Drive device, diagnostic device, and diagnostic method

Abstract

This drive device drives a load by means of an electric motor. The drive device comprises an inverter, a controller, and a status identification unit. The inverter drives the electric motor. The controller controls the inverter on the basis of a command value, and outputs information indicating the operation status of the electric motor due to the control. The status identification unit detects an abnormality in the electric motor and a machine portion of the load of the electric motor on the basis of a machine learning model learned in advance by learning processing in which information on the vibration of the electric motor is included in learning data, the information on the vibration of the electric motor, and the operation status.

Description

Drive device, diagnostic device, and diagnostic method

[0001] FIELD Embodiments of the present invention relate to a drive device, a diagnostic device, and a diagnostic method.

[0002] A drive device drives a load using the shaft output of an electric motor. Drive devices are used depending on their application and under various operating conditions. The rotational stability of the electric motor driven by the drive device can be affected by load fluctuations. Furthermore, as the mechanical parts of the electric motor and its load change over time, the rotational stability of the electric motor can decrease. It has sometimes been difficult to determine whether such a decrease in the rotational stability of the electric motor is within the range of fluctuations allowed under appropriate operating conditions or is due to an abnormality caused by the change over time in the mechanical parts.

[0003] Japanese Patent Application Laid-Open No. 2022-20512

[0004] An object of the present invention is to provide a drive device, a diagnostic device, and a diagnostic method that improve the accuracy of estimating abnormalities in mechanical parts.

[0005] A drive device according to an embodiment drives a load using an electric motor. The drive device includes an inverter, a controller, and a state identification unit. The inverter drives the electric motor. The controller controls the inverter based on a command value and outputs information indicating the operating status of the electric motor as a result of the control. The state identification unit detects abnormalities in the electric motor and mechanical parts of the load based on a machine learning model previously trained through a learning process that includes information about the vibration of the electric motor in learning data, the information about the vibration of the electric motor, and the operating status.

[0006] 1 is a configuration diagram of a drive system according to an embodiment. 2 is a schematic configuration diagram of a drive device according to an embodiment. 3 is a functional block diagram showing an example of a drive device according to an embodiment. 4 is a functional block diagram showing an example of a control device according to an embodiment. 5 is a diagram for explaining a rolling process in a hot rolling line according to an embodiment. 6 is a diagram for explaining a rolling process in a hot rolling line according to an embodiment. 7 is a diagram for explaining definitions of operation patterns (pattern classification) according to an embodiment. 8 is a diagram for explaining vibrations observed when driven with a specific operation pattern according to an embodiment. 9 is a diagram for explaining compression processing of frequency component features according to an embodiment. 10 is a schematic configuration diagram of a machine learning model according to an embodiment. 11 is a diagram for explaining a machine learning model according to an embodiment.

[0007] Hereinafter, a drive device, a diagnostic device, and a diagnostic method according to embodiments will be described with reference to the drawings. In the following description, a drive system that performs speed matching control will simply be referred to as a drive system. Components having the same or similar functions will be assigned the same reference numerals. Duplicate descriptions of those components may be omitted. Electrical connection may also be simply referred to as "connected."

[0008] FIG. 1 is a configuration diagram of a drive system 1 according to an embodiment.

[0009] The drive system 1 includes, for example, a master device 10 and drive devices 30A to 30D. The electric motors 21A to 21D are examples of electric motors to be controlled by the drive system 1.

[0010] Figure 1 also shows an example of a manufacturing facility related to the drive system 1. The drive system 1 in the manufacturing facility shown in Figure 1 includes, for example, rolling stands STA to STD for transporting objects. Rolling stand STA is equipped with an electric motor 21A, second rolling stand ST2 is equipped with an electric motor 21B, third rolling stand ST3 is equipped with an electric motor 21C, and rolling stand STD is equipped with an electric motor 21D. Rolling stands STA to STD are driven by the power of electric motors 21A to 21D, respectively.

[0011] The electric motor 21A is driven by a drive unit 30A. The electric motor 21B is driven by a drive unit 30B. The electric motor 21C is driven by a drive unit 30C. The electric motor 21D is driven by a drive unit 30D. When there is no need to distinguish between the electric motors 21A and 21D, they are simply referred to as the electric motors 21. When there is no need to distinguish between the electric motors 30A and 30D, they are simply referred to as the drive units 30. When each drive unit 30 simultaneously operates each electric motor 21, the rolling stands STA to STD are driven by the power output by each electric motor 21. This allows a relatively long object in the first direction to be transported at least in the elongation direction (first direction) of the object. This manufacturing equipment may be applied, for example, to a rolling process for steel plates (objects). The following description shows an example of application to a steel plate rolling process, and provides a detailed explanation of one example.

