input data, the temporal vibration values ​​and the temporal outputs corresponding to the temporal vibration values, which are acquired from the data logging device 102 via the data processing unit 1, to the diagnostic model M3 in the storage unit 2C. The output unit 3C outputs, as the degree of deviation, the distance between the first distribution and the second distribution output from the diagnostic model M3.

[0144] The diagnostic unit 4C compares the degree of deviation from the output unit 3C with a threshold value stored in the memory unit 2C. When the degree of deviation is less than the threshold value, the diagnostic unit 4C diagnoses that the state of the rotating element is normal. When the degree of deviation is equal to or greater than the threshold value, the diagnostic unit 4C diagnoses that the state of the rotating element is abnormal.

[0145] As described above, the condition monitoring system 100C according to this embodiment includes the vibration sensor 50, the power meter 60, and the diagnostic device 103C.

[0146] According to the above configuration, when input data whose values ​​are significantly different from the normal input data used for learning is input to diagnostic device 103C, the features obtained by compression differ from the features used during learning. As a result, the input data cannot be correctly restored, and the difference between the input data and the output data becomes larger than the difference during learning. Therefore, the second distribution of the difference and the features exists at a large distance from the first distribution. Therefore, it is possible to diagnose that the state of the rotation element is abnormal based on the distance between the first distribution and the second distribution. In this way, the features obtained by compressing the input data and the distribution of the difference between the input data and the output data are used as diagnostic elements, thereby improving the accuracy of the diagnosis.

[0147] The condition monitoring system 100C also includes a diagnostic model creation device 104C, which allows the first distribution to be learned based on normal input data, making it possible to create a highly accurate diagnostic model without requiring training data.

[0148] [Contribution to SDGs] As described above, according to the configurations of the respective embodiments, the state of the rotating elements can be diagnosed more accurately than with conventional diagnostic methods. This improves the maintainability of the wind turbine generator 101. Therefore, it becomes possible to operate the wind turbine generator 101 for a longer period of time. This contributes to achieving Goal 7 of the Sustainable Development Goals (SDGs), "Affordable and clean energy (renewable energy, etc.)."

[0149] [Software implementation example] The functions of diagnostic devices 103, 103A-103C (hereinafter referred to as "diagnostic devices") and diagnostic model creation devices 104A-104C (hereinafter referred to as "creation devices") are realized by programs that cause a computer to function as the diagnostic devices and the creation devices. Furthermore, the functions of the diagnostic devices and the creation devices can be realized by programs that cause a computer to function as each control block of the diagnostic devices and the creation devices. The control blocks correspond to, in particular, output units 3, 3A-3C, diagnosis units 4, 4A-4C, setting units 7A-7C, and model creation units 8A-8C.

[0150] In this case, the diagnostic device and the creation device each include a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in the first to fourth embodiments are realized by executing the program using the control device and storage device.

[0151] The program may be stored non-transitory on one or more computer-readable storage media. The storage media may or may not be included in the diagnostic device and the creation device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0152] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0153] In particular, each process described in the second to fourth embodiments is executed by AI (Artificial Intelligence). In this case, the AI ​​may be operated by the control device or another device (for example, an edge computer or a cloud server).

[0154] [Additional Notes] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Furthermore, embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention. [Explanation of symbols]

[0155] 3,3A,3B,3C output section 4,4A,4B,4C Diagnosis section 7,7A,7B,7C Setting section 8A, 8B, 8C Model Creation Section 11 Main bearing (rotating element) 12 Main axis (rotating element) 13 Speed ​​increaser (rotating element) 14 Power transmission shaft (rotating element) 15 Generator (rotating element) 50 Vibration sensor (detector) 60 Power meter (measuring instrument) 100, 100A, 100B, 100C Condition Monitoring System 101 Wind power generation equipment 103, 103A, 103B, 103C Diagnostic equipment 104A, 104B, 104C Diagnostic model creation device M1~M3 diagnostic models

Claims

1. A condition monitoring system for monitoring the condition of a rotating element constituting a wind turbine generator, a detector for detecting a vibration value of the rotating element; a measuring instrument for measuring the output of the rotating element; a diagnostic device that diagnoses the condition of the rotating element based on the degree of deviation between a reference vibration value, which is the vibration value corresponding to a reference output obtained from the measuring instrument when the rotating element is operating normally and in a range where the rotational speed of the rotating element is constant relative to the output, and a time-dependent vibration value, which is the vibration value corresponding to a time-dependent output obtained from the measuring instrument when the rotating element is operating continuously and in a range where the rotational speed of the rotating element is constant relative to the output.

