Abnormality detection system

Abstract

To detect an abnormality of an apparatus comprising an inverter without preparing a special rule.SOLUTION: An abnormality detection system 1 for detecting an abnormality of an apparatus 10 comprising an inverter 13 includes: a sensor 31 for detecting a current of any one of a power source side power line 40 which supplies power to the apparatus 10 and an output side power line 50 which supplies power from the inverter 13 to the outside; an analysis unit 32 which generates one or more pieces of analysis data on the basis of the detected current; a learning unit 33 for learning a machine learning model 34 using one or more pieces of first analysis data which are generated while defining as a target a current which is detected during a first period; an acquisition unit 35 for acquiring an evaluation result from the machine learning model 34 by inputting one or more pieces of second analysis data, which are generated while defining as a target a current which is detected during a second period, to the learnt machine learning model 34 and performing evaluation regarding the abnormality of the apparatus 10; and a diagnosis unit 36 which diagnoses the abnormality of the apparatus 10 on the basis of the evaluation result.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to an anomaly detection system that detects an anomaly in an apparatus equipped with an inverter. [Background technology]

[0002] Patent Document 1 discloses a technique for detecting an abnormality in a device equipped with an inverter. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2003-156547 A [Patent Document 2] International Publication No. 2020 / 261639 Summary of the Invention [Problem to be solved by the invention]

[0004] The technology disclosed in Patent Document 1 employs a rule-based algorithm as an algorithm for detecting abnormalities. Therefore, in order to use this technology to detect abnormalities in equipment equipped with an inverter, it is necessary to prepare special rules for that equipment in advance.

[0005] In order to prepare these rules, detailed information on the specifications of the device, the conditions of use, etc. is required. For this reason, it is difficult to relatively easily introduce the technology disclosed in Patent Document 1 into a detection system that detects abnormalities in devices equipped with inverters.

[0006] Therefore, an object of the present disclosure is to provide an anomaly detection system that can detect an anomaly in an equipment equipped with an inverter without having to prepare special rules for the target equipment in advance. [Means for solving the problem]

[0007] An anomaly detection system according to one embodiment of the present disclosure is an anomaly detection system that detects an anomaly in an equipment having an inverter, and includes: a sensor that detects a current in either a power supply side power line that supplies power to the equipment from an outside source or an output side power line that supplies power from the inverter to an outside source; an analysis unit that generates one or more analysis data based on the current detected by the sensor; an acquisition unit that inputs one or more second analysis data generated by the analysis unit for a current detected by the sensor in a second time period to a machine learning model trained using the one or more first analysis data generated by the analysis unit for a current detected by the sensor in a first time period, and causes the machine learning model to evaluate an anomaly in the equipment, thereby acquiring an evaluation result from the machine learning model; and a diagnosis unit that diagnoses an anomaly in the equipment based on the evaluation result. Effect of the Invention

[0008] According to one aspect of the present disclosure, an anomaly detection system is provided that can detect an anomaly in an equipment equipped with an inverter without having to prepare special rules for the target equipment in advance. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing a configuration of an anomaly detection system according to an embodiment. [Diagram 2] FIG. 2 is a schematic diagram illustrating an example of analysis data generated by an analysis unit according to the embodiment. [Diagram 3] FIG. 3 is a schematic diagram illustrating an example of how the learning unit according to the embodiment determines whether or not the analysis data is appropriate as data to be used for learning a machine learning model. [Figure 4] FIG. 4 is a schematic diagram showing how a trained machine learning model according to an embodiment outputs a degree of change as an evaluation result. [Diagram 5] FIG. 5 is a flowchart of the learning process according to the embodiment. [Figure 6]FIG. 6 is a flowchart of the diagnosis process according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] (How one aspect of the present disclosure was achieved) The inventors conducted extensive experiments and research in order to realize an anomaly detection system that can detect anomalies in equipment equipped with an inverter without the need to prepare special rules in advance for the target equipment.

[0011] As a result, the inventors came up with the idea of ​​an anomaly detection system according to one aspect of the present disclosure described below.

[0012] An anomaly detection system according to one embodiment of the present disclosure is an anomaly detection system that detects an anomaly in an equipment having an inverter, and includes: a sensor that detects a current in either a power supply side power line that supplies power to the equipment from an outside source or an output side power line that supplies power from the inverter to an outside source; an analysis unit that generates one or more analysis data based on the current detected by the sensor; an acquisition unit that inputs one or more second analysis data generated by the analysis unit for a current detected by the sensor in a second time period to a machine learning model trained using the one or more first analysis data generated by the analysis unit for a current detected by the sensor in a first time period, and causes the machine learning model to evaluate an anomaly in the equipment, thereby acquiring an evaluation result from the machine learning model; and a diagnosis unit that diagnoses an anomaly in the equipment based on the evaluation result.

