Estimation device, power conversion device, motor drive device, and refrigeration cycle application apparatus

Abstract

Provided is an estimation device (410) all or part of which is mounted to a power conversion device (1), said estimation device comprising: a preprocessing unit (411) that converts, into a second physical amount, a first physical amount that indicates an operating state of a configuration provided to the power conversion device (1) or a configuration provided to a device including the power conversion device (1); a feature amount extraction unit (412) that extracts a feature amount from the second physical amount; and an estimation unit (413) that diagnoses or predicts a configuration on the basis of the feature amount. The second physical amount is a physical amount to be input to the feature amount extraction unit (412) and with which the feature amount extraction unit (412) is able to extract the feature amount.

Description

Estimation Device, Power Conversion Device, Motor Drive Device, and Refrigeration Cycle Application Equipment

[0001] The present disclosure relates to an estimation device for estimating the state of a power conversion device, a power conversion device, a motor drive device, and a refrigeration cycle application device.

[0002] Conventionally, high-precision diagnosis, estimation, etc. of devices have been performed using machine learning models and the like. In the field of computer science, since there is a large amount of data on the Internet, it is easy to create a machine learning model. On the other hand, in the field of power electronics such as motor control, it is difficult to acquire a large amount of data such as current waveforms and voltage waveforms from the Internet. As for a system for performing motor control and the like, for example, Patent Document 1 discloses a technique for estimating the state of a system that is difficult to model by a transfer function or the like by using a statistic as a feature quantity. Also, as another method different from the technique disclosed in Patent Document 1, a method of generating a machine learning model within a possible range with a small amount of data and extracting feature quantities can also be considered.

[0003] Japanese Patent No. 6899897

[0004] However, according to the above conventional techniques, there is a problem that a slight change in the shape of waveforms such as current and voltage cannot be extracted from the statistic. Also, there is a problem that a machine learning model generated with a small amount of data has low versatility.

[0005] The present disclosure has been made in view of the above, and an object thereof is to obtain an estimation device capable of improving the estimation accuracy of the state of a device for performing motor control.

[0006] To solve the aforementioned problems and achieve the objectives, this disclosure provides an estimation device that is mounted in whole or in part on a power converter. The estimation device comprises a preprocessing unit that converts a first physical quantity indicating the operating state of the configuration of the power converter or the configuration of a device including the power converter into a second physical quantity; a feature extraction unit that extracts feature quantities from the second physical quantity; and an estimation unit that performs a diagnosis or prediction about the configuration based on the feature quantities. The second physical quantity is a target physical quantity input to the feature extraction unit and is a physical quantity from which the feature extraction unit can extract feature quantities.

[0007] The estimation device described herein has the effect of improving the accuracy of state estimation for devices that perform motor control.

[0008] Figure showing an example configuration of a power conversion device including all of the estimation device according to Embodiment 1. Figure showing an example of operation of the estimation device according to Embodiment 1. Figure showing an example of the first process performed by the pre-processing unit of the estimation device according to Embodiment 1 as a process of converting a first physical quantity to a second physical quantity. Figure showing an example of the second process performed by the pre-processing unit of the estimation device according to Embodiment 1 as a process of converting a first physical quantity to a second physical quantity. Figure showing an example of the third process performed by the pre-processing unit of the estimation device according to Embodiment 1 as a process of converting a first physical quantity to a second physical quantity. Figure showing an example of YAMNet, a pre-trained model used in the estimation device according to Embodiment 1. Figure showing the operation of the estimation device according to Embodiment 1. Figure 1 shows an example of the hardware configuration for realizing the estimation device provided in the control unit of the power converter according to Embodiment 1. Figure 2 shows an example of the configuration of a power converter including a part of the estimation device according to Embodiment 2. Figure 3 shows an example of operation of the estimation device according to Embodiment 2. Figure 4 shows an example of diagnosis by the estimation device according to Embodiment 3. Figure 4 shows an example of the estimation device according to Embodiment 4 being applied to an air conditioner. Figure 5 shows an example of the estimation device according to Embodiment 5 performing fault diagnosis of the sensor of an air conditioner. Figure 6 shows an example of the configuration of equipment applied to a refrigeration cycle.

[0009] The estimation device, power conversion device, motor drive device, and refrigeration cycle application equipment according to embodiments of this disclosure will be described in detail below with reference to the drawings.

[0010] Embodiment 1. Figure 1 shows an example configuration of a power conversion device 1 including the entire estimation device 410 according to Embodiment 1. The power conversion device 1 converts a first AC voltage supplied from a power supply 110, which is an AC power source, into a second AC voltage having a desired amplitude and phase, and supplies it to a motor 314. The power supply 110 may be a single-phase AC power source or a three-phase AC power source. The power conversion device 1 comprises a physical quantity detection unit 500a, a converter 130, a physical quantity detection unit 500b, an inverter 310, a physical quantity detection unit 500c, and a control unit 400. The estimation device 410 is included in the control unit 400 as shown in Figure 1. The power conversion device 1 and the motor 314 constitute a motor drive device 2.

[0011] The physical quantity detection unit 500a detects physical quantities that indicate the operating state of the power converter 1. For example, the physical quantity detection unit 500a detects physical quantities such as the voltage value and current value of the first AC voltage supplied from the power supply 110 to the converter 130. The physical quantity detection unit 500a outputs the detected physical quantities to the control unit 400. The physical quantity detection unit 500a may also detect zero-crossings of the first AC voltage supplied from the power supply 110 to the converter 130.

[0012] Converter 130 is a power converter that converts a first AC voltage supplied from power supply 110 into a DC voltage. Converter 130, although not shown in the figures, includes a rectifier element, a reactor, a switching element, a freewheeling diode, a capacitor, etc. The switching element is turned on and off by the control unit 400. The switching element is, for example, an IGBT (Insulated Gate Bipolar Transistor), a MOSFET (Metal Oxide Semiconductor Field Effect Transistor), a bipolar transistor, etc., but is not limited to these. The configuration of converter 130 will differ depending on whether power supply 110 is a single-phase AC power supply or a three-phase AC power supply, but a general configuration is sufficient, so a detailed explanation of the configuration and operation will be omitted.