2. The condition monitoring system according to claim 1 , wherein the diagnostic device acquires the reference vibration value for the reference output and the time-dependent vibration value for the time-dependent output in an output range in which the output is equal to or less than a predetermined value.

3. the degree of deviation is a ratio between the time-dependent vibration value and the reference vibration value, a ratio between the dimensionless time-dependent vibration value and the dimensionless reference vibration value, or a difference between the time-dependent vibration value and the reference vibration value, 3. The condition monitoring system according to claim 2, wherein the diagnostic device diagnoses that the condition of the rotating element is not normal when the degree of deviation is equal to or greater than a predetermined threshold value.

4. The diagnostic device comprises: an output unit that inputs the time-dependent vibration value acquired from the detector and the time-dependent output corresponding to the time-dependent vibration value into a diagnostic model that is trained to output a distance between input data and the kth (k is any natural number) normal value closest to input data from a normal value group prepared in advance and including only a reference vibration value and the reference output corresponding to the reference vibration value as normal values, and outputs the distance output from the diagnostic model as the degree of deviation; 3. The condition monitoring system according to claim 2, further comprising a diagnostic unit that diagnoses that the condition of the rotating element is not normal when the degree of deviation is equal to or greater than a predetermined threshold value.

5. 5. A diagnostic model creation device for creating the diagnostic model used in the diagnostic device in the condition monitoring system according to claim 4, comprising: a model creation unit that creates the diagnostic model using the reference vibration value and the reference output corresponding to the reference vibration value as input data and outputs the distance; a setting unit that sets a predetermined quantile in the normal value group as the threshold.

6. The diagnostic device comprises: an output unit that inputs the time-dependent vibration value acquired from the detector and the time-dependent output corresponding to the time-dependent vibration value as input data to the diagnostic model that extracts features by compressing input data and restores the features as output data, the diagnostic model being machine-learned based on normal input data so as to minimize a difference between the input data and the output data, and outputs a difference between the input data and the output data output from the diagnostic model as the degree of deviation; 3. The condition monitoring system according to claim 2, wherein when the degree of deviation is equal to or greater than a predetermined threshold, the condition of the rotating element is diagnosed as abnormal.

7. 7. A diagnostic model creation device for creating the diagnostic model used in the diagnostic device in the condition monitoring system according to claim 6, comprising: a model creation unit that creates the diagnostic model by using the normal vibration value over time and the output over time corresponding to the vibration value over time as input data, and adjusting a first weight for compressing the input data and a second weight for restoring the features to the output data so as to minimize a difference between the input data and the output data; a setting unit that sets the threshold based on the minimum difference.

8. The diagnostic device comprises: an output unit that extracts features by compressing the reference vibration value and the reference output corresponding to the reference vibration value as input data, restores the features as output data, forms a first distribution of the difference between the input data and the output data and the features, and outputs the distance between a second distribution of the difference and the features when the time-varying vibration value and the time-varying output corresponding to the time-varying vibration value are input and the first distribution as the deviation degree; 3. The condition monitoring system according to claim 2, further comprising a diagnostic unit that diagnoses that the condition of the rotating element is not normal when the degree of deviation is equal to or greater than a predetermined threshold value.

9. 9. A diagnostic model creation device for creating the diagnostic model used in the diagnostic device in the condition monitoring system according to claim 8, comprising: a model creation unit that creates the diagnostic model using the reference vibration value and the reference output corresponding to the reference vibration value as input data and outputs the distance; a setting unit that sets the threshold value based on the first distribution.

10. A state monitoring method for monitoring a state of a rotating element of a wind turbine generator including a rotating element, a detector for detecting a vibration value of the rotating element, and a measuring instrument for measuring an output of the rotating element, comprising: A condition monitoring method characterized by diagnosing the condition of the rotating element based on the degree of deviation between a reference vibration value, which is the vibration value corresponding to a reference output obtained from the measuring instrument when the rotating element is operating normally and in a range where the rotational speed of the rotating element is constant relative to the output, and a time-dependent vibration value, which is the vibration value corresponding to a time-dependent output obtained from the measuring instrument when the rotating element is operating continuously and in a range where the rotational speed of the rotating element is constant relative to the output.

Images