[0013] According to the anomaly detection system having the above configuration, it is possible to diagnose an anomaly in an inverter-equipped device based on the evaluation result of the anomaly evaluation of the inverter-equipped device obtained from the trained machine learning model, and therefore it is not necessary to prepare special rules for the target device in advance.

[0014] In this way, according to the anomaly detection system configured as described above, an anomaly detection system is provided that can detect anomalies in equipment equipped with an inverter without having to prepare special rules for the target equipment in advance.

[0015] The analysis unit may generate the one or more pieces of analysis data by performing a frequency analysis on the current.

[0016] This allows one or more pieces of analysis data to reflect the results of the frequency analysis.

[0017] The sensor may also be a harmonic sensor having a frequency-dependent characteristic in which the detection sensitivity increases in proportion to the frequency.

[0018] It is generally known that when an abnormality occurs in equipment equipped with an inverter, there will be a difference in the harmonic components of the fundamental frequency among the frequency components of the current flowing in the power supply side power line and / or the output side power line compared to when no abnormality occurs.

[0019] Therefore, the anomaly detection system configured as described above can detect anomalies in equipment equipped with an inverter more accurately than a configuration using a sensor whose detection sensitivity is constant regardless of frequency.

[0020] The current detected by the sensor may be a current in the output power line.

[0021] This makes it possible to detect an abnormality in a device equipped with an inverter by utilizing the output power line.

[0022] The system may further include a learning unit configured to train the machine learning model using the one or more pieces of first analytical data.

[0023] This allows the machine learning model to be trained.

[0024] Furthermore, the learning of the machine learning model performed by the learning unit may be unsupervised learning.

[0025] This makes it possible to train the machine learning model without preparing any data other than the one or more first analysis data.

[0026] In addition, the evaluation result may include a first evaluation result and a second evaluation result in which at least a portion of the evaluation targets for the current are different from each other, and the diagnosing unit may diagnose an abnormality in the device based on the first evaluation result and the second evaluation result, and may further diagnose an abnormality in a facility to which power is supplied by the output power line.

[0027] This makes it possible to detect abnormalities in equipment to which power is supplied via the output power line.

[0028] The first evaluation result is an evaluation result obtained by evaluating a current component of the current in a first frequency band, and the second evaluation result is an evaluation result obtained by evaluating a current component of the current in a second frequency band including a frequency band higher than the first frequency band. It is also possible to use the following.

[0029] This allows the second evaluation result to be an evaluation result in which the second frequency band including a frequency band higher than the first frequency band that is the evaluation target of the first evaluation result is evaluated.

[0030] A specific example of an anomaly detection system according to one aspect of the present disclosure will be described below with reference to the drawings. Each embodiment shown here shows one specific example of the present disclosure. Therefore, the numerical values, shapes, components, the arrangement and connection of the components, as well as the steps (processes) and the order of the steps shown in the following embodiments are merely examples and are not intended to limit the present disclosure. In addition, each figure is a schematic diagram and is not necessarily an exact illustration. In each figure, the same reference numerals are given to substantially the same configurations, and duplicated explanations are omitted or simplified.

[0031] (Embodiment) Hereinafter, an abnormality detection system according to an embodiment capable of detecting an abnormality in an apparatus equipped with an inverter will be described.

[0032] <Configuration> FIG. 1 is a block diagram showing a configuration of an anomaly detection system 1 according to an embodiment.

[0033] As shown in FIG. 1, the anomaly detection system 1 includes a device 10, a facility 20, an anomaly detection device 30, a power supply side power line 40, an output side power line 50, and a power supply 60.

[0034] The power supply 60 is a power supply for supplying power to the device 10. Here, the power supply 60 will be described as being an AC power supply. In this case, the power supply 60 is, for example, a commercial AC power supply.

[0035] The power supply side power line 40 is connected to the power source 60 and the device 10, and transmits AC power output from the power source 60 to the device 10. For example, when the power source 60 is a commercial AC power source, the power supply side power line 40 is a single-phase two-wire cable. An appropriate cable is selected for the power supply side power line 40 according to the type of the power source 60. For example, when the power source 60 is a three-phase AC power source, the power supply side power line 40 may be a three-phase three-wire cable.

[0036] The device 10 includes a converter 11, a smoothing circuit 12, and an inverter 13. Here, the device 10 will be described as a servo amplifier that drives a servo motor 21, which will be described later.