[0013] The physical quantity detection unit 500b detects physical quantities that indicate the operating state of the power converter 1. For example, the physical quantity detection unit 500b detects physical quantities such as the voltage value and current value of the DC voltage supplied from the converter 130 to the inverter 310. The physical quantity detection unit 500b outputs the detected physical quantities to the control unit 400.

[0014] The inverter 310 is a power converter that converts the DC voltage supplied from the converter 130 into a second AC voltage. The inverter 310 includes switching elements, freewheeling diodes, etc., although these are not shown in the diagram. The switching elements are turned on and off by the control unit 400. The switching elements are, for example, IGBTs, MOSFETs, bipolar transistors, etc., but are not limited to these. The inverter 310 can be a general type, so a detailed explanation of its configuration and operation is omitted.

[0015] The physical quantity detection unit 500c detects physical quantities that indicate the operating state of the power converter 1. For example, the physical quantity detection unit 500c detects physical quantities such as the voltage value and current value of the second AC voltage supplied from the inverter 310 to the motor 314. The physical quantity detection unit 500c outputs the detected physical quantities to the control unit 400.

[0016] The power converter 1 may also include physical quantity detection units other than the physical quantity detection units 500a, 500b, and 500c. In the following description, when the physical quantity detection units of the power converter 1 are not distinguished, they may be referred to as the physical quantity detection unit 500. The physical quantity detection unit 500 may, for example, detect a physical quantity indicating the operating state of a configuration of a compressor (not shown) including a motor 314, or it may detect a physical quantity indicating the operating state of a configuration of a refrigeration cycle application device such as an air conditioner (not shown) including the power converter 1. The physical quantity detection unit 500 may also be simply referred to as the detection unit. The physical quantity detected by the physical quantity detection unit 500 may also be referred to as the first physical quantity. The first physical quantity is, for example, a current, voltage, or a physical quantity whose average value is not zero, detected by the physical quantity detection unit 500 of the power converter 1, but is not limited to these.

[0017] The control unit 400 acquires the first physical quantities detected by each physical quantity detection unit 500 from each physical quantity detection unit 500. Based on the acquired first physical quantities, the control unit 400 controls the operation of the converter 130 and the inverter 310. Specifically, based on the acquired first physical quantities, the control unit 400 controls the on / off status of switching elements (not shown) in the converter 130 and the on / off status of switching elements (not shown) in the inverter 310. The control unit 400 may also control the operation of the converter 130 and the inverter 310 based on the first physical quantities acquired from some of the physical quantity detection units 500 of the power conversion device 1. The method by which the control unit 400 controls the operation of the converter 130 and the inverter 310 can be a general method, so a detailed explanation is omitted.

[0018] As shown in Figure 1, the control unit 400 includes an estimation device 410. The estimation device 410 estimates the state of the configuration of the power converter 1 or the configuration of the device including the power converter 1 based on a first physical quantity obtained. As state estimation, the estimation device 410 performs, for example, abnormality diagnosis of the configuration of the power converter 1 or the configuration of the device including the power converter 1, and life prediction. As shown in Figure 1, the estimation device 410 includes a preprocessing unit 411, a feature extraction unit 412, and an estimation unit 413. In Embodiment 1, a pattern in which the entire configuration of the estimation device 410 is mounted on the power converter 1 is described, but as will be described in the embodiments to be described later, it is also possible that a part of the configuration of the estimation device 410 is mounted on the power converter 1, that is, another part of the configuration of the estimation device 410 exists outside the power converter 1. The power converter 1 shown in Figure 1 is a power converter 1 in which the entire estimation device 410 is mounted.

[0019] The preprocessor 411 acquires a first physical quantity from the physical quantity detection unit 500. The preprocessor 411 converts the first physical quantity, which indicates the operating state of the configuration of the power converter 1 or the configuration of the device including the power converter 1, into a second physical quantity. The preprocessor 411 outputs the second physical quantity to the feature extraction unit 412. Here, the second physical quantity is the target physical quantity input to the feature extraction unit 412 and is a physical quantity from which the feature extraction unit 412 can extract features. The device including the power converter 1 is, for example, a refrigeration cycle application device such as the aforementioned air conditioner (not shown). The configuration of the power converter 1 is, for example, the aforementioned converter 130, inverter 310, etc., but also includes DC (Direct Current) DC converters (not shown). The configuration of the device including the power converter 1 is, for example, a four-way valve (not shown) mounted on the refrigeration cycle application device such as an air conditioner including the power converter 1.

[0020] The feature extraction unit 412 obtains a second physical quantity from the preprocessing unit 411. The feature extraction unit 412 extracts features from the second physical quantity obtained from the preprocessing unit 411. The feature extraction unit 412 outputs the features to the estimation unit 413.

[0021] The estimation unit 413 obtains feature quantities from the feature quantity extraction unit 412. Based on the feature quantities, the estimation unit 413 estimates the state of the configuration of the power converter 1 or the configuration of the device including the power converter 1, i.e., performs a diagnosis or prediction. Diagnosis or prediction may include, for example, determining whether there are any abnormalities in each configuration, or predicting the lifespan of each configuration, but is not limited to these. The estimation unit 413 may display the results of the diagnosis or prediction on the display unit of a refrigeration cycle application device such as an air conditioner (not shown) on which the power converter 1 is installed, so that users of the refrigeration cycle application device such as an air conditioner can recognize the results. The estimation unit 413 may also include alerts in its output depending on the results of the diagnosis or prediction, and may change the output method of the diagnosis or prediction results as appropriate.

[0022] Here, the operation flow of the estimation device 410 will be explained using a specific example. Figure 2 is a diagram showing an example of the operation of the estimation device 410 according to Embodiment 1. In the example in Figure 2, it is assumed that the pre-trained machine learning model used by the estimation unit 413, i.e., the classifier, is a speech classifier.