[0012] The master unit 10 (high-level controller) sends command values ​​to each drive unit 30 via the network NW and controls each drive unit 30 using the command values ​​to adjust the operating conditions when transporting objects by each electric motor 21. The master unit 10 causes each drive unit 30 to adjust (synchronize) the rotation speed of each electric motor 21, thereby maintaining a good transport condition for the objects.

[0013] A more specific example of each drive device 30 will be described with reference to Figures 2 to 4. Figure 2 is a schematic configuration diagram of the drive device 30 of the embodiment. Figure 3 is a functional block diagram showing an example of the drive device 30 of the embodiment. Figure 4 is a functional block diagram showing an example of the control device 31 of the embodiment.

[0014] The drive device 30 includes, for example, a control device 31 and a state identification unit 32 .

[0015] The control device 31 includes, for example, a power converter 311 and a control unit 312. 、 The system includes a communication processing unit 313 (FIG. 4), a storage unit 314 (denoted as STRAGE in FIG. 4), and an input / output unit 315 (FIG. 4).

[0016] The power converter 311 supplies power to the windings (not shown) of the electric motor 21 for transporting an object.

[0017] To control the AC motor 21 using the DC power POW, the power converter 311 functions as a so-called inverter to convert the DC power POW into AC power. The above is an example, and is not limiting. For example, the motor 21 may be a DC motor. To control the DC motor 21 using the AC power POW, the power converter 311 functions as a so-called converter to convert the AC power POW into DC power. The specifications of the power converter 311 are determined by the specifications of the motor 21 and the specifications of the power supply, and therefore may be a power converter of a type other than those described above. In the following description, a case will be described in which the motor 21 is an AC motor.

[0018] The power converter 311 (referred to as a PC in FIG. 2 ) includes, for example, a plurality of semiconductor switches (not shown in FIG. 2 ) and their drive circuits. For example, the plurality of semiconductor switches are configured as a full-bridge or half-bridge type. The circuit configuration of the power converter 311 is not limited to this and may be changed as appropriate. The type of the plurality of semiconductor switches may be, for example, either an IGBT (Insulated Gate Bipolar Transistor) or an FET (Field-Effect Transistor). However, the type is not limited to this and other types of semiconductor switches may be applied, or diodes may be appropriately combined.

[0019] The communication processing unit 313 communicates with the master device 10 via the network NW, and can supply various information acquired from the master device 10 to the control unit 312. The communication processing unit 313 may communicate with the state identification unit 32 and acquire the identification result of the state identification unit 32.

[0020] The storage unit 314 is realized by a ROM, a RAM, a HDD, a flash memory, etc. The storage unit 314 is allocated a storage area for storing various setting information for causing the control unit 312 to function, basic programs such as an OS, application programs, etc.

[0021] The input / output unit 315 receives, for example, information on the shaft position of the electric motor 21 and information on the output state of the power converter 311 as input information, and outputs a gate signal GP. The input / output unit 315 may include, for example, a display unit such as a liquid crystal display that displays various information, and an operation detection unit. The display unit and operation detection unit may be configured as a touch panel that is a combination of them.

[0022] The control unit 312 sends a gate signal GP to the power converter 311 via the input / output unit 315 to control the power converter 311 .

[0023] The control unit 312 includes a function of executing a software program (described below). The control unit 312 is connected to a bus BUS together with a communication processing unit 313, a storage unit 314, and an input / output unit 315. The control unit 312 executes a software program to form some or all of its functions. The software program for realizing the functions of the control unit 312 may be stored in the storage unit 314 in advance, or may be downloaded to the storage unit 314 from an external device or portable storage medium (not shown), or via a communication line.

[0024] 3, the state identification unit 32 acquires information indicating the driving conditions such as speed and torque from the control device 31. The speed may be a speed reference designated by the master device 10, a speed detection value detected by a speed sensor (not shown), or a speed estimation value generated within the control device 31. The torque may be a torque reference generated according to the speed or the like.

[0025] Furthermore, the state identification unit 32 acquires the vibration detection result from the sensor 21S. If an accelerometer is applied to the sensor 21S, the vibration can be detected as acceleration. The sensor 21S may output the detection result of acceleration components in three orthogonal axial directions. The type of the sensor 21S is not limited to this. The sensor 21S is disposed on the housing of the electric motor 21 or on the stand on which the electric motor 21 is disposed, and detects vibration at the position where it is disposed.