[0037] In the following, the device 10 will be described as including the converter 11, the smoothing circuit 12, and the inverter 13, but it is sufficient that the device 10 is configured to include at least the inverter 13, and the converter 11 and the smoothing circuit 12 are not necessarily essential components of the device 10. For example, when the power source 60 is a DC power source, the device 10 may be configured such that the DC power supplied from the power source 60 via the power source side power line 40 is directly input to the inverter 13.

[0038] The converter 11 converts AC power supplied from the power source 60 via the power source side power line 40 into pulsating power. The converter 11 functions as a pulsating power supply source that supplies the pulsating power to the smoothing circuit 12.

[0039] The smoothing circuit 12 smoothes the pulsating power supplied from the converter 11 and converts it into DC power. The smoothing circuit 12 functions as a DC power supply source that supplies the DC power to the inverter 13.

[0040] The inverter 13 converts the DC power output from the smoothing circuit 12 into AC power. Here, the inverter 13 will be described as generating three-phase AC power for driving a servo motor 21, which will be described later, by switching the power supply voltage of the DC power supplied from the smoothing circuit 12. The inverter 13 functions as a three-phase AC power supply source that supplies the three-phase AC power to the servo motor 21.

[0041] The timing of switching performed by the inverter 13 is determined, for example, based on an operation command (e.g., a speed command) for the servo motor 21 output from a higher-level device of the equipment 10 and an encoded signal indicating the position of the servo motor 21 that is fed back from the servo motor 21, so as to operate the servo motor 21 in accordance with the operation command.

[0042] The output power line 50 is connected to the device 10 and the facility 20, and transmits the three-phase AC power output from the device 10 to the facility 20. The output power line 50 is, for example, a three-phase three-wire cable. An appropriate cable is selected for the output power line 50 depending on the power output from the device 10. For example, when the power output from the device 10 is single-phase AC power, the output power line 50 may be a single-phase two-wire cable.

[0043] The facility 20 includes a servo motor 21. The facility 20 is, for example, a machine tool installed on a production line in a factory.

[0044] The servo motor 21 is a motor driven by the device 10, which is a servo amplifier, and operates using three-phase AC power supplied from the inverter 13 via the output power line 50 as a power source.

[0045] The abnormality detection device 30 is a device for detecting an abnormality in the device 10.

[0046] As shown in FIG. 1, the anomaly detection device 30 includes a sensor 31, an analysis unit 32, a learning unit 33, a machine learning model 34, an acquisition unit 35, a diagnosis unit 36, and an output unit 37.

[0047] The anomaly detection device 30 may be configured to include, for example, one or more computer devices as hardware resources, and the functions of the analysis unit 32, the learning unit 33, the machine learning model 34, the acquisition unit 35, the diagnosis unit 36, and the output unit 37 may be realized by one or more processors (e.g., CPUs (Central Processing Units)) executing programs stored in one or more memories (e.g., ROMs (Read Only Memory) and / or RAMs (Random Access Memory)) in the one or more computer devices.

[0048] The sensor 31 detects a current flowing through either the output side power line 50 or the power supply side power line 40 .

[0049] FIG. 1 illustrates sensor 31 as if it were detecting only the current in output power line 50, but sensor 31 can also detect the current flowing in output power line 50 and can also detect the current flowing in power supply power line 40.

[0050] Although the following describes a case where the sensor 31 detects a current flowing through the output-side power line 50, the same applies to a case where the sensor 31 detects a current flowing through the power supply-side power line 40.

[0051] The sensor 31 may be, for example, a harmonic sensor having a frequency-dependent characteristic in which the detection sensitivity increases in proportion to the frequency, or may be, for example, a current sensor in which the detection sensitivity does not depend on the frequency.

[0052] A known example of a harmonic sensor having a frequency-dependent characteristic in which the detection sensitivity increases in proportion to the frequency is the current measurement system disclosed in Patent Document 2. As shown in Fig. 5 of Patent Document 2, this current measurement system has a frequency-dependent characteristic in which the detection sensitivity increases in proportion to the frequency.

[0053] It is generally known that when an abnormality occurs in a device equipped with an inverter, or in equipment equipped with a servo motor driven by an inverter, a difference occurs in the harmonic components of the fundamental frequency among the frequency components of the current flowing in the power line supplying power from the inverter to the servo motor and / or the power line supplying power to the equipment, compared to when no abnormality occurs.

[0054] More specifically, for example, as shown in Table 6 of Patent Document 1, it is known that (1) when an abnormality occurs in equipment equipped with a servo motor driven by an inverter, the second to tenth harmonic components of the current flowing in the power line supplying power from the inverter to the servo motor differ from when no abnormality occurs; and as shown in Table 7 of Patent Document 1, (2) when an abnormality occurs in a device equipped with an inverter, the second to thirteenth harmonic components of the current flowing in the power line supplying power to the device differ from when no abnormality occurs; and the second to 38th harmonic components of the current flowing in the power line supplying power from the inverter to the servo motor differ from when no abnormality occurs.