[0023] Figure 2(a) shows the waveform of a current detected by a physical quantity detection unit 500 as the first physical quantity. The current waveform shown in Figure 2(a) has a sampling period of Ts1, an arbitrary amplitude range, n samples, and a total waveform length of Ts1 × n. Generally, in a feature extraction configuration such as the feature extraction unit 412 of this embodiment, there is a limit on the magnitude of the input signal, i.e., the magnitude of the physical quantity. Therefore, as shown from Figure 2(a) to Figure 2(c), the preprocessing unit 411 converts the first physical quantity into a second physical quantity, which is the target physical quantity input to the feature extraction unit 412 and from which the feature extraction unit 412 can extract features. In the example in Figure 2, the voice conversion converts the waveform of the current into a waveform of sound. In the example shown in Figure 2, the preprocessing unit 411 performs a process to convert the first physical quantity into a second physical quantity, i.e., a speech conversion, which involves converting the sampling period from Ts1 to Ts2, converting the amplitude range from an arbitrary magnitude to -1 to +1, and converting the overall waveform length from Ts1 × n to Ts2 × n. The preprocessing unit 411 also includes a process to convert the first physical quantity into a second physical quantity so that the second physical quantity can be input to the feature extraction unit 412. The first physical quantity was represented by a current waveform as shown in Figure 2(a), but the second physical quantity will be represented by a current waveform after speech conversion, as shown in Figure 2(c).

[0024] The preprocessing unit 411 includes at least one of the following processes for converting a first physical quantity into a second physical quantity: a first process of subtracting the average value of the first physical quantity from the first physical quantity; a second process of adjusting the range of the first physical quantity so that the maximum and minimum values ​​of the first physical quantity fall within the maximum and minimum values ​​of the numerical values ​​that can be input to the speech classifier; and a third process of repeatedly combining the first physical quantity at a defined period.

[0025] Figure 3 is a diagram showing an example of a first process performed by the preprocessing unit 411 of the estimation device 410 according to Embodiment 1 as a process to convert a first physical quantity into a second physical quantity. The first process is to remove the DC component so that the average value of the waveform that changes with respect to 0 becomes 0, as shown in the upper part of Figure 3, or to remove the offset so that the average value of the waveform on which the offset is superimposed becomes 0, as shown in the lower part of Figure 3. The first process is to subtract the average value from the first physical quantity, as shown in the upper and lower parts of Figure 3.

[0026] Figure 4 shows an example of a second process performed by the preprocessing unit 411 of the estimation device 410 according to Embodiment 1 as a process to convert a first physical quantity into a second physical quantity. As shown in Figure 4, the second process is a process to convert the amplitude range of the waveforms to a desired range by dividing the target waveforms by the absolute value of the waveform with the largest absolute value among the target waveforms. In the second process, the preprocessing unit 411 may exclude data that is thought to be an outlier, such as waveform 4 in the upper part of Figure 4, from the conversion process. In the example in Figure 4, the waveform with the largest absolute value among the target waveforms is set to waveform 3. Note that in the example in Figure 4, the case where the amplitude range is -1 to +1 is assumed, but it is not limited to this. The preprocessing unit 411 only needs to be able to convert the first physical quantity into a second physical quantity so that the amplitude range is from the minimum value to the maximum value of the input specifications assumed by the feature extraction unit 412.

[0027] Figure 5 shows an example of a third process performed by the preprocessing unit 411 of the estimation device 410 according to Embodiment 1 as a process to convert a first physical quantity into a second physical quantity. The third process, as shown in Figure 5, is a process of repeatedly combining waveforms of only one pulse to create an AC waveform. Although the description is simplified in Figure 5, the preprocessing unit 411 generates a waveform in which a waveform in phase with the waveform of only one pulse shown on the left side of Figure 5 and a waveform that is an inverted version of the waveform of only one pulse are repeated by repeatedly combining waveforms of only one pulse. In the third process, the preprocessing unit 411 inverts and combines the base waveform, but it may also combine waveforms in phase and remove the DC component so that the average value is 0. Furthermore, if the starting value and the ending value do not match when combining waveforms, the preprocessing unit 411 may adjust the starting value and the ending value of the waveform to match by integrating a window function. The type of window function is not particularly limited. As a result, the preprocessing unit 411 can generate an AC waveform, i.e., an audio waveform, from a waveform of only one pulse by the third process.

[0028] The preprocessing unit 411 can improve the accuracy of the processing in the feature extraction unit 412 and the estimation unit 413 by performing the processing shown in the first to third processes as a process to convert the first physical quantity into a second physical quantity.

[0029] As shown in Figure 2(d), the feature extraction unit 412 extracts features from a second physical quantity using a pre-trained model. The pre-trained model used in the feature extraction unit 412 is, for example, a pre-trained machine learning model obtained by transfer learning from an existing model such as YAMNet or vggish. Transfer learning is a well-known method of machine learning. The pre-trained model used in the feature extraction unit 412 of this embodiment is a modified version of the output layer of an existing model such as YAMNet or vggish, adapted to this embodiment so that the feature extraction unit 412 can extract features from a second physical quantity.

[0030] Figure 6 shows an example of YAMNet, a pre-trained model used in the estimation device 410 according to Embodiment 1. The YAMNet shown in Figure 6 is the same as the one shown in Figure 5 in the paper "Deep learning Binary / Multi classification for music's brainwave entertainment beats" by Rowayda A. Sadek et al., Peer J Computer Science, (November 3, 2023), with the classifier part removed. The "feature extraction as m-dimensional vector" between Figure 2(d) and Figure 2(e) specifically becomes "feature extraction as a 1024-dimensional vector" in the example of Figure 6. The feature extraction unit 412 outputs the extracted m-dimensional vector features, or in the example of Figure 6, the 1024-dimensional vector features, to the estimation unit 413. In this way, the feature extraction unit 412 can extract features even with a small amount of training data, without using a large amount of training data, by using a pre-trained machine learning model that has undergone transfer learning.