[0026] The state identification unit 32 includes a driving pattern identification unit 321 (driving pattern), an FFT calculation processing unit 322 (FFT), a machine learning model 323, a state determination unit 324 (state determination), and a memory unit (not shown).

[0027] The driving pattern identification unit 321 receives information such as speed and load (torque) from the control device 31 via communication and identifies a driving pattern based on the information. The FFT calculation processing unit 322 converts information (vibration information) supplied from the sensor 21S into time-series information and performs FFT calculation on the time-series information (vibration information), as well as upstream and downstream processing. The FFT calculation and upstream and downstream processing will be described later. The machine learning model 323 includes a neural network formed with multiple layers. The machine learning model 323 is configured to obtain desired characteristics by pre-training the machine learning model 323 using training data. The state determination unit 324 determines a state based on input data to the machine learning model 323, output data generated by the machine learning model 323, results identified by the machine learning model 323, and the like. These will be described in detail later.

[0028] The state identification unit 32 includes, for example, a function of executing arithmetic processing, as described below. The storage unit of the state identification unit 32 is realized by a ROM, RAM, HDD, SSD, flash memory, etc. The storage unit of the state identification unit 32 may be configured internally, or may be configured as a partial area of ​​the storage unit 314 described above. Each unit of the state identification unit 32 writes data to a predetermined area of ​​the storage unit and stores the data when acquiring data, when temporarily storing the calculation process or calculation results, etc. Each unit of the state identification unit 32 reads the data as needed and performs the desired processing. For example, a program for implementing the function of the state identification unit 32 may be stored in the storage unit of the state identification unit 32, or may be stored in the storage unit 314 described above. Some or all of the functions of the state identification unit 32 are implemented by executing a software program.

[0029] The rolling process in the hot rolling line of the embodiment will be described with reference to FIGS. 5 and 6. FIG.

[0030] 5 and 6 are diagrams for explaining the rolling process in a hot rolling line according to an embodiment. In a hot rolling line, red-hot iron is pulled through the gap between two rollers (cylinders) included in a rolling stand STA (FIG. 1) or the like to produce the desired thickness. During this process, the speed and torque of the electric motor 21 for feeding the iron vary depending on the material and finishing method, and the operating pattern. In the following example, three types of material, A to C, are used as examples to explain the differences between them.

[0031] The timing chart in the upper part (a) of Figure 5 shows three materials with different rolling process requirements, each subjected to a rolling process. Material A has an operating pattern in which the speed of the electric motor 21 is fast and the torque (load) is small (see Case 1 in Figure 6). Material B has an operating pattern in which the speed of the electric motor 21 is slow and the torque (load) is large (see Case 2 in Figure 6). Material C has an operating pattern in which the speed of the electric motor 21 is fast and the torque (load) is medium.

[0032] The timing chart in the lower part (b) of Figure 5 shows one rolling process of the above-mentioned material B, with the time axis expanded. At time t0, material B begins to be applied to the rollers of the rolling stand STA. Thereafter, material B is moved in a predetermined direction without applying a load (torque) to the electric motor 21 until time t1. At time t1, the electric motor 21 of the rolling stand STA applies a predetermined load (torque) to move material B in the predetermined direction. From time t1 to time t2, both the speed and load (torque) of the electric motor 21 in the rolling stand STA are increased. At time t2, the electric motor 21 of the rolling stand STA is adjusted so that both the speed and load (torque) are constant, and the rolling process of material B proceeds under these conditions.

[0033] As the rolling of material B progresses and the material approaches the trailing edge (time t4), the load (torque) of the electric motor 21 is set to 0, and the speed is gradually reduced.

[0034] During such a rolling process, vibrations occur in the mechanical parts of the rolling stand STA. These vibrations occur even when there is no abnormality in the rolling stand STA, and the magnitude of the vibrations tends to increase during periods when a load (torque) is applied.

[0035] Incidentally, if the rolling process is carried out while there is an abnormality in the mechanical parts of the rolling stand STA, the above vibrations tend to become even larger, but it can be difficult to determine whether or not an abnormality has occurred from the magnitude (amplitude) of the vibrations. Therefore, in this embodiment, vibrations that occur during the period from time t2 to time t4 when adjustments are being made to keep both the speed and the load (torque) constant are detected, and the data is recorded as time-series information.