[0055] For this reason, it is more desirable for the sensor 31 to be a harmonic sensor having frequency-dependent characteristics in which the detection sensitivity increases in proportion to the frequency, rather than a current sensor in which the detection sensitivity does not depend on the frequency. This is because the sensor 31 being a harmonic sensor can prevent the signal of higher-order harmonic components from being buried in noise, making it impossible to perform significant detection.

[0056] Here, the sensor 31 will be described as a harmonic sensor disclosed as a current measurement system in Patent Document 2.

[0057] The sensor 31 is configured, for example, by being retrofitted and clamped on to an existing output side power line 50 or an existing power supply side power line 40, so as to be able to detect the current flowing in the output side power line 50 or the power supply side power line 40.

[0058] In this case, for example, when detecting a current flowing through the output side power line 50, the sensor 31 may detect the current by clamping on to one of the three-phase three-wire cables, and when detecting a current flowing through the power supply side power line 40, the sensor 31 may detect the current by clamping on to one of the single-phase two-phase cables.

[0059] The analysis unit 32 generates one or more pieces of analysis data based on the current detected by the sensor 31 .

[0060] More specifically, the analysis unit 32 performs a frequency analysis on the current detected by the sensor 31, and generates one or more pieces of analysis data.

[0061] Here, the analysis unit 32 will be described as performing a Fourier transform as the frequency analysis on the current detected by the sensor 31.

[0062] More specifically, the analysis unit 32 divides the signal output in time series from the sensor 31 into one or more time series signals for each predetermined period (e.g., one second), and performs a Fourier transform on each of the divided time series signals to generate a frequency spectrum for each of the divided time series signals as one or more pieces of analysis data.

[0063] Note that, here, the analysis unit 32 is described as generating the one or more pieces of analysis data by performing the above-mentioned frequency analysis on the current detected by the sensor 31; however, as long as the analysis unit 32 is capable of generating one or more pieces of analysis data based on the current detected by the sensor 31, the analysis unit 32 is not necessarily limited to a configuration in which it generates the one or more pieces of analysis data by performing the above-mentioned frequency analysis on the current.

[0064] The analysis unit 32 may be configured, for example, to treat the current detected by the sensor 31 as it is as one or more pieces of analysis data, or may be configured, for example, to treat the current detected by the sensor 31 as one or more pieces of analysis data by filtering the data obtained so that only components in the frequency band of several to several tens of harmonics remain.

[0065] FIG. 2 is a schematic diagram showing an example of the analysis data generated by the analysis unit 32. As shown in FIG.

[0066] As shown in FIG. 2, the analysis unit 32 generates, as analysis data, a frequency spectrum at frequencies for each predetermined frequency resolution.

[0067] Returning to FIG. 1, the explanation of the anomaly detection device 30 will be continued.

[0068] The learning unit 33 trains the machine learning model 34 using one or more pieces of first analysis data generated by the analysis unit 32 with respect to the current detected by the sensor 31 during the first period.

[0069] Here, the first period refers to a period during which the device 10 and the facility 20 are operating normally.

[0070] The machine learning model 34 is, for example, a neural network.

[0071] More specifically, the learning unit 33 inputs each of the one or more first analysis data generated by the analysis unit 32 for the current detected by the sensor 31 during a first period in which the equipment 10 and the facility 20 are operating normally into the machine learning model 34 as normal data, and trains the machine learning model 34 so that the degree of change indicating the degree of deviation of each normal data from the distribution of the one or more normal data becomes zero.

[0072] Here, the learning unit 33 is described as training the machine learning model 34 by unsupervised learning that uses only the one or more normal data. However, the learning unit 33 may be configured to train the machine learning model 34 by supervised learning in which, in addition to the one or more normal data, one or more pieces of analysis data generated by the analysis unit 32 for a current detected during a period when the device 10 and / or the facility 20 are not operating normally are input to the machine learning model 34 as one or more pieces of abnormal data.

[0073] The learning unit 33 may further determine whether or not each of the one or more analysis data generated by the analysis unit 32 for the current detected by the sensor 31 during the first period is appropriate as data to be used for learning the machine learning model 34, and may train the machine learning model 34 on the one or more analysis data determined to be appropriate as one or more first analysis data.

[0074] In this case, the learning unit 33 may determine that analysis data that satisfies specified conditions indicating that the equipment 10 and / or the facility 20 is performing significant operation is appropriate as data to be used for learning the machine learning model 34.