[0031] The estimation unit 413 uses a voice classifier as a classifier and performs a diagnosis or prediction about the configuration of the power converter 1 or the configuration of the device including the power converter 1, based on the features obtained from the feature extraction unit 412.

[0032] Here, the estimation unit 413 can update the contents of the classifier used for diagnosis or prediction, i.e., the contents of the learning process, using the features obtained from the feature extraction unit 412. However, the number of training data samples, which are the features used in the learning process of the estimation unit 413, is less than the number of samples used for training by the pre-trained machine learning model used by the feature extraction unit 412. By updating the contents of the learning process, the estimation unit 413 can perform optimal estimation tailored to the environment in which the power converter 1 is installed, the operating conditions of the power converter 1, and so on. For example, if the power converter 1 is used in refrigeration cycle equipment such as an air conditioner, the outside temperature of such refrigeration cycle equipment may differ depending on the region in which it is used, and the features may change depending on the season. Therefore, by updating the contents of the learning process according to the region in which it is used, the estimation unit 413 can improve the accuracy of diagnosis or prediction, enabling diagnosis or prediction tailored to the region in which it is used. The learning of the classifier used in the estimation unit 413 may be performed before or after shipment of the product including the estimation device 410.

[0033] Although the explanation has been given assuming that the pre-trained machine learning model used by the estimation unit 413, i.e., the classifier, is a speech classifier, it is not limited to this. The pre-trained machine learning model used by the estimation unit 413, i.e., the classifier, may be an image classifier. In this case, the pre-processing unit 411 converts the first physical quantity into a second physical quantity by image transformation. In this case as well, the pre-processing unit 411 includes at least one of the following processes for converting the first physical quantity into a second physical quantity: a first process of subtracting the average value of the first physical quantity from the first physical quantity; a second process of adjusting the range of the first physical quantity so that the maximum and minimum values ​​of the first physical quantity fall within the maximum and minimum values ​​of the numerical values ​​that can be input to the image classifier; and a third process of repeatedly combining the first physical quantity at a defined period. The same applies to subsequent embodiments.

[0034] Figure 7 is a flowchart illustrating the operation of the estimation device 410 according to Embodiment 1. In the estimation device 410, the preprocessing unit 411 converts a first physical quantity, which is a physical quantity detected by the physical quantity detection unit 500 and indicates the operating state of the configuration of the power converter 1 or the configuration of the device including the power converter 1, into a second physical quantity (step S1). The feature quantity extraction unit 412 extracts feature quantities from the second physical quantity (step S2). The estimation unit 413 estimates the state of the configuration of the power converter 1 or the configuration of the device including the power converter 1 based on the feature quantities (step S3).

[0035] Next, the hardware configuration of the estimation device 410 provided in the control unit 400 of the power converter 1 will be described. Figure 8 is a diagram showing an example of the hardware configuration for realizing the estimation device 410 provided in the control unit 400 of the power converter 1 according to Embodiment 1. In the estimation device 410, the preprocessing unit 411, the feature extraction unit 412, and the estimation unit 413 are realized by the processor 91 and the memory 92.

[0036] The processor 91 is a CPU (Central Processing Unit, also known as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, processor, or DSP (Digital Signal Processor)), or a system LSI (Large Scale Integration). Memory 92 can be exemplified by non-volatile or volatile semiconductor memories such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), and EEPROM® (Electrically Erasable Programmable Read Only Memory). Memory 92 is not limited to these and may also be a magnetic disk, optical disk, compact disk, minidisc, or DVD (Digital Versatile Disc).

[0037] As described above, according to this embodiment, in the estimation device 410, the preprocessing unit 411 converts a first physical quantity indicating the operating state of the configuration of the power converter 1 or the configuration of the device including the power converter 1 into a second physical quantity. The feature extraction unit 412 extracts features from the second physical quantity using a pre-trained model. The estimation unit 413 uses a speech classifier as a classifier and performs a diagnosis or prediction about the configuration of the power converter 1 or the configuration of the device including the power converter 1 based on the features obtained from the feature extraction unit 412. As a result, the estimation device 410 can make highly accurate estimations of the state of the configuration of the power converter 1 or the configuration of the device including the power converter 1 from power electronics data such as current and voltage. Furthermore, by using a pre-trained model, the estimation device 410 can make highly accurate estimations even with a small amount of training data, without using a large amount of training data. The estimation device 410 can improve the accuracy of state estimation for devices that perform motor control.

[0038] In this embodiment, the feature extraction unit 412 extracted features using a pre-trained model, and the estimation unit 413 estimated the configuration of the power converter 1 or the configuration of the device including the power converter 1 using a classifier, but this is not limited to this. The feature extraction unit 412 may extract features using statistics if possible, or may extract features using statistics in combination with a pre-trained model. Statistics are, for example, the effective value, mean value, and maximum value of a second physical quantity, but are not limited to these. The estimation unit 413 may perform estimation using a machine learning model, or it may perform estimation by simply comparing with a threshold, or it may use these methods in combination with a classifier.

[0039] Furthermore, although the case in which the estimation device 410 is composed of a preprocessing unit 411, a feature extraction unit 412, and an estimation unit 413 has been described, it is not limited to this. The estimation device 410 may be configured such that the preprocessing unit 411 and the feature extraction unit 412 perform processing as a single unit, or the feature extraction unit 412 and the estimation unit 413 perform processing as a single unit.

[0040] Embodiment 2. Embodiment 2 describes the case where the estimation unit 413 of the estimation device 410 is located outside the power converter 1.