[0036] 6, the vibration components differ depending on the speed and load of the electric motor 21. For example, the frequency components of the vibration in a high-speed, low-torque operation pattern (Case 1) are compared with the frequency components of the vibration in a low-speed, high-torque operation pattern (Case 2).

[0037] It can be seen that the distribution of frequencies where energy is concentrated in the frequency spectrum is different from one another. The drive device 30 utilizes the fact that the frequency spectrum changes depending on the driving pattern in the process of identifying the driving situation. This makes it easier to evaluate the driving situation compared to a comparative example that does not utilize this feature.

[0038] The evaluation of the driving situation using the characteristics of each driving pattern according to the embodiment will be described with reference to Fig. 7 and Fig. 8. Fig. 7 is a diagram for explaining the definition (pattern classification) of the driving patterns according to the embodiment.

[0039] As shown in Figure 7, speed and load (torque) are assigned to two orthogonal axes, and the coordinate space including the two axes is divided into a plurality of cells. For example, for the speed axis, the range from 0% to 100% rated speed is divided into, for example, five equal intervals. For the load (torque) axis, the range from 0% to 150% rated load (exceeding 100%) is divided into, for example, five intervals. By dividing in this way, the coordinate space including the above two axes can be divided into 25 cells.

[0040] Each of the above cells is associated with a "driving pattern," and identification information is assigned to each association. For example, identification information ranging from 0 to 24 is assigned to each pair of a cell and a "driving pattern."

[0041] 8 is a diagram for explaining vibrations observed when the vehicle is driven in a specific driving pattern according to the embodiment. As shown in FIG. 8, the vibrations observed when the vehicle is driven in a specific driving pattern have different characteristics from the vibrations observed when the vehicle is driven in other driving patterns.

[0042] A selected specific driving pattern is shown in Figure 8(a). The driving pattern shown here is identified by identification number 19 (Figure 7). It corresponds to the fourth cell from the bottom on the torque axis and the fifth cell from the bottom on the time axis.

[0043] (b) in Figure 8 shows vibration data observed during operation with this operating pattern. This vibration data was sampled at a sampling period corresponding to the number of data points suitable for FFT calculation processing and the required frequency resolution. Time is assigned to the horizontal axis, and signal magnitude is assigned to the vertical axis. Data within the range extracted by the time window of the period specified above is subjected to FFT calculation processing. Note that a window function such as a Hanning window may be applied to the sampled data prior to FFT calculation processing. Window function processing is an example of processing performed before FFT calculation processing. FFT calculation processing is performed on the sampled data or data to which a window function has been applied, to obtain a frequency spectrum as shown in (c) in Figure 8. Frequency is assigned to the horizontal axis, and signal magnitude is assigned to the vertical axis.

[0044] Increasing the number of data points in the FFT calculation process increases the frequency resolution, but also increases the calculation load. If the number of data points in the FFT calculation process is limited, the frequency resolution decreases, and it becomes necessary to properly detect the energy of frequency components that do not match the frequencies specified by the frequency resolution.

[0045] The compression process of frequency component features according to the embodiment will be described with reference to Fig. 9. Fig. 9 is a diagram for explaining the compression process of frequency component features according to the embodiment. In general, increasing the frequency resolution of the FFT calculation results increases the accuracy of converting time series data into frequency domain information. On the other hand, the judgment of the similarity of frequency spectra becomes stricter, resulting in a lower evaluation of the similarity.

[0046] Therefore, the FFT processor 322 may reduce the number of poles contained in the frequency spectrum of the FFT processing result calculated by itself, thereby somewhat reducing the strictness of the similarity evaluation and preventing the omission of a determination that the signal is similar. For example, the FFT processor 322 may compress the waveform of the frequency spectrum by taking the sum of the squares of several consecutive frequency components from the FFT processing result. In the example shown in Figure 9, the sum of the squares of six consecutive frequency components is taken and the results are plotted on the same graph. In this way, the compression operation described above allows a waveform indicating the characteristics of the frequency spectrum to be obtained from the frequency spectrum of the FFT processing result. It can be seen that the number of poles is clearly reduced compared to the waveform of the original frequency spectrum.

[0047] The machine learning model 323 of the embodiment will be described with reference to Fig. 10A and Fig. 10B. Fig. 10A is a schematic configuration diagram showing the machine learning model 323 of the embodiment and its surroundings. Fig. 10B is a configuration diagram of the machine learning model 323 of the embodiment.

[0048] The machine learning model 323 includes, for example, an encoder 3231, a decoder 3232, and an encoder 3233.

