Machine learning device and anomaly estimation device

Abstract

To provide a machine learning device and an anomaly estimation device that can build a learning model for more quickly estimating the presence or absence of various anomalies and the causes of any anomalies that have occurred. [Solution] The machine learning device 1 includes an input data acquisition unit 11 that acquires a data set including at least vibration waveforms detected by a vibration sensor 31 placed on the wind power generation device 5 as input data; a label acquisition unit 12 that acquires the presence or absence of an abnormality in the wind power generation device 5 and the cause of the abnormality if an abnormality occurs as labels; and a learning model construction unit 14 that constructs a learning model for estimating the presence or absence of an abnormality in the wind power generation device 5 and the cause of the abnormality by performing supervised learning using the input data and label pair as training data.

Description

Technical Field

[0001] The present invention relates to a machine learning device and an abnormality estimation device.

Background Art

[0002] Conventionally, techniques for estimating the occurrence of an abnormality in a wind power generator are known. For example, Patent Document 1 describes a technique for performing abnormality diagnosis of a generator bearing based on the generator winding temperature and the generator bearing temperature measured during power generation of the generator of the wind power generator.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Incidentally, there are various types and causes of abnormalities in wind power generators, and if these abnormalities are overlooked and countermeasures are taken only after a failure of the device occurs, power generation stoppage and an increase in repair costs are unavoidable, leading to a decrease in overall operational efficiency and economic performance. In Patent Document 1, there is room for improvement in terms of rapidly estimating the occurrence of various abnormalities other than those of the generator.

[0005] An object of the present invention is to provide a machine learning device and an abnormality estimation device capable of constructing a learning model for more rapidly estimating the presence or absence of occurrence of various abnormalities and the cause of an abnormality that has occurred.

Means for Solving the Problems

[0006] (1) The machine learning device includes an input data acquisition unit that acquires a data set including at least vibration waveforms detected by vibration sensors placed on the wind power generation device as input data; a label acquisition unit that acquires the presence or absence of an abnormality in the wind power generation device and the cause of the abnormality if an abnormality occurs as labels; and a learning model construction unit that constructs a learning model for estimating the presence or absence of an abnormality in the wind power generation device and the cause of the abnormality by performing supervised learning using the pair of the input data and the labels as training data.

[0007] (2) The machine learning apparatus described in (1) further includes temperatures detected by temperature sensors placed in the wind power generation apparatus.

[0008] (3) In the machine learning apparatus described in (1) or (2), the data set further includes output information relating to the output of electricity generated by the generator of the wind power generation apparatus, and the wind speed and wind direction around the wind power generation apparatus.

[0009] (4) In the machine learning apparatus described in any of (1) to (3), the data set includes at least vibration waveforms detected by vibration sensors placed in the bearings of the gearbox of the wind power generation apparatus.

[0010] (5) The anomaly estimation device is an anomaly estimation device that uses the learning model constructed with the machine learning device described in any of (1) to (4), and comprises: an estimation data acquisition unit that acquires a data set including at least the vibration waveform as estimation data; and an estimation unit that estimates whether or not the anomaly has occurred in the wind power generation device and the cause of the anomaly based on the estimation data and the learning model. [Effects of the Invention]

[0011] According to the present invention, it is possible to provide a machine learning device and an anomaly estimation device that can construct a learning model for more quickly estimating the presence or absence of various anomalies and the causes of any anomalies that have occurred. [Brief explanation of the drawing]

[0012] [Figure 1] This figure shows the overall configuration of an anomaly estimation system according to one embodiment of the present invention. [Figure 2] This is a schematic diagram showing a management device for an anomaly estimation system according to one embodiment of the present invention and an example of a wind power generation system in which the anomaly estimation system is used. [Figure 3] Figure 2 is a schematic diagram showing the structure of the wind power generation device and the arrangement of each sensor. [Figure 4] This is a functional block diagram of a machine learning device relating to one embodiment of the present invention. [Figure 5] This is a functional block diagram of an anomaly estimation device according to one embodiment of the present invention. [Figure 6] This is a flowchart illustrating the operation of a machine learning device relating to one embodiment of the present invention. [Figure 7] This is a schematic diagram showing a flowchart of the operation of an anomaly estimation device according to one embodiment of the present invention. [Modes for carrying out the invention]

[0013] Embodiments of the present invention will be described below with reference to the drawings. However, the present invention is not limited to the following embodiments.