[0041] Figure 9 shows an example configuration of a power converter 1 including a part of the estimation device 410 according to Embodiment 2. In Embodiment 2, the power converter 1 differs from the power converter 1 of Embodiment 1 shown in Figure 1 in that the estimation unit 413 of the estimation device 410 is located outside, and a communication unit 414 is added. That is, the estimation device 410 of Embodiment 2 further includes a communication unit 414 compared to the estimation device 410 of Embodiment 1. The communication unit 414 transmits the feature quantities extracted by the feature quantity extraction unit 412 to the estimation unit 413 located outside the power converter 1. The communication unit 414 may communicate via the Internet or via a dedicated network. Based on the feature quantities acquired via the communication unit 414, the estimation unit 413 performs a diagnosis or prediction about the configuration of the power converter 1 or the configuration of the device including the power converter 1. The power converter 1 shown in Figure 9 is a power converter 1 on which a part of the estimation device 410 is mounted. As in the case of Figure 1, the motor drive unit 2 is composed of the power converter 1 and the motor 314.

[0042] Figure 10 shows an example of the operation of the estimation device 410 according to Embodiment 2. In the example in Figure 10, it is assumed that the pre-trained machine learning model used by the estimation unit 413, i.e., the classifier, is a speech classifier. As mentioned above, the pre-trained machine learning model used by the estimation unit 413, i.e., the classifier, may be an image classifier. In Embodiment 2, the operation of the pre-processing unit 411, the feature extraction unit 412, and the estimation unit 413 of the estimation device 410 is the same as in Embodiment 1. However, in Embodiment 2, it is assumed that the pre-processing unit 411 and the feature extraction unit 412 are located inside the power converter 1, i.e., on the edge side, and the estimation unit 413 is located outside the power converter 1, i.e., on the cloud side. In Figure 10, for comparison with Figure 2, the description is omitted, but in reality, the features extracted by the feature extraction unit 412 are transmitted to the estimation unit 413 by the communication unit 414.

[0043] As shown in Figures 9 and 10, operators using multiple power converters 1 can share the estimation unit 413 of the estimation device 410. The estimation unit 413 acquires feature quantities from the feature extraction units 412 of the multiple estimation devices 410 via the communication unit 414 of each estimation device 410, and performs a diagnosis or prediction about the configuration of each power converter 1 or the configuration of the device including each power converter 1. At this time, the estimation unit 413 may update the contents of the classifier used when performing the diagnosis or prediction, i.e., the contents of the learning process, using the feature quantities acquired from the multiple feature extraction units 412. As a result, the estimation unit 413 can use many feature quantities from multiple power converters 1 compared to the case where only the feature quantities for one power converter 1 are used, thereby improving the accuracy of the estimation. In the example of Figure 10, it is assumed that the estimation unit 413 is installed in the cloud, but the installation pattern of the estimation unit 413 is not limited to this. The estimation unit 413 may be installed on a specific server or the like.

[0044] In the above example, the estimation device 410 has been described in the case where the preprocessing unit 411 and the feature quantity extraction unit 412 are provided inside the power conversion device 1, that is, on the edge side, and the estimation unit 413 is provided outside the power conversion device 1, that is, on the cloud side. However, the present invention is not limited to this. The estimation device 410 may be configured such that the preprocessing unit 411 is provided inside the power conversion device 1, that is, on the edge side, and the feature quantity extraction unit 412 and the estimation unit 413 are provided outside the power conversion device 1, that is, on the cloud side. In this case, the communication unit 414 acquires the second physical quantity from the preprocessing unit 411 and transmits it to the feature quantity extraction unit 412 on the cloud side. Further, the estimation device 410 may be configured such that the preprocessing unit 411, the feature quantity extraction unit 412, and the estimation unit 413 are provided outside the power conversion device 1, that is, on the cloud side. In this case, the communication unit 414 acquires the first physical quantity from each physical quantity detection unit 500 and transmits it to the preprocessing unit 411 on the cloud side. However, in the two patterns described here, it is assumed that the communication volume of the communication unit 414 is larger than that in the case of transmitting the feature quantity. Therefore, as the configuration of the estimation device 410, a pattern as shown in FIGS. 9 and 10 in which information is compressed and the communication volume is reduced is more preferable.

[0045] As described above, according to the present embodiment, the estimation device 410 adds the communication unit 414 as compared with the configuration of the estimation device 410 in Embodiment 1, and the communication unit 414 transmits the feature quantity extracted by the feature quantity extraction unit 到 the estimation unit 413 on the cloud side. As a result, the estimation device 410 enables the estimation unit 413 to acquire feature quantities from a plurality of power conversion devices 1, so that a more accurate estimation is possible as compared with Embodiment 1.

[0046] Embodiment 3. In Embodiment 3, a specific diagnosis example performed by the estimation unit 413 of the estimation device 410 will be described. Note that the case where the configuration of the estimation device 410 is the configuration shown in Embodiment 1 will be described as an example, but Embodiment 3 is also applicable to the case where the configuration of the estimation device 410 is the configuration shown in Embodiment 2.

[0047] FIG. 11 is a diagram showing a diagnostic example by the estimation device 410 according to Embodiment 3. FIG. 11 shows an example in which the estimation device 410 performs oil level diagnosis using a common current. In FIG. 11, only the configurations necessary for the description are shown, and the parts unnecessary for the description are omitted. The compressor 315 shown in FIG. 11 is assumed to include the motor 314 shown in FIG. 1 although not shown.

[0048] Generally, a common current may flow from the compressor 315 to the reference plane through the capacitance to the ground. The compressor 315 holds oil for lubricating the movement of internal parts, and the impedance of the common current changes depending on the height of the oil level inside the compressor 315. The power conversion device 1 can detect the shape change of the current waveform of the common current using the physical quantity detection unit 500. However, since the amount of change in impedance is small, it is difficult to perform diagnosis by a method using the peak value, effective value, etc. of the common current. Therefore, in the present embodiment, the estimation device 410 acquires the information of the current waveform of the common current as shown in FIG. 11 as the first physical quantity, converts the first physical quantity into the second physical quantity, and then extracts the feature quantity from the second physical quantity. The estimation device 410 diagnoses the height of the oil level inside the compressor 315 from the extracted feature quantity, diagnoses it as normal when the oil level height is higher than the threshold value, and diagnoses it as abnormal when the oil level height is lower than the threshold value.