[0049] The encoder 3231 compresses the spectral data (hereinafter referred to as data X) that is output information from the FFT calculation processing unit 322, and outputs the compressed data to the encoder 3233 and the decoder 3232. This compressed data is an example of a feature.

[0050] The decoder 3232 decodes the compressed data (features) input from the encoder 3231 to generate reconstructed data, and outputs the reconstructed data (data Xest) to the state determination unit 324. The combination of the encoder 3231 and the decoder 3232 is an example of an autoencoder.

[0051] The encoder 3233 processes the compressed data (features) input from the encoder 3231 to generate final layer data and intermediate layer data. The final layer data corresponds to the output value of the neural network, and the intermediate layer data corresponds to data of all layers other than the final layer data and input layer data.

[0052] For example, the machine learning model 323 may be configured as shown in FIG. 10B . The encoder 3231 is configured as a neural network including an intermediate layer 32311 located after the input layer and an output layer 3231O. The number of intermediate layers and the number of neurons in each layer may be determined appropriately as needed. The intermediate layer 32311 includes a fully connected layer (Dense) 32311D and an activation function processing unit 32311A located after it. The activation function illustrated here is ReLU (Rectified Linear Unit). The type of activation function can be selected appropriately. The output layer 3221O includes a fully connected layer 3231OD and an activation function processing unit 3231OA located after it. The activation function illustrated here is ReLU. The type of activation function can be selected appropriately. The output from the activation function processing unit 3231OA becomes the output of the encoder 3231.

[0053] The decoder 3232 is configured as a neural network including an intermediate layer 32321 located after the output layer 3221O of the encoder 3231, an intermediate layer 32322, and an output layer 3232O. The number of intermediate layers and the number of neurons in each layer in the decoder 3232 may be determined appropriately as needed. The intermediate layer 32321 includes a fully connected layer 32321D and an activation function processing unit 32321A located after it. The activation function shown here is ReLU. The type of activation function can be selected appropriately. The intermediate layer 32322 includes a fully connected layer 32322D and an activation function processing unit 32322A located after it. The activation function shown here is ReLU. The type of activation function can be selected appropriately. The output layer 3222O includes a fully connected layer. The output from the fully connected layer 3232OD becomes the output of the decoder 3232.

[0054] The encoder 3233 is configured as a neural network including an intermediate layer 32331 located after the output layer 3221O of the encoder 3231, an intermediate layer 32332, and an output layer 3233O. The number of intermediate layers and the number of neurons in each layer in the encoder 3233 may be determined appropriately as needed. The intermediate layer 32331 includes a fully connected layer 32331D and an activation function processing unit 32331A located after it. The activation function shown here is ReLU. The type of activation function can be selected appropriately. The intermediate layer 32332 includes a fully connected layer 32332D and an activation function processing unit 32332A located after it. The activation function shown here is ReLU. The type of activation function can be selected appropriately. The output layer 3222O includes softmax function calculation processing. The output from the fully connected layer 3232OD becomes the output of the decoder 3232. The result of the softmax function calculation is a probability.

[0055] The machine learning model 323 is trained in advance using appropriate training data, thereby making it possible to reproduce input data (data X) of the machine learning model 323. In the machine learning model 323, the decoder 3232 outputs data Xest estimated from the data X. If the machine learning model 323 has been properly trained on the predetermined data X, the data Xest will be approximate to the data X.

[0056] Therefore, it is preferable to determine the similarity between data X and data Xest using a predetermined evaluation function. For example, the mean square error (MSE) of data Xest with respect to data X may be used as the evaluation function. The state determination unit 324 downstream of the machine learning model 323 may calculate the mean square error (MSE) of data Xest with respect to data X, and use the result to evaluate the normality of the state indicated by data X.

[0057] Furthermore, the machine learning model 323 is trained in advance using appropriate training data, thereby identifying the data X input to the machine learning model 323, and the identification result is output by the encoder 3233. If the machine learning model 323 has been properly trained for the predetermined data X, it becomes possible to identify the data X from the identification result.

[0058] As described above, the machine learning model 323 includes an autoencoder (encoder 3231 and decoder 3232).

[0059] The state determination unit 324 downstream of the machine learning model 323 can use a desired determination process from among several determinations. For example, the first determination process may include a determination (first determination) based on input information (data X) to the autoencoder and output information (data Xest) generated by the autoencoder from the input information (data X). In this case, the state determination unit 324 can perform the first determination based on the input information (data X) to the autoencoder and the output information (data Xest) generated by the autoencoder from the input information (data X).