[0014] Figure 1 shows the overall configuration of the anomaly estimation system 100 according to this embodiment. Figure 2 is a schematic diagram showing an example of a management device 9 for the anomaly estimation system 100 and a wind power generation device 5 in which the anomaly estimation system 100 is used.

[0015] The anomaly estimation system 100 is a system for estimating whether or not an anomaly has occurred in the wind power generation device 5 and the cause of the anomaly (hereinafter referred to as the cause of the anomaly). Before describing the anomaly estimation system 100, an example of a wind power generation device 5 in which the anomaly estimation system 100 is used will be described with reference to Figures 2 and 3. Figure 3 is a schematic diagram showing the structure of the wind power generation device 5 shown in Figure 2.

[0016] The wind turbine generator 5 includes a tower 51, a nacelle 50, a hub 52, and a plurality of blades 53.

[0017] As shown in FIG. 2, the tower 51 is installed on an installation surface 8 such as the ground, and is a hollow columnar member extending in the vertical direction. The tower 51 is configured to support the hub 52, the plurality (three) of blades 53, and the nacelle 50.

[0018] The nacelle 50 is a casing connected to the upper end of the tower 51 and accommodating the main shaft 54, which is the rotation axis of the blades 53, and the like. The nacelle 50 is connected to the main shaft 54 at the end on the hub 52 side via the bearing 71 of the main shaft 54. In addition to the main shaft 54, the nacelle 50 accommodates a gearbox 55, a brake device 56, a generator 57, a transformer 58, a cooling device 59, and the like.

[0019] The hub 52 is connected to the plurality of blades 53 and the main shaft 54, and serves as the rotation center of the plurality of blades 53.

[0020] The three blades 53 are connected to the hub 52 and arranged to extend radially from the hub 52. The blades 53 are rotatable by receiving wind force, and transmit the rotational energy to the main shaft 54. That is, when the blades 53 receive wind force and rotate around the hub 52, the hub 52 rotates around the main shaft 54, and the main shaft 54 is rotationally driven. The three blades 53 are arranged at equal intervals in the rotation direction of the blades 53 when the wind turbine generator 5 is in a normal state. The number of the blades 53 is not particularly limited, and may be, for example, 2, or 4 or more.

[0021] The main shaft 54 is accommodated in the nacelle 50, one end thereof is connected to the hub 52, and the other end thereof is connected to the gearbox 55. The main shaft 54 is a shaft for transmitting the rotation of the blades 53 to the gearbox 55.

[0022] The gearbox 55 is a device that converts rotational energy from the spindle 54 into an appropriate rotational speed and transmits it to the generator 57. The gearbox 55 is a speed increaser that converts the rotational speed of the spindle 54 into a higher rotational speed and transmits it to the output shaft 571 of the generator 57. The gearbox 55 has bearings 72 and 73 located at both ends thereof, connected to the spindle 54 via bearing 72 and connected to the output shaft 571 of the generator 57 via bearing 73.

[0023] The generator 57 is a device for converting rotational energy from the main shaft 54 ​​into electricity. The generator 57 mainly consists of an output shaft 571, a coil (not shown), a housing 572 that houses part of the output shaft 571 and the coil, and bearings 74. The output shaft 571 is rotatably connected to the gearbox 55 and is rotationally driven by the gearbox 55. The generator 57 generates electricity through the rotation of this output shaft 571. The housing 572 is attached to the output shaft 571 via bearings 74. In the following description, bearings 71 to 74 may be collectively referred to as bearing 7.

[0024] The braking device 56 is attached to the output shaft 571 of the generator 57 and is a device for stopping the rotation of the output shaft 571.

[0025] The transformer 58 is a device that transforms the power generated by the generator 57 and outputs it. The power transformed by the transformer 58 is transmitted to each piece of equipment via transmission cables (not shown) that extend into the nacelle 50 and tower 51.

[0026] The cooling device 59 is a device that cools the gearbox 55, generator 57, transformer 58, etc., in order to suppress the temperature rise of these components. The cooling device 59 may be, for example, a fan that supplies cool air to the gearbox 55, generator 57, transformer 58, etc.