[0049] Note that the specific diagnosis method by the estimation device 410 is not limited to the above example. The estimation device 410 may perform a rare short diagnosis using, for example, the phase current of the motor 314 detected as the first physical quantity, or may perform a solder deterioration diagnosis using the drain-source voltage of the IPM (Intelligent Power Module) detected as the first physical quantity. As long as the feature quantity is included in the current waveform or voltage waveform, the estimation device 410 can perform various diagnoses using the first physical quantity represented by the current waveform or voltage waveform. The diagnosis method by the estimation device 410 may be a multi-value classification instead of the binary classification method as shown in FIG. 11.

[0050] As described above, according to this embodiment, the estimation device 410 can perform a specific diagnosis because the current waveform or voltage waveform, which is represented by the first physical quantity detected by the physical quantity detection unit 500, contains a characteristic quantity.

[0051] Embodiment 4. Embodiment 4 describes a case in which the estimation device 410 diagnoses an air conditioner as the aforementioned refrigeration cycle application equipment. Although the description will use the configuration of the estimation device 410 shown in Embodiment 1 as an example, Embodiment 4 is also applicable when the configuration of the estimation device 410 is shown in Embodiment 2.

[0052] Figure 12 is a first diagram showing an example in which the estimation device 410 according to Embodiment 4 is applied to an air conditioner 700. The air conditioner 700 comprises an outdoor unit 710 and an indoor unit 720. The outdoor unit 710 comprises a physical quantity detection unit 500d, a noise filter 120a, a converter 130a, a DC-DC converter 320a, inverters 310a and 310b, a fan 316a, a compressor 315, a control unit 400, and a four-way valve 902. The indoor unit 720 comprises a noise filter 120b, a converter 130b, a DC-DC converter 320b, an inverter 310c, and a fan 316b. Although the method of description in Figure 12 differs from that in Figure 1, Figure 12 shows that the estimation device 410 is located inside the control unit 400, similar to Figure 1.

[0053] In the outdoor unit 710, the physical quantity detection unit 500d detects the current value, voltage value, etc., of the first AC voltage supplied from the power supply 110 as the first physical quantity and outputs it to the control unit 400. In the example of Figure 12, only one physical quantity detection unit 500d is installed inside the air conditioner 700, but multiple physical quantity detection units 500 may be installed inside the air conditioner 700. The noise filter 120a removes noise from the first AC voltage supplied from the power supply 110 to the converter 130a and the four-way valve 902. The converter 130a converts the first AC voltage to a DC voltage. The DC-DC converter 320a converts the voltage value of the DC voltage obtained from the converter 130a. The inverter 310a converts the DC voltage obtained from the converter 130a to a second AC voltage. The inverter 310b converts the DC voltage obtained from the converter 130a to a second AC voltage. The second AC voltages generated by inverters 310a and 310b may be the same or different. Fan 316a is driven by the second AC voltage obtained from inverter 310a. Compressor 315 is driven by the second AC voltage obtained from inverter 310b. Four-way valve 902 is driven by the first AC voltage.

[0054] In the indoor unit 720, the noise filter 120b removes noise from the first AC voltage supplied from the power supply 110 to the converter 130b. The converter 130b converts the first AC voltage into a DC voltage. The DC voltages generated by converters 130a and 130b may be the same or different. The DC-DC converter 320b converts the voltage value of the DC voltage obtained from converter 130b. The inverter 310c converts the DC voltage obtained from converter 130b into a second AC voltage. The fan 316b is driven by the second AC voltage obtained from inverter 310c.

[0055] The estimation device 410 uses the first physical quantity detected by the physical quantity detection unit 500d to diagnose the state of the various drive units of the air conditioner 700. The internal processing of the estimation device 410 shown in Figure 12 is the same as the processing described in Embodiment 1, so a detailed explanation is omitted. In Figure 12, the estimation unit 413 outputs information indicating the probability of abnormality for each component as shown in the upper right of Figure 12. That is, the estimation unit 413 outputs the probability of abnormality of the compressor 315, the probability of abnormality of the fan 316a, the probability of abnormality of the fan 316b, and the probability of abnormality of the four-way valve 902 based on the feature quantities. The estimation unit 413 may perform the diagnosis using binary classification, which indicates either abnormal or normal as the diagnosis result, rather than multi-class classification such as abnormality probability which can take various values.

[0056] The air conditioner 700 consists of an outdoor unit 710 and an indoor unit 720, and a first AC voltage is supplied in parallel from the power supply 110 to the outdoor unit 710 and the indoor unit 720. Therefore, the estimation device 410 may use a first physical quantity obtained from the outdoor unit 710, a first physical quantity obtained from the indoor unit 720, or a first physical quantity obtained from both the outdoor unit 710 and the indoor unit 720, as long as the first physical quantity includes a feature quantity for the object to be diagnosed. Furthermore, if the first physical quantity at the input terminal or output terminal of the converter 130a, 130b, inverter 310a, 310b, 310c, etc., which are the object to be diagnosed, includes a feature quantity, the air conditioner 700 may be equipped with multiple physical quantity detection units 500 at positions where the first physical quantity containing the feature quantity can be detected. In this case, the estimation device 410 can obtain the first physical quantity from the multiple physical quantity detection units 500 and perform the diagnosis.

[0057] In the example shown in Figure 12, the estimation unit 413 of the estimation device 410 diagnosed abnormalities in multiple drive units, but it is not limited to this. The estimation device 410 includes an estimation unit 413 corresponding to the drive unit to be diagnosed, and each estimation unit 413 may perform diagnosis targeting only the drive unit to be diagnosed. Figure 13 is a second diagram showing an example in which the estimation device 410 according to Embodiment 4 is applied to an air conditioner 700. In Figure 13, the estimation unit 413 is replaced with estimation units 413a to 413d compared to Figure 12. In the example shown in Figure 13, estimation unit 413a outputs the probability of abnormality of the compressor 315 based on feature quantities. Estimation unit 413b outputs the probability of abnormality of the fan 316a based on feature quantities. Estimation unit 413c outputs the probability of abnormality of the fan 316b based on feature quantities. Estimation unit 413d outputs the probability of abnormality of the four-way valve 902 based on feature quantities. Each of the estimation units 413a to 413d may perform a diagnosis using a binary classification that indicates either abnormal or normal, rather than a multi-level classification such as an abnormality probability that can take various values, as a method for diagnosing the target drive unit.