[0075] In this case, the predetermined condition may be, for example, a condition that the amplitude of the signal output from the sensor 31 corresponding to the analysis data to be determined is equal to or greater than a predetermined value. By setting such a condition, the learning unit 33 can suppress training of the machine learning model 34 using analysis data generated based on the current detected by the sensor 31 during a period in which the device 10 and / or the facility 20 is not performing a significant operation (for example, a period in which the device 10 and / or the facility 20 is energized but is not performing a physical operation).

[0076] FIG. 3 is a schematic diagram showing an example of how the learning unit 33 determines whether or not the analysis data generated by the analysis unit 32 is appropriate as data to be used for learning the machine learning model 34.

[0077] As shown in Figure 3, the learning unit 33 may, for example, determine that the analysis data is appropriate as data to be used for training the machine learning model 34 when the amplitude of the signal output from the sensor 31 during a period corresponding to the target analysis data is equal to or greater than a predetermined value, and may determine that the analysis data is not appropriate as data to be used for training the machine learning model 34 when the amplitude is less than the predetermined value.

[0078] Returning to FIG. 1, the explanation of the anomaly detection device 30 will be continued.

[0079] When new analytical data output from the analysis unit 32 is input to the machine learning model 34 trained by the learning unit 33, the machine learning model 34 outputs a degree of change indicating the degree of deviation of the new analytical data from the distribution of one or more normal data.

[0080] That is, when new analysis data is input, the machine learning model 34 trained by the learning unit 33 evaluates abnormalities in the equipment 10 and the facility 20, and outputs the degree of change as the evaluation result.

[0081] FIG. 4 is a schematic diagram showing how, when new analytical data is input, the machine learning model 34 trained by the learning unit 33 outputs a degree of change indicating the degree of deviation of the new analytical data from the distribution of normal data as an evaluation result for the evaluation of abnormalities in the equipment 10 and the facility 20.

[0082] Here, the degree of change is a scalar value that increases as the degree of deviation from the distribution of normal data increases.

[0083] Here, when new analysis data is input, the machine learning model 34 trained by the learning unit 33 is described as outputting, as evaluation results for the evaluation of abnormalities in the equipment 10 and the facility 20, a first degree of change indicating the degree of deviation of the current detected by the sensor 31 from the distribution of normal data in a frequency band ranging from the second to the tenth harmonics of the fundamental frequency (hereinafter also referred to as the "first frequency band"), and a second degree of change indicating the degree of deviation of the current detected by the sensor 31 from the distribution of normal data in a frequency band ranging from the 11th to the 38th harmonics of the fundamental frequency (hereinafter also referred to as the "second frequency band").

[0084] That is, here, the machine learning model 34 learned by the learning unit 33 is described as outputting a first degree of change as a first evaluation result, in which the current component of the current detected by the sensor 31 in a first frequency band is evaluated, and outputting a second degree of change as a second evaluation result, in which the current component of the current detected by the sensor 31 in a second frequency band is evaluated.

[0085] In addition, the machine learning model 34 learned by the learning unit 33 does not necessarily have to be limited to a configuration that outputs the first evaluation result and the second evaluation result as described above, as long as it is able to output a first evaluation result and a second evaluation result in which at least some of the evaluation targets for the current detected by the sensor 31 are different from each other.

[0086] Returning to FIG. 1, the explanation of the anomaly detection device 30 will be continued.

[0087] The acquisition unit 35 inputs one or more second analysis data generated by the analysis unit 32 for the current detected by the sensor 31 during a second period into the machine learning model 34 learned by the learning unit 33, and acquires one or more evaluation results output from the machine learning model 34 by having the machine learning model 34 evaluate abnormalities in the equipment 10 and the facility 20.

[0088] That is, the acquisition unit 35 inputs one or more second analysis data to the machine learning model 34 learned by the learning unit 33, acquires one or more first change degrees output from the machine learning model 34 as one or more first evaluation results, and acquires one or more second change degrees output from the machine learning model 34 as one or more second evaluation results.

[0089] For example, the second period may be a period after the first period. In this case, the second period may be a period during which some abnormality occurs in the device 10 and / or the facility 20. That is, if some abnormality occurs in the device 10 and / or the facility 20 after the first period ends and before the second period ends, the second period will be a period during which an abnormality occurs in the device 10 and / or the facility 20, and if no abnormality occurs in the device 10 and the facility 20 after the first period ends and before the second period ends, the second period will be a period during which no abnormality occurs in the device 10 and the facility 20.

[0090] It should be noted that, when the machine learning model 34 is made to evaluate abnormalities in the equipment 10 and the facility 20 going back to the past, the second period is a period prior to the first period.