[0027] Next, the anomaly estimation system 100 will be described with reference to Figures 1 to 3. As shown in Figures 1 and 2, the anomaly estimation system 100 consists of a management device 9 having a machine learning device 1 and an anomaly estimation device 2, a sensor group 3, and an IoT gateway 4. As shown in Figures 2 and 3, the sensor group 3 and the IoT gateway 4 are located in the wind power generation device 5.

[0028] Sensor group 3 consists of various sensors placed in each part of the wind power generation device 5. Sensor group 3 in this embodiment includes a vibration sensor 31, a temperature sensor 32, a wind speed and wind direction sensor 33, and a power sensor 34.

[0029] The vibration sensor 31 is a sensor that detects vibration waveforms in the wind power generation device 5. In this embodiment, as shown in Figure 3, vibration sensors 311 to 318 are arranged in the wind power generation device 5 as the vibration sensor 31.

[0030] The vibration sensor 311 is located inside the tower 51 and primarily detects vibration waveforms in the tower 51. The vibration sensor 312 is located inside the hub 52 and primarily detects vibration waveforms in the hub 52. The vibration sensors 313 and 314 are located on each blade 53. Specifically, the vibration sensor 313 is located in the middle of the length of each blade 53. The vibration sensor 314 is located at the base on the hub 52 side. The vibration sensors 313 and 314 primarily detect vibration waveforms in the blade 53.

[0031] The vibration sensor 315 is located on the bearing 71 of the main shaft 54 ​​and mainly detects vibration waveforms in the main shaft 54. The vibration sensor 316 is located on the bearing 72 of the gearbox 55 and mainly detects vibration waveforms in the main shaft 54 ​​and the gearbox 55. The vibration sensor 317 is located on the bearing 73 of the gearbox 55 and mainly detects vibration waveforms in the gearbox 55 and the output shaft 571. The vibration sensor 318 is located on the output shaft 571 of the generator 57 and mainly detects vibration waveforms in the generator 57.

[0032] The temperature sensor 32 is a sensor that detects the temperature in each part of the wind power generation device 5. In this embodiment, the wind power generation device 5 is equipped with temperature sensors 321 to 324, which serve as the temperature sensor 32, as shown in Figure 3.

[0033] Temperature sensor 321 is located inside the blade 53 and detects the temperature inside the blade 53. Temperature sensor 322 is located inside the gearbox 55 and detects the oil temperature of the gearbox 55. Temperature sensor 323 is located in the coil of the generator 57 and detects the temperature of the coil. Temperature sensor 324 is located in the transformer 58 and detects the temperature of the transformer 58.

[0034] As shown in Figure 3, the wind speed and wind direction sensor 33 is positioned on the upper surface of the nacelle 50 and is a sensor that detects wind speed and wind direction around the wind power generation device 5.

[0035] The power sensor 34 is a sensor that detects information regarding the output of power generated by the generator 57 (hereinafter referred to as output information). Examples of output information detected by the power sensor 34 include the current and voltage output from the generator 57.

[0036] The IoT gateway 4 acquires a data set from the sensor group 3 via wireless communication, including vibration waveforms, temperature, wind speed, wind direction, output information, etc., detected by the sensor group 3. As shown in Figure 2, the IoT gateway 4 transmits the aggregated data set to the management device 9 in real time via the network 6. In this embodiment, the IoT gateway 4 is located inside the nacelle 50, but its placement is not particularly limited.

[0037] The management device 9 is a device for managing the wind power generation equipment 5 and includes a machine learning device 1 and an anomaly estimation device 2 that estimates whether or not an anomaly has occurred in the wind power generation equipment 5 and the cause of the anomaly if one has occurred.

[0038] Let's describe the machine learning device 1. Figure 4 shows the functional blocks of the machine learning device 1.

[0039] Machine learning device 1 constructs a learning model for use in anomaly estimation device 2 through supervised learning. The learning model constructed by machine learning device 1 is used to estimate whether or not an anomaly has occurred in the wind power generation device 5, and, if an anomaly has occurred, to estimate the cause of the anomaly. As shown in Figure 4, machine learning device 1 comprises an input data acquisition unit 11, a label acquisition unit 12, a storage unit 13, and a learning model construction unit 14.