[0058] In the examples in Figures 12 and 13, the case in which the air conditioner 700 comprises one outdoor unit 710 and one indoor unit 720 was described, but the air conditioner 700 can also comprise one outdoor unit 710 and multiple indoor units. Figure 14 is a third figure showing an example in which the estimation device 410 according to Embodiment 4 is applied to the air conditioner 700. The air conditioner 700 shown in Figure 14 is the air conditioner 700 shown in Figure 12 with the addition of indoor units 720a and 720b. Indoor unit 720a comprises a noise filter 120c, a converter 130c, a DC-DC converter 320c, an inverter 310d, and a fan 316c. Indoor unit 720b comprises a noise filter 120d, a converter 130d, a DC-DC converter 320d, an inverter 310e, and a fan 316d. Since indoor units 720a and 720b have the same configuration as indoor unit 720, a detailed explanation will be omitted.

[0059] Even with the configuration shown in Figure 14, the estimation device 410 can diagnose one outdoor unit 710 and multiple indoor units 720, 720a, and 720b, provided that the acquired first physical quantity includes a characteristic quantity for the unit to be diagnosed. Furthermore, if the current flowing through each indoor unit includes a characteristic quantity for the unit to be diagnosed, the air conditioner 700 may be provided with a physical quantity detection unit 500 capable of detecting current at the input terminal from the power supply 110 of each indoor unit, and the estimation device 410 may acquire the first physical quantity from the physical quantity detection unit 500 provided in each indoor unit. This allows the estimation device 410 to identify which indoor unit has malfunctioned. Although the configuration of the estimation device 410 shown in Figure 12 was used as an example when the air conditioner 700 has multiple indoor units, the device is not limited to this. The configuration of the estimation device 410 shown in Figure 13 is also applicable when the air conditioner 700 has multiple indoor units. Furthermore, the number of indoor units in the air conditioner 700 is not limited to the three cases shown in Figure 14. The number of indoor units in the air conditioner 700 may be two, or four or more.

[0060] As described above, according to this embodiment, the estimation device 410 can perform specific diagnoses for multiple diagnostic targets because the first physical quantity detected by one or more physical quantity detection units 500 includes characteristic quantities for multiple diagnostic targets.

[0061] Embodiment 5. Embodiment 5 describes a case in which the estimation device 410 performs fault diagnosis of a sensor used in an air conditioner 700 or the like.

[0062] Figure 15 shows an example of the estimation device 410 according to Embodiment 5 performing fault diagnosis of the sensor 600 of the air conditioner 700. The air conditioner 700 shown in Figure 15 is the same as the air conditioner 700 shown in Figure 12, but with the physical quantity detection unit 500d replaced by a physical quantity detection unit 500e, and a sensor 600 added. The physical quantity detection unit 500e acquires, for example, the current value supplied to the compressor 315 as the first physical quantity. In the example of Figure 15, the sensor 600 detects the temperature of the compressor 315. The sensor 600 may also be in the form of the physical quantity detection unit 500.

[0063] The estimation device 410 estimates the temperature of the compressor 315 based on a first physical quantity obtained from the physical quantity detection unit 500e, namely the current value supplied to the compressor 315. The estimation device 410 also obtains temperature information of the compressor 315 from the sensor 600. Using this, the estimation device 410 can obtain a relationship between the feature quantities shown in the upper right of Figure 15 and the estimated temperature of the compressor 315, using past estimation results and past temperature data obtained from the sensor 600. By using the relationship between the feature quantities shown in the upper right of Figure 15 and the estimated temperature of the compressor 315, the estimation device 410 can estimate the temperature of the compressor 315 when a certain feature quantity is present. Therefore, if a large discrepancy occurs between the estimated temperature and the temperature obtained from the sensor 600, the estimation device 410 can estimate that the sensor 600 is malfunctioning, i.e., perform regression analysis.

[0064] The method for diagnosing sensor failures using the estimation device 410 is not limited to the above example. The estimation device 410 can diagnose sensor failures based on the feature quantity obtained from the first physical quantity detected by the physical quantity detection unit 500 if there is a correlation between the feature quantity obtained from the first physical quantity detected by the physical quantity detection unit 500 and the output from a sensor that detects the operating state of a certain object to be diagnosed. Furthermore, if such a correlation is obtained, the type of sensor is not limited to a temperature-detecting sensor. For example, if the power converter 1 is equipped with a sensor that detects the rotation speed of a fan, the estimation device 410 can estimate whether or not the sensor that detects the rotation speed of the fan is faulty if a feature quantity about the rotation speed of the fan is obtained from the first physical quantity. Although the example of the estimation device 410 performing fault diagnosis of sensor 600 was given using the configuration of the estimation device 410 shown in Figure 12, it is not limited to this. The estimation device 410 can also be applied to the case where the configuration of the estimation device 410 is shown in Figure 13 when the estimation device 410 performs fault diagnosis of sensor 600.

[0065] As described above, according to this embodiment, if there is a correlation between the feature quantity obtained from the first physical quantity detected by the physical quantity detection unit 500 and the output from the sensor that detects the operating state of the object to be diagnosed, the estimation device 410 can perform a fault diagnosis of the sensor based on the feature quantity obtained from the first physical quantity detected by the physical quantity detection unit 500.

[0066] Embodiment 6. Figure 16 shows an example configuration of a refrigeration cycle application device 900 according to Embodiment 6. The refrigeration cycle application device 900 according to Embodiment 6 includes the power converter 1 described in Embodiment 1. The refrigeration cycle application device 900 according to Embodiment 6 may also include the power converter 1 described in Embodiments 2 to 5. The refrigeration cycle application device 900 according to Embodiment 6 can be applied to products equipped with a refrigeration cycle, such as air conditioners, refrigerators, freezers, and heat pump water heaters. In Figure 16, components having the same functions as those described in Embodiments 1 to 5 are denoted by the same reference numerals as in Embodiments 1 to 5.