[0060] If the machine learning model 323 is configured to include an encoder 3233 (estimation unit) in addition to the above-mentioned autoencoder, the state determination unit 324 will also be able to perform the following second and third determination processes.

[0061] For example, the second determination process may include a determination (second determination) based on identification information (driving pattern) of the driving status of the drive device 30 and the estimated information generated by the encoder 3233. In this case, the state determination unit 324 can perform the second determination based on the identification information of the driving status of the drive device 30 and the estimated information generated by the encoder 3233.

[0062] The third determination process is a combination of the first and second determination processes. In this case, the state determination unit 324 performs the first and second determinations and detects an abnormality in the mechanical parts of the electric motor 21 and its load based on the results of the first and second determinations.

[0063] Next, learning of the machine learning model 323 will be described. When learning the machine learning model 323 in advance, it is preferable to use vibration data observed using the drive system 1 in a good state and vibration data observed using the drive system 1 in an ungood state, respectively. As this "ungood state," it is preferable to use a state simulating a typical failure or deterioration, and evaluate the events that occur in that case.

[0064] More specifically, it is advisable to use as learning data the results of FFT calculation processing of vibration data detected when the rolling stand STA of the drive system 1 is operating normally, and the results of FFT calculation processing of vibration data detected when it is operated in a faulty state.

[0065] In the embodiment, the machine learning model 323 is associated with each driving pattern. Learning of the machine learning model 323 may be performed for each driving pattern using learning data corresponding to the driving pattern, thereby forming a machine learning model for each driving pattern.

[0066] In this learning process, it is advisable to use an evaluation index that considers a data set that has a better evaluation value according to an evaluation function such as the mean square error (MSE) of the data Xest for the data X and that has a higher degree of certainty in being identified as being due to a predetermined driving pattern by the evaluation by the encoder 3233 as more suitable.

[0067] According to the above embodiment, the drive device 30 drives its load using the electric motor 21. The drive device 30 includes a power converter 311 (inverter) of the control device 31, a control unit 312 (controller), and a state identification unit 32. The power converter 311 is an inverter that drives the electric motor. The control unit 312 is a controller that controls the power converter 311 based on a command value to drive the electric motor 21 and outputs information indicating the operating status of the electric motor 21 as a result of the control. The state identification unit 32 detects abnormalities in the mechanical parts of the electric motor 21 and its load based on a machine learning model 323 that has been trained in advance through a learning process in which information indicating the rotation stability of the electric motor 21 is included in learning data, the information indicating the rotation stability of the electric motor 21, and the operating status. This allows the drive device 30 to further improve the accuracy of estimating abnormalities in the mechanical parts.

[0068] The information indicating the operating status of the electric motor 21 may include any one of a speed reference based on a command value, a torque reference, and a detected speed. By defining a plurality of operating patterns that can be identified using any one of the speed reference based on the command value, the torque reference, and the detected speed, the state identification unit 32 may detect an abnormality in the electric motor 21 and the mechanical parts of its load using a machine learning model 323 that corresponds to the plurality of operating patterns.

[0069] The information indicating the rotation stability of the electric motor 21 includes information on the magnitude of each frequency component of the frequency spectrum. This frequency spectrum may be based on the frequency spectrum included in the time-series information obtained by detecting the vibration of the electric motor 21.

[0070] As described above, the speed and required torque of the electric motor 21 in the drive device 30 vary depending on the operating conditions. Even when rolling iron, the operating conditions vary depending on the material, such as a high speed and low torque, or a low speed and high torque. The vibration of the electric motor 21 and other components varies depending on the operating conditions.

[0071] In the comparative example, when detecting the occurrence of a mechanical abnormality in the electric motor 21 or the like from the magnitude (peak value) of the vibration, it was necessary to set the threshold level used for the determination relatively high so as not to erroneously determine that various operating conditions are abnormal. With such a determination method based on the magnitude of the vibration, it was not possible to detect an abnormality until a state in which large vibrations were generated occurred.

[0072] In contrast, the drive device 30 of the embodiment can recognize driving conditions by dividing them into patterns, and utilizes this driving condition information to carry out deterioration diagnosis of the mechanical parts.

[0073] For example, in the speed control of the drive device 30, a speed command is input from the master device 10 or externally, a torque command is generated to follow this speed command, and the necessary torque is output to follow this torque command, thereby allowing the drive device 30 to continue operation.