[0040] The input data acquisition unit 11 acquires various information detected by the sensor group 3, including at least vibration waveforms, as input data. Preferably, the input data includes temperature, wind speed, wind direction, and output information in addition to the vibration waveforms detected by the sensor group 3.

[0041] The label acquisition unit 12 acquires a label indicating whether or not an abnormality has occurred in the wind turbine 5 when the sensor group 3 detects the data group, and if an abnormality has occurred, the cause of the abnormality. Examples of abnormality causes include damage to the blades 53, uneven balance between multiple blades 53, eccentricity of the main shaft 54, missing gears in the gearbox 55, insufficient lubrication, wear, distortion, loosening of bolts and nuts, failure of the cooling device 59, overload operation of the generator 57, poor electrical connection, insulation deterioration, and instability of the wind turbine 5 due to ground subsidence or deterioration of the foundation. Examples of wear of mechanical parts include wear of the bearings 7 and wear of other mechanical parts that make up the wind turbine 5.

[0042] Here, we will explain the relationship between the input data (data set) and the labels (whether or not an anomaly occurred and the cause of the anomaly). First, we will explain the relationship between the vibration waveform detected by the vibration sensor 31 and the cause of the anomaly.

[0043] For example, if the cause of the malfunction is wear of bearing 7 or a gear failure in gearbox 55, the high-frequency components of the vibration waveform tend to increase. Conversely, when no malfunction occurs, the vibration waveform tends to be dominated by low-frequency components.

[0044] Furthermore, for example, if the cause of the abnormality is an uneven balance between multiple blades 53 or eccentricity of the main spindle 54, the periodic peaks in the vibration waveform tend to increase.

[0045] Furthermore, for example, if the wind turbine 5 becomes unstable due to loosening of bolts or nuts on the wind turbine 5, or due to ground subsidence on the installation surface 8 where the wind turbine 5 is installed, or deterioration of the foundation, the amplitude of the vibration waveform tends to increase randomly.

[0046] Furthermore, for example, if lubrication is insufficient or wear occurs on mechanical parts other than bearing 7, the vibration waveform tends to increase steadily. Since the vibration waveform tends to change depending on the cause of the abnormality, by using machine learning to study the vibration waveform patterns under normal conditions and the vibration waveform patterns for each cause of abnormality, it is possible to construct a learning model that can quickly estimate the presence or absence of an abnormality and its cause based on the vibration waveform.

[0047] Furthermore, the vibration sensor 31 is preferably placed in at least one of the main shaft 54, the generator 57 bearing 7, and the gearbox 55 bearing 7, in order to estimate a wider variety of abnormal causes. Among these, the gearbox 55 bearing 7 is more preferable as a placement for the vibration sensor 31 because it easily transmits not only vibrations due to malfunctions in the gearbox 55, but also vibrations from the blade 53 via the main shaft 54, vibrations from the generator 57 via the output shaft 571, and vibrations of the main shaft 54.

[0048] The relationship between the temperature detected by the temperature sensor 32 and the cause of the abnormality will be explained.

[0049] For example, if the bearing 7 wears out or the gearbox 55 malfunctions, the temperature tends to rise rapidly. Also, for example, if the cooling system 59 fails or the wind power generator 5 is operating under overload, the temperature of the generator 57's coils and the transformer 58 tends to remain high for an extended period.

[0050] For example, if electrical connection failures or insulation deterioration occur in the wind turbine 5, a localized temperature rise tends to occur in the wind turbine 5. Also, for example, if lubrication failures or mechanical distortions occur, the temperature distribution in the wind turbine 5 tends to become uneven. Since the temperature tends to change according to each cause of abnormality, by learning the temperature patterns along with the vibration waveforms for each cause of abnormality in both normal and abnormal conditions, it is possible to construct a learning model that can estimate the presence or absence of abnormalities and, if abnormalities occur, the cause of the abnormality with higher accuracy.

[0051] For example, if wear occurs on the mechanical parts of the wind turbine 5 or if there is a malfunction in the blades 53, the output of the generator 57 tends to decrease. Also, for example, the power generated and output by the generator 57 tends to increase when the wind speed is strong and the wind direction is such that it is easy to rotate the blades 53. For this reason, the presence or absence of an abnormality and its cause can be estimated based on the wind speed, wind direction and output information. Specifically, if no power corresponding to the detected wind speed and wind direction is detected, it can be estimated that an abnormality has occurred in at least one of the blades 53, main shaft 54, gearbox 55, generator 57, transformer 58, etc.