[0091] In addition, when an evaluation of abnormalities in other equipment and facilities similar to the equipment 10 and the equipment 20 is performed using a machine learning model 34 trained using one or more first analysis data generated by the analysis unit 32 for the current detected by the sensor 31 during a first period, the first period and the second period may be the same period.

[0092] The acquisition unit 35 may further determine whether or not each of the one or more analysis data generated by the analysis unit 32 for the current detected by the sensor 31 during the second period is appropriate as data to be used by the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20, and may cause the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20 using the one or more analysis data determined to be appropriate as one or more second analysis data.

[0093] In this case, the acquisition unit 35 may determine that the analysis data that satisfies predetermined conditions indicating that the equipment 10 and / or the facility 20 is performing significant operation is appropriate as data to be used by the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20.

[0094] In this case, the predetermined condition may be, for example, a condition that the amplitude of the signal output from the sensor 31 corresponding to the analysis data to be determined is equal to or greater than a predetermined value. By setting such a condition, the acquisition unit 35 can suppress the machine learning model 34 from evaluating anomalies in the equipment 10 and the equipment 20 using analysis data generated based on the current detected by the sensor 31 during a period in which the equipment 10 and / or the equipment 20 are not performing a significant operation (for example, a period in which the equipment 10 and / or the equipment 20 are energized but are not performing a physical operation).

[0095] The diagnosing unit 36 ​​diagnoses an abnormality in the device 10 based on the evaluation result acquired by the acquiring unit 35.

[0096] More specifically, the diagnosis unit 36 ​​diagnoses that there is an abnormality in the device 10 when (1) a representative value of one or more first rates of change acquired by the acquisition unit 35 as one or more first evaluation results is less than a first threshold value and a representative value of one or more second rates of change acquired by the acquisition unit 35 as one or more second evaluation results is greater than or equal to a second threshold value, and (2) a representative value of one or more first rates of change is less than the first threshold value and a representative value of one or more second rates of change is less than the second threshold value, the diagnosis unit 36 ​​diagnoses that there is no abnormality in the device 10.

[0097] Here, the representative value of the first rate of change that is 1 or more may be any value that represents the first rate of change that is 1 or more, and may be, for example, an average value, a maximum value, a minimum value, a median, or a mode of the first rate of change that is 1 or more.

[0098] Furthermore, the representative value of the second rate of change of 1 or more may be any value that represents the second rate of change of 1 or more, and may be, for example, an average value, a maximum value, a minimum value, a median, or a mode of the second rate of change of 1 or more.

[0099] The diagnosis unit 36 ​​may further diagnose, for example, (3) that there is an abnormality in the equipment 20 when the representative value of the first change rate is equal to or greater than a first threshold value and the representative value of the second change rate is less than a second threshold value, (4) that there is no abnormality in the equipment 20 when the representative value of the first change rate is less than the first threshold value, or (5) that there is an abnormality in the equipment 10 and the equipment 20 when the representative value of the first change rate is equal to or greater than the first threshold value and the representative value of the second change rate is equal to or greater than the second threshold value.

[0100] Here, the first threshold value and the second threshold value may be set in advance, or may be set by a user who uses the anomaly detection system 1.

[0101] When the diagnosing unit 36 ​​diagnoses an abnormality in the device 10 and / or an abnormality in the facility 20, the output unit 37 outputs the diagnosis result to the outside.

[0102] The output unit 37 may, for example, be configured to include a display and output the diagnosis result to the outside by displaying an image indicating the diagnosis result on the display, or may, for example, be configured to include a wireless communication interface and output the diagnosis result to the outside by transmitting information indicating the diagnosis result to a mobile terminal owned by a user who uses the anomaly detection system 1.

[0103] <Operation> The operation of the anomaly detection system 1 configured as above will be described below.

[0104] The anomaly detection system 1 performs a learning process to train a machine learning model 34, and a diagnosis process to diagnose an anomaly in the device 10 using the trained machine learning model 34.

[0105] FIG. 5 is a flowchart of the learning process.

[0106] The learning process is started, for example, when a user of the anomaly detection system 1 performs an operation on the anomaly detection device 30 to start the learning process during a period in which the equipment 10 and the facility 20 are operating normally.

[0107] When the learning process is started, the sensor 31 detects the current flowing through the power supply side power line 40 during a first period in which the device 10 and the facility 20 are operating normally (step S100).

[0108] When the sensor 31 detects the current in the first period, the analysis unit 32 generates one or more pieces of analysis data based on the current detected by the sensor 31 (step S110).