[0074] Furthermore, the characteristics of data such as the current flowing through the motor 21 and the vibration of the motor 21 will differ depending on the operating conditions of the drive device 30. For example, when focusing on the relationship between the speed and torque of the motor 21 in terms of operating conditions, as the torque increases, the amplitude generally increases, and as the speed changes, the frequency components and amplitude change. It is advisable to define operating patterns so that differences in load (torque) and speed can be identified.

[0075] To detect an abnormality in the electric motor 21, vibration data of the electric motor 21 is input to the drive device 30. A machine learning model 323 is created that learns a waveform obtained by frequency analysis of the vibration from the driving pattern of the drive device 30 and the vibration data. If the detected data differs from the output of the machine learning model 323 (autoencoder), it is recognized as an abnormality, and information indicating when an abnormality is detected or when a deterioration trend is expected is output.

[0076] Since the occurrence of vibrations differs depending on the installation conditions of each device (machine), it is advisable to obtain data after installing and operating each device, and use that data as learning data to perform the learning process of the machine learning model 323.

[0077] The results of the FFT calculation process (FFT data) based on the vibration data of the electric motor 21 are provided to the IN terminal of the machine learning model 323 in FIG. 10B, and reproduction data of the FFT data is obtained from the OUT1 terminal in FIG. 10B. Furthermore, the estimated results of the driving pattern are obtained from the OUT2 terminal in FIG. 10B. The state determination unit 324 performs an analysis based on the results of the FFT calculation process (FFT data), the reproduced data of the FFT data, and the estimated results of the driving pattern. The state determination unit 324 may diagnose the deterioration tendency of the drive device 30 based on the above analysis results, including the determination of the driving pattern based on the command value, and the differences in the reproduced data.

[0078] According to at least one embodiment described above, a drive device drives a load using an electric motor. The drive device includes an inverter, a controller, and a state identification unit. The inverter drives the electric motor. The controller controls the inverter based on a command value and outputs information indicating the operating status of the electric motor as a result of the control. The state identification unit detects an abnormality in the mechanical parts of the electric motor and its load based on a learning model previously trained through a learning process in which information about the vibration of the electric motor is included in learning data, the information about the vibration of the electric motor, and the operating status. This can further improve the accuracy of estimating an abnormality in the mechanical parts.

[0079] The functions described herein and implemented by components such as the control unit 312 and the state identification unit 32 may be implemented in a circuit or processing circuitry. The circuit or processing circuitry may include a general-purpose processor, an application-specific processor, an integrated circuit, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a central processing unit (CPU), a graphics processing unit (GPU), a conventional circuit, and / or a combination thereof, programmed to implement the described functions. A processor implementing the above functions includes transistors and other circuits and is considered to be a circuit or processing circuitry. A processor for implementing the above functions may include or be a programmable processor that executes a program stored in a memory and / or a programmable device that can be reconfigured by data stored in a memory. In this specification, a circuit, unit, or means refers to hardware that is programmed to implement the described functions or hardware that executes the functions. The hardware may be any hardware disclosed herein or any hardware known to be programmed or capable of performing the described functions. If the hardware is a processor considered to be a type of circuit, the circuit, means, or unit is a combination of hardware and software used to configure the hardware and / or processor.

[0080] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention described in the claims and their equivalents.

[0081] For example, the control device 31 and the state identification unit 32 may be configured with separate processors, or a processor common to the control device 31 and the state identification unit 32 may perform the respective processes.

[0082] 1... Drive system 21, 21A, 21B, 21C, 21D... Electric motor 30, 30A, 30B, 30C, 30D... Drive device 31... Control device 32... State identification unit 311... Power converter (inverter) 312... Control unit (controller) 321... Operation pattern identification unit (operation pattern) 322... FFT calculation processing unit (FFT) 323... Machine learning model 324... State determination unit (state determination) 3231, 3233... Encoder 3232... Decoder

Claims

1. A drive device that drives a load using an electric motor, comprising: an inverter that drives the electric motor; a controller that controls the inverter based on a command value to drive the electric motor and outputs information indicating the operating status of the electric motor as a result of the control; and a state identification unit that detects abnormalities in the mechanical parts of the electric motor and its load based on a machine learning model that has been trained in advance by a learning process in which information indicating the rotation stability of the electric motor is included in learning data, and the information indicating the rotation stability of the electric motor and the operating status.

2. The drive device according to claim 1, wherein the information indicating the operating status of the electric motor includes any one of a speed reference based on the command value, a torque reference, and a detected speed.