[0052] The memory unit 13 uses the data set acquired by the input data acquisition unit 11 as input data and stores the presence or absence of an anomaly and the cause of the anomaly acquired by the label acquisition unit 12 as labels. Specifically, the memory unit 13 stores the vibration waveform, temperature, wind speed, wind direction, and output information detected at the same time as a set of data that are linked together. The memory unit 13 then links the input data, which is a set of data that are linked together, with the labels, which are the presence or absence of an anomaly and the cause of the anomaly corresponding to the input data, and stores them as a set of training data. The presence or absence of an anomaly and the cause of the anomaly corresponding to the input data refers to the presence or absence of an anomaly and the cause of the anomaly at the time the data set was detected. The memory unit 13 also stores the learning model constructed by the learning model construction unit 14.

[0053] The learning model building unit 14 uses the input data stored in the memory unit 13 and the labels as training data. By performing supervised learning on the data, a learning model is constructed to estimate whether or not an anomaly has occurred, and to estimate the cause of the anomaly if one does occur. The learning model construction unit 14 then transmits the constructed learning model to the anomaly estimation device 2.

[0054] The learning model construction unit 14 can be implemented, for example, using a Support Vector Machine (SVM).

[0055] The anomaly estimation device 2 will be explained with reference to Figure 5. Figure 5 is a functional block diagram of the anomaly estimation device 2. The anomaly estimation device 2 comprises a control unit 20, a storage unit 21, a communication unit 22, and an output unit 23.

[0056] The control unit 20 controls the entire anomaly estimation device 2 and realizes the various functions in this embodiment by appropriately reading and executing various programs from a storage area such as ROM, RAM, flash memory, or hard disk (HDD). The control unit 20 may be a processor. A processor is a CPU (central processing unit), MPU (micro processing unit), SoC (system on a chip), DSP (digital signal processor), GPU (graphics processing unit), ASIC (application specific integrated circuit), PLD (programmable logic device), or FPGA (field-programmable gate array). Alternatively, a processor may be a combination of several of these. Furthermore, a processor may be a combination of these with a hardware accelerator or the like.

[0057] As shown in Figure 5, the control unit 20 includes an estimation data acquisition unit 201, an estimation unit 202, and an output processing unit 203.

[0058] The estimation data acquisition unit 201 acquires estimation data to estimate whether or not an abnormality has occurred in the wind power generation device 5 and, if an abnormality has occurred, the cause of the abnormality. The estimation data acquisition unit 201 acquires various current data detected by the sensor group 3 arranged in the wind power generation device 5 as estimation data. The estimation data includes at least the vibration waveform detected by the vibration sensor 31, and more preferably includes the temperature detected by the temperature sensor 32, the wind speed and wind direction detected by the wind speed and wind direction sensor 33, and the output information detected by the power sensor 34, along with the vibration waveform.

[0059] The estimation unit 202 estimates whether or not an abnormality has occurred in the wind power generation device 5 at the present time, based on the estimation data acquired by the estimation data acquisition unit 201 and the learning model constructed by the machine learning device 1, and if an abnormality is estimated to have occurred, it estimates the cause of the abnormality. Specifically, the estimation unit 202 inputs the estimation data into the learning model and executes a process to cause the learning model to estimate whether or not an abnormality has occurred and the cause of the abnormality.

[0060] The output processing unit 203 executes a process to output to the output unit 23 whether or not an abnormality has occurred in the wind power generation device 5, as estimated by the estimation unit 202, and, if an abnormality has occurred, the cause of the abnormality.

[0061] The memory unit 21 stores the learning model acquired from the machine learning device 1. The memory unit 21 may also use the estimation data acquired by the estimation data acquisition unit 201 as input data and store the presence or absence of an anomaly determined to have occurred and the cause of the anomaly as labels.

[0062] The communication unit 22 sends and receives data to and from the machine learning device 1. For example, the management device 9 can transmit a learning model constructed by the learning model construction unit 14 from the machine learning device 1 to the anomaly estimation device 2 via the communication unit 22. Conversely, the management device 9 can transmit a set of new estimation data acquired by the estimation data acquisition unit 201 and estimation results such as whether or not an anomaly occurred and the cause of the anomaly, based on the estimation data, from the anomaly estimation device 2 to the machine learning device 1 via the communication unit 22.