[0109] When one or more pieces of analysis data are generated, the learning unit 33 selects one unselected piece of analysis data from the one or more pieces of analysis data that have been generated (step S120).

[0110] Here, unselected analysis data refers to analysis data that has not yet been selected in the process of step S120 in the loop process formed by proceeding to No in the process of step S120 to step S150 described below.

[0111] When unselected analytical data is selected, the learning unit 33 determines whether the selected analytical data is appropriate as data to be used for learning the machine learning model 34 (step S130).

[0112] In the processing of step S130, if it is determined that the selected analysis data is appropriate as data to be used for training the machine learning model 34 (step S130: Yes), the learning unit 33 stores the selected analysis data as one or more first analysis data for training the machine learning model 34 (step S140).

[0113] When the processing of step S140 is completed, or when the processing of step S130 does not determine that the selected analytical data is appropriate as data to be used for training the machine learning model 34 (step S130: No), the learning unit 33 checks whether all of the one or more analytical data generated in the processing of step S110 have been selected (step S150).

[0114] In the process of step S150, if all of the one or more pieces of analysis data have not been selected (step S150: No), the process proceeds to step S120.

[0115] In the processing of step S150, if all of the one or more pieces of analysis data have been selected (step S150: Yes), the learning unit 33 trains the machine learning model 34 using the one or more pieces of first analysis data stored therein (step S160).

[0116] When the process of step S160 ends, the anomaly detection system 1 ends its learning process.

[0117] FIG. 6 is a flowchart of the diagnosis process.

[0118] The diagnostic process is started, for example, after the machine learning model 34 has been learned by the learning process, when a user of the anomaly detection system 1 performs an operation on the anomaly detection device 30 to start the diagnostic process.

[0119] As another example, if the anomaly detection system 1 is a system that appropriately monitors the current detected by the sensor 31, the diagnostic process may be automatically started when the current detected by the sensor 31 satisfies a predetermined condition.

[0120] Alternatively, as another example, the diagnostic process may be automatically initiated periodically.

[0121] When the diagnostic process is started, the sensor 31 detects the current flowing through the power supply side power line 40 during the second period (step S200).

[0122] When the sensor 31 detects the current in the second period, the analysis unit 32 generates one or more pieces of analysis data based on the current detected by the sensor 31 (step S210).

[0123] When one or more pieces of analysis data are generated, the acquisition unit 35 selects one unselected piece of analysis data from the one or more pieces of analysis data that have been generated (step S220).

[0124] Here, unselected analysis data refers to analysis data that has not yet been selected in the process of step S220 in the loop process formed by proceeding to No in the process of step S220 to step S250 described below.

[0125] When unselected analysis data is selected, the acquisition unit 35 determines whether the selected analysis data is appropriate as data to be used by the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20 (step S230).

[0126] In the processing of step S230, if it is determined that the selected analysis data is appropriate as data to be used by the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20 (step S230: Yes), the acquisition unit 35 stores the selected analysis data as one or more second analysis data for the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20 (step S240).

[0127] When the processing of step S240 is completed, or when the processing of step S230 does not determine that the selected analysis data is appropriate as data to be used by the machine learning model 34 to evaluate abnormalities in the equipment 10 and the facility 20 (step S230: No), the acquisition unit 35 checks whether all of the one or more analysis data generated in the processing of step S210 have been selected (step S250).

[0128] If all of the one or more pieces of analysis data have not been selected in the process of step S250 (step S250: No), the process proceeds to step S220.

[0129] In the processing of step S250, if all of the one or more pieces of analysis data have been selected (step S250: Yes), the acquisition unit 35 inputs the one or more pieces of second analysis data stored to the machine learning model 34, has the machine learning model 34 evaluate abnormalities in the equipment 10 and the facility 20 (step S260), and acquires one or more evaluation results output from the machine learning model 34 (step S270).

[0130] When the acquisition unit 35 acquires one or more evaluation results, the diagnoser 36 diagnoses abnormalities in the device 10 and the facility 20 based on the evaluation results acquired by the acquisition unit 35 (step S280).

[0131] When the diagnosing unit 36 ​​diagnoses the device 10 and the facility 20, the output unit 37 outputs the diagnosis result to the outside (step S290).

[0132] When the process of step S290 ends, the anomaly detection system 1 ends its diagnosis process.

[0133] <Consideration> According to the anomaly detection system 1 having the above configuration, it is possible to diagnose an abnormality in the device 10 based on the evaluation result of the evaluation regarding the abnormality in the device 10 including the inverter 13, which is acquired from the trained machine learning model 34. Therefore, it is not necessary to prepare a special rule for the target device 10 in advance.