3. The drive device of claim 2, wherein a plurality of operating patterns are defined that can be identified using any of a speed reference, a torque reference, and a detected speed based on the command value, and the state identification unit detects abnormalities in the mechanical parts of the electric motor and its load using the machine learning model corresponding to the plurality of operating patterns.

4. The drive device according to claim 1, wherein the machine learning model includes an autoencoder.

5. The drive device according to claim 4, wherein the state identification unit performs a first determination based on input information to the autoencoder and output information generated by the autoencoder from the input information.

6. The drive device according to claim 4, wherein the machine learning model includes an estimation unit that identifies features generated by the autoencoder.

7. The drive device according to claim 6, wherein the state identification unit performs a second determination based on the identification information of the driving situation and the estimation information generated by the estimation unit.

8. The drive device of claim 1, wherein the machine learning model includes an autoencoder and an estimation unit that identifies features generated by the autoencoder, and the state identification unit performs a first judgment based on input information to the autoencoder and output information generated by the autoencoder from the input information, and a second judgment based on identification information of the driving situation and estimation information generated by the estimation unit, and detects abnormalities in the mechanical parts of the electric motor and its load based on the results of the first judgment and the results of the second judgment.

9. The drive device according to claim 1, wherein the information indicating the rotational stability of the motor includes information on the magnitude of each frequency component of the frequency spectrum of time-series information obtained by detecting vibrations of the motor.

10. A diagnostic device for a drive system that drives a load using an electric motor, comprising: an inverter that drives an electric motor; and a controller that controls the inverter based on a command value to drive the electric motor and outputs information indicating the operating status of the electric motor as a result of the control; the diagnostic device further comprising: a machine learning model that has been trained in advance by a learning process in which information indicating the rotation stability of the electric motor is included in learning data; and a condition identification unit that detects abnormalities in the mechanical parts of the electric motor and its load based on the information indicating the rotation stability of the electric motor and the operating status.

11. The diagnostic device according to claim 10, wherein the information indicating the operating condition of the electric motor includes any one of a speed reference, a torque reference, and a detected speed based on the command value.

12. A diagnostic device as described in claim 11, wherein a plurality of operating patterns that can be identified using any of a speed reference, a torque reference, and a detected speed based on the command value are defined, and the state identification unit detects abnormalities in the mechanical parts of the motor and its load using the machine learning model corresponding to the plurality of operating patterns.

13. The diagnostic device according to claim 10, wherein the machine learning model includes an autoencoder.

14. The diagnostic device according to claim 13, wherein the state identification unit performs a first determination based on input information to the autoencoder and output information generated by the autoencoder from the input information.

15. The diagnostic device according to claim 13, wherein the machine learning model includes an estimation unit that identifies features generated by the autoencoder.

16. The diagnostic device according to claim 15, wherein the state identification unit performs a second determination based on the identification information of the driving situation and the estimation information generated by the estimation unit.

17. The diagnostic device described in claim 10, wherein the machine learning model includes an autoencoder and an estimation unit that identifies features generated by the autoencoder, and the state identification unit performs a first judgment based on input information to the autoencoder and output information generated by the autoencoder from the input information, and a second judgment based on identification information of the driving situation and estimation information generated by the estimation unit, and detects abnormalities in the mechanical parts of the electric motor and its load based on the results of the first judgment and the results of the second judgment.

18. A diagnostic method for a drive system including an inverter that drives an electric motor, and a controller that controls the inverter based on a command value to drive the electric motor and output information indicating the operating status of the electric motor as a result of the control, and which drives a load using the electric motor, the diagnostic method including a process of detecting abnormalities in mechanical parts of the electric motor and its load based on a machine learning model that has been trained in advance by a learning process in which information indicating the rotation stability of the electric motor is included in learning data, the information indicating the rotation stability of the electric motor, and the operating status.

19. The diagnostic method described in claim 18, further comprising: defining a plurality of operating patterns identifiable using any one of a speed reference, a torque reference, and a detected speed based on the command value; and detecting abnormalities in the mechanical parts of the motor and its load using the machine learning model corresponding to the plurality of operating patterns.

20. The diagnostic method described in claim 19, wherein the machine learning model identifies an autoencoder and features generated by the autoencoder, performs a first judgment based on input information to the autoencoder and output information generated by the autoencoder from the input information, and a second judgment based on identification information of the driving situation and the generated estimated information, and detects abnormalities in the mechanical parts of the electric motor and its load based on the results of the first judgment and the results of the second judgment.

Images