[0063] The output unit 23 outputs the presence or absence of an abnormality transmitted from the output processing unit 203 of the control unit 20, and if an abnormality is estimated to have occurred, the cause of the abnormality as an estimated result. The output unit 23 may be a monitor that outputs the estimated result transmitted from the output processing unit 203 as an image, or it may be a speaker that outputs it as sound.

[0064] Next, the operation of the anomaly estimation system 100 according to this embodiment during machine learning will be described. Figure 6 is a flowchart showing the operation of the machine learning device 1 during machine learning.

[0065] In step S11, the input data acquisition unit 11 of the machine learning device 1 acquires a data set that includes at least vibration waveforms detected by the sensor group 3 as input data.

[0066] In step S12, the label acquisition unit 12 of the machine learning device 1 acquires a label indicating whether or not an anomaly has occurred corresponding to the data set acquired in step S11. At this time, if the label acquisition unit 12 finds that an anomaly has occurred in the label corresponding to the data set, it also acquires the cause of the anomaly as a label.

[0067] In step S13, the learning model building unit 14 of the machine learning device 1 receives pairs of input data and labels as training data.

[0068] In step S14, the learning model building unit 14 of the machine learning device 1 performs machine learning using this training data.

[0069] In step S15, the learning model building unit 14 determines whether to terminate machine learning or repeat it. If the learning model building unit 14 determines to repeat machine learning (No in step S15), it returns to step S11. Then, the machine learning device 1 repeats the same operation. On the other hand, if the learning model building unit 14 determines to terminate machine learning (Yes in step S15), it proceeds to step S16. The conditions for terminating machine learning can be arbitrarily defined. For example, machine learning may be terminated after repeating machine learning a predetermined number of times.

[0070] In step S16, the machine learning device 1 transmits the learning model constructed by machine learning up to that point to the anomaly estimation device 2.

[0071] Furthermore, the memory unit 13 of the machine learning device 1 stores this learned model. This allows the device to transmit the learned model to the anomaly estimation device 2 when the device requests it. Additionally, when new training data is acquired, further machine learning can be performed on the learned model.

[0072] Next, an example of the process by which the abnormality estimation device 2 estimates whether or not an abnormality has occurred and the cause of the abnormality will be explained with reference to Figure 7. Figure 7 is a flowchart showing the process flow of the abnormality estimation device 2 estimating whether or not an abnormality has occurred and the cause of the abnormality in the wind power generation device 5.

[0073] In step S21, the estimation data acquisition unit 201 acquires a data set including vibration waveforms detected by the sensor group 3 arranged in the wind power generation device 5 as estimation data.

[0074] In step S22, the estimation unit 202 reads the learning model constructed by the machine learning device 1 from the memory unit 21.

[0075] In step S23, the estimation unit 202 estimates whether or not an abnormality has occurred in the wind power generation device 5 based on the estimation data, including the vibration waveform acquired in step S21, and the learning model read out in step S22.

[0076] If the estimation unit 202 estimates that an abnormality has occurred (step S23; YES), in step S24, it estimates the cause of the abnormality based on the estimation data acquired in step S21 and the learning model read out in step S22. In step S25, the output processing unit 203 outputs to the output unit 23 that an abnormality has occurred in the wind power generation device 5 and the cause of the abnormality estimated in step S24, and then terminates processing.

[0077] On the other hand, if the estimation unit 202 estimates that no abnormality has occurred (step S23; NO), it skips the processing in step S24 and proceeds to step S25. In step S25, the output processing unit 203 outputs to the output unit 23 that no abnormality has occurred in the wind power generation device 5, and then terminates the processing.

[0078] According to the embodiments described above, the following effects are achieved.

[0079] The machine learning apparatus 1 according to this embodiment includes an input data acquisition unit 11 that acquires a data set including at least vibration waveforms detected by a vibration sensor 31 placed in the wind power generation apparatus 5 as input data; a label acquisition unit 12 that acquires the presence or absence of an abnormality in the wind power generation apparatus 5 and the cause of the abnormality if an abnormality occurs as labels; and a learning model construction unit 14 that constructs a learning model for estimating the presence or absence of an abnormality in the wind power generation apparatus 5 and the cause of the abnormality by performing supervised learning using the input data and label pair as training data.