[0067] The refrigeration cycle application equipment 900 includes a compressor 315 with a built-in motor 314 as in Embodiment 1, a four-way valve 902, an indoor heat exchanger 906, an expansion valve 908, and an outdoor heat exchanger 910, all of which are connected via refrigerant piping 912.

[0068] Inside the compressor 315 are a compression mechanism 904 for compressing the refrigerant and a motor 314 for operating the compression mechanism 904.

[0069] The refrigeration cycle equipment 900 can operate in heating or cooling mode by switching the four-way valve 902. The compression mechanism 904 is driven by a variable-speed controlled motor 314.

[0070] During heating operation, as indicated by the solid arrows, the refrigerant is pressurized by the compression mechanism 904 and sent out, then returns to the compression mechanism 904 after passing through the four-way valve 902, indoor heat exchanger 906, expansion valve 908, outdoor heat exchanger 910 and the four-way valve 902.

[0071] During cooling operation, as indicated by the dashed arrows, the refrigerant is pressurized by the compression mechanism 904 and sent out, then returns to the compression mechanism 904 after passing through the four-way valve 902, the outdoor heat exchanger 910, the expansion valve 908, the indoor heat exchanger 906 and the four-way valve 902.

[0072] During heating operation, the indoor heat exchanger 906 acts as a condenser to release heat, and the outdoor heat exchanger 910 acts as an evaporator to absorb heat. During cooling operation, the outdoor heat exchanger 910 acts as a condenser to release heat, and the indoor heat exchanger 906 acts as an evaporator to absorb heat. The expansion valve 908 reduces the pressure of the refrigerant and causes it to expand.

[0073] The configurations shown in the above embodiments are merely examples, and it is possible to combine them with other known technologies, combine different embodiments, and omit or modify parts of the configuration without departing from the gist of the invention.

[0074] 1 Power converter, 2 Motor drive unit, 91 Processor, 92 Memory, 110 Power supply, 120a, 120b, 120c, 120d Noise filter, 130, 130a, 130b, 130c, 130d Converter, 310, 310a, 310b, 310c, 310d, 310e Inverter, 320a, 320b, 320c, 320d DCDC converter, 314 Motor, 315 Compressor, 316a, 316b, 316c, 316d Fan, 400 Control unit, 410 Estimation unit, 411 Preprocessing unit, 412 Feature extraction unit, 413, 413a, 413b, 413c, 413d Estimation unit, 414 Communication unit, 500a, 500b, 500c, 500d, 500e; Physical quantity detection unit, 600; Sensor, 700; Air conditioner, 710; Outdoor unit, 720, 720a, 720b; Indoor unit, 900; Refrigeration cycle application equipment, 902; Four-way valve, 904; Compression mechanism, 906; Indoor heat exchanger, 908; Expansion valve, 910; Outdoor heat exchanger, 912; Refrigerant piping.

Claims

1. An estimation device, which is mounted in whole or in part on a power converter, comprising: a preprocessing unit that converts a first physical quantity indicating the operating state of the configuration of the power converter or the configuration of a device including the power converter into a second physical quantity; a feature extraction unit that extracts feature quantities from the second physical quantity; and an estimation unit that performs diagnosis or prediction about the configuration based on the feature quantities, wherein the second physical quantity is a target physical quantity input to the feature extraction unit and is a physical quantity from which the feature extraction unit can extract the feature quantities.

2. The estimation apparatus according to claim 1, wherein the preprocessing unit includes a process for converting the first physical quantity to the second physical quantity such that the second physical quantity can be input to the feature extraction unit, and the feature extraction unit extracts the features using a pre-trained machine learning model.

3. The estimation device according to claim 1 or 2, wherein the estimation unit updates the content of the learning process used when performing the diagnosis or prediction using the features, and the number of samples of the learning data which are the features used in the learning process of the estimation unit is less than the number of samples used for learning by the pre-trained machine learning model used by the feature extraction unit.

4. The estimation device according to any one of claims 1 to 3, wherein the pre-trained machine learning model used by the feature extraction unit is a speech classifier, and the pre-processing unit includes at least one of the following processes for converting the first physical quantity to the second physical quantity: a first process of subtracting the average value of the first physical quantity from the first physical quantity; a second process of adjusting the range of the first physical quantity so that the maximum and minimum values ​​of the first physical quantity fall within the maximum and minimum values ​​of numerical values ​​that can be input to the speech classifier; and a third process of repeatedly combining the first physical quantity at a defined period.

5. The estimation device according to any one of claims 1 to 3, wherein the pre-trained machine learning model used by the feature extraction unit is an image classifier, and the pre-processing unit includes at least one of the following processes for converting the first physical quantity to the second physical quantity: a first process of subtracting the average value of the first physical quantity from the first physical quantity; a second process of adjusting the range of the first physical quantity so that the maximum and minimum values ​​of the first physical quantity fall within the maximum and minimum values ​​of numerical values ​​that can be input to the image classifier; and a third process of repeatedly combining the first physical quantity at a defined period.

6. An estimation device according to any one of claims 1 to 5, comprising: a communication unit that transmits the feature quantities extracted by the feature quantity extraction unit to the estimation unit located outside the power converter, wherein the estimation unit performs a diagnosis or prediction of the configuration based on the feature quantities obtained via the communication unit.

7. The estimation device according to any one of claims 1 to 6, wherein the first physical quantity is a physical quantity in which the current, voltage, or average value detected by a detection unit provided in the power converter is not zero.

8. A power conversion device equipped with all or part of the estimation device described in any one of claims 1 to 7.

9. A motor drive device comprising the power conversion device described in claim 8.

10. A refrigeration cycle application device comprising the power conversion device described in claim 8.

Images