[0134] In this way, according to the anomaly detection system 1 configured as described above, it is possible to detect an anomaly in the device 10 including the inverter 13 without having to prepare a special rule for the target device 10 in advance.

[0135] (supplement) As described above, the present disclosure has been described based on the embodiment as an example of the technology disclosed in the present application. However, the present disclosure is not limited to the present embodiment. As long as it does not deviate from the gist of the present disclosure, various modifications conceived by a person skilled in the art to the present embodiment, and forms constructed by combining components in the present embodiment and various modifications may also be included within the scope of one or more aspects of the present disclosure.

[0136] (1) The components constituting the anomaly detection device 30 do not necessarily need to be realized within a single housing.

[0137] The anomaly detection device 30 may be realized, for example, by a sensor 31 having a structure that can be attached and detached to the power supply side power line 40 and the output side power line 50, an edge device which is a computer device that realizes an analysis unit 32 and communicates with the sensor 31 via a wired connection, a server which is a computer device that realizes a learning unit 33, a machine learning model 34, an acquisition unit 35, and a diagnosis unit 36 ​​and communicates wirelessly with the edge device, and a personal computer which is a computer device that realizes an output unit 37 and communicates wirelessly with the server.

[0138] (2) In the embodiment, the machine learning model 34 learned by the learning unit 33 has been described as outputting, when new analysis data output from the analysis unit 32 is input, a first degree of change in which the current component of the current detected by the sensor 31 in a first frequency band is evaluated as a first evaluation result, and outputting, as a second evaluation result, a second degree of change in which the current component of the current detected by the sensor 31 in a second frequency band is evaluated as a second evaluation result.

[0139] However, the machine learning model 34 learned by the learning unit 33 does not necessarily have to be limited to a configuration that outputs the first degree of change as the first evaluation result and the second degree of change as the second evaluation result as described above.

[0140] The machine learning model 34 learned by the learning unit 33 may be configured to output, for example, one degree of change that evaluates the current components of the current detected by the sensor 31 in all frequency bands as one evaluation result, or may be configured to output, for example, three or more degrees of change that evaluate the current components of the current detected by the sensor 31 in three or more frequency bands as three or more evaluation results. [Industrial Applicability]

[0141] The present disclosure is widely applicable to systems for detecting abnormalities in devices equipped with inverters. [Explanation of symbols]

[0142] 1. Anomaly detection system 10 equipment 11 Converter 12 Smoothing circuit 13 Inverter 20 Equipment 21 Servo motor 30 Anomaly detection device 31 Sensors 32 Analysis Department 33 Learning Department 34 Machine Learning Models 35 Acquisition Department 36 Diagnostic Department 37 Output section 40 Power supply line 50 Output power line 60 power supply

Claims

1. An abnormality detection system for detecting an abnormality in an apparatus having an inverter, a sensor for detecting a current in either a power supply side power line for supplying power from an external source to the device or an output side power line for supplying power from the inverter to an external source; an analysis unit that generates one or more analysis data based on the current detected by the sensor; an acquisition unit that inputs one or more second analysis data generated by the analysis unit for a current detected by the sensor in a second time period to a machine learning model trained using one or more first analysis data generated by the analysis unit for a current detected by the sensor in a first time period, and causes the machine learning model to perform an evaluation regarding an abnormality of the device, thereby acquiring an evaluation result from the machine learning model; A diagnosis unit that diagnoses an abnormality in the device based on the evaluation result. Anomaly detection system.

2. The analysis unit generates the one or more pieces of analysis data by performing a frequency analysis on the current. The anomaly detection system according to claim 1 .

3. The sensor is a harmonic sensor having a frequency-dependent characteristic in which the detection sensitivity increases in proportion to the frequency. The anomaly detection system according to claim 1 or 2.

4. The current detected by the sensor is the current in the output power line. The anomaly detection system according to claim 1 or 2.

5. The method further includes a learning unit that trains the machine learning model using the one or more first analysis data. The anomaly detection system according to claim 1 or 2.

6. The learning of the machine learning model performed by the learning unit is unsupervised learning. The anomaly detection system according to claim 5 .

7. the evaluation result includes a first evaluation result and a second evaluation result in which at least a part of an evaluation target for the current is different from each other; The diagnosing unit diagnoses an abnormality in the device based on the first evaluation result and the second evaluation result, and further diagnoses an abnormality in a facility to which power is supplied through the output side power line. The anomaly detection system according to claim 1 or 2.

8. the first evaluation result is an evaluation result in which a current component of the current in a first frequency band is evaluated; The second evaluation result is an evaluation result in which a current component of the current in a second frequency band including a frequency band higher than the first frequency band is evaluated. The anomaly detection system according to claim 7.

Images