[0080] This allows for the use of vibration waveforms with different patterns depending on the presence or absence of an anomaly and its cause as input data, enabling the construction of a learning model for more quickly estimating the presence or absence of various anomalies and their causes in the wind power generation device 5.

[0081] Furthermore, in the machine learning device 1 according to this embodiment, the data set further includes the temperature detected by the temperature sensor 32 located in the wind power generation device 5.

[0082] This allows for the construction of a learning model that also takes into account temperature data, whose trends change depending on whether or not an anomaly occurs and the cause of the anomaly. As a result, a learning model can be constructed to more accurately estimate whether or not an anomaly occurs and the cause of the anomaly in the wind power generation device 5.

[0083] Furthermore, in the machine learning device 1 according to this embodiment, the data set further includes output information relating to the power output generated by the generator 57 of the wind power generation device 5, and wind speed and wind direction around the wind power generation device 5.

[0084] This allows for the construction of a learning model that takes into account the power output generated by the generator 57 and the wind speed and wind direction that affect that output, thereby enabling the construction of a learning model that more accurately estimates whether or not an abnormality has occurred in the wind power generation device 5.

[0085] Furthermore, in the machine learning device 1 according to this embodiment, the data set includes at least vibration waveforms detected by a vibration sensor 31 located on the bearing 7 of the gearbox 55 of the wind power generation device 5.

[0086] As a result, vibration waveforms detected by vibration sensors 31 located in the bearings 7 of the gearbox 55, which receive vibrations from more parts of the wind turbine 5, are used as input data for machine learning. This allows for the more efficient construction of learning models to more quickly estimate the presence or absence of various abnormalities and their causes.

[0087] Furthermore, the anomaly estimation device 2 according to this embodiment is an anomaly estimation device 2 that uses the learning model constructed by a machine learning device, and comprises an estimation data acquisition unit 201 that acquires a data set including at least vibration waveforms as estimation data, and an estimation unit 202 that estimates whether or not an anomaly has occurred in the wind power generation device 5 and the cause of the anomaly based on the estimation data and the learning model.

[0088] As a result, by inputting vibration waveforms with different patterns depending on the cause of the anomaly as estimation data into the learning model according to this embodiment, the presence or absence of an anomaly and its cause can be estimated, making it possible to more quickly estimate the presence or absence of various anomalies and their causes in the wind power generation device 5.

[0089] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above and can be modified as appropriate. [Explanation of Symbols]

[0090] 1. Machine Learning Device 2 Abnormality estimation device 5. Wind power generation equipment 11 Input data acquisition unit 12 Label acquisition unit 14. Learning Model Construction Section 31. Vibration Sensor 201 Data acquisition unit for estimation 202 Estimation section

Claims

1. An input data acquisition unit acquires a data set that includes at least vibration waveforms detected by vibration sensors placed on a wind power generation device as input data, A label acquisition unit that acquires, as a label, whether or not an abnormality has occurred in the wind power generation device and, if an abnormality has occurred, the cause of the abnormality. A machine learning device comprising: a learning model construction unit that constructs a learning model for estimating whether or not an abnormality occurs in the wind power generation device and the cause of the abnormality by performing supervised learning using the input data and the label as training data.

2. The machine learning apparatus according to claim 1, wherein the data set further includes temperatures detected by temperature sensors placed in the wind power generation apparatus.

3. The machine learning apparatus according to claim 1, further comprising output information relating to the output of electricity generated by the generator of the wind power generation apparatus, and wind speed and wind direction around the wind power generation apparatus.

4. The machine learning apparatus according to claim 1, wherein the data set includes at least vibration waveforms detected by vibration sensors positioned in the bearings of the gearbox of the wind power generation apparatus.

5. An anomaly estimation device using the learning model constructed with the machine learning device described in any one of claims 1 to 4, An estimation data acquisition unit that acquires a data set including at least the vibration waveform as estimation data, An anomaly estimation device comprising: an estimation unit that estimates whether or not an anomaly has occurred in the wind power generation device and the cause of the anomaly based on the estimation data and the learning model.

Images