Driving state identification device and power conversion device

The driving state identification device uses coordinate system mapping and machine learning to identify abnormal states in power conversion systems, enhancing control precision and system balance.

JP7591932B2Active Publication Date: 2024-11-29TMEIC CORP (100 00)
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2021010634
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-01-26
Publication Date
2024-11-29
Estimated Expiration
2041-01-26

AI Technical Summary

Technical Problem

Existing drive state identification devices struggle to accurately identify operating states of multiple switching elements that deviate from the normal state in power conversion systems.

Method used

A driving state identification device that maps current command values and detected values in a stationary coordinate system, analyzing the difference between their trajectories to identify abnormal operating states using determination criteria and machine learning techniques.

Benefits of technology

Effectively detects abnormal operating states of switching elements, ensuring precise control and maintaining balance in power conversion systems.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007591932000001
    Figure 0007591932000001
  • Figure 0007591932000002
    Figure 0007591932000002
  • Figure 0007591932000003
    Figure 0007591932000003
Patent Text Reader

Abstract

To provide a drive state identification device capable of identifying that an operation state different from a normal state occurs in an operation state of a plurality of switching elements and a power conversion device.SOLUTION: A drive state identification device includes a coordinate conversion unit, a mapping processing unit, and a state identification unit. The coordinate conversion unit performs coordinate conversion of data of a detected value of a current flowing in a winding of a three-phase AC motor from a three-phase stationary coordinate system into a two-phase stationary coordinate system. The mapping processing unit maps results of the coordinate conversion to a two-phase stationary coordinate plane. The state identification unit identifies a drive state of the AC motor using a determination reference and mapped results.SELECTED DRAWING: Figure 13
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] An embodiment of the present invention relates to a driving state identification device and a power conversion device. [Background technology]

[0002] A power conversion device generates polyphase AC power by switching a plurality of switching elements included in the power conversion device. A drive state identification device identifies the drive state of an electric motor driven by the power conversion device. Such a drive state identification device is sometimes required to identify that an operating state different from a normal operating state occurs in the plurality of switching elements. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-158858 A Summary of the Invention [Problem to be solved by the invention]

[0004] An object of the present invention is to provide a drive state identification device and a power conversion device capable of identifying that an operating state different from normal operating states has occurred in the operating states of a plurality of switching elements. [Means for solving the problem]

[0005] The driving state identification device according to the embodiment , state State Identification Department Prepare. Status Identification Unit teeth, Controlling the switching elements of the power converter For 3-phase AC A current command value for adjusting the current flowing through the windings of the motor; The detected value of the current flowing through the motor windings Shown in stationary coordinate system data , stillness coordinate plane To Mapped Based on the difference between the trajectory of the current command value and the trajectory of the current detection value The above The switching element is in a state different from normal operation. Identify. [Brief description of the drawings]

[0006] [Figure 1A] 1 is a configuration diagram showing an electric motor drive system including a power conversion device according to an embodiment; [Figure 1B] FIG. 2 is a configuration diagram of an inverter in the power conversion device according to the embodiment. [Figure 1C] FIG. 2 is a configuration diagram of a PWM controller according to an embodiment. [Figure 1D] FIG. 4 is a configuration diagram of an analysis processing unit 62 according to the embodiment. [Figure 2A] 5A and 5B are diagrams for explaining the relationship between three-phase signals and two-phase signals in a normal state according to the embodiment. [Figure 2B] 5A and 5B are diagrams for explaining the relationship between two-phase signals in a normal state and corresponding command values ​​in the embodiment. [Figure 2C] 6A and 6B are diagrams for explaining images of detection results in a normal state according to the embodiment; [Figure 3A] 13 is a diagram for explaining a current waveform and its corresponding trajectory when a specific gate pulse is intentionally distorted. [Figure 3B] 13 is a diagram for explaining a current waveform and its corresponding trajectory when a specific gate pulse is intentionally distorted. [Figure 3C] 13 is a diagram for explaining a current waveform and its corresponding trajectory when a specific gate pulse is intentionally distorted. [Figure 4] 4 is a flowchart relating to a determination process according to the embodiment. [Diagram 5] 11 is a flowchart relating to a process of generating a determination criterion according to the embodiment. [Figure 6] 5A to 5C are diagrams for explaining a first determination criterion and a method for generating the first determination criterion according to the embodiment. [Figure 7] 3 is a diagram for explaining the phase relationship of currents (vectors) of each phase of the electric motor 3 according to the embodiment. FIG. [Figure 8] 13 is a diagram for explaining a circuit module that executes a rotation process according to a modified example. [Figure 9A] 13A and 13B are diagrams for explaining the results of a rotation process on a trajectory in a modified example. [Figure 9B] 13A and 13B are diagrams for explaining the results of a rotation process on a trajectory in a modified example. [Figure 10]FIG. 11 is a diagram for explaining a determination process according to the second embodiment. [Figure 11] 13A to 13C are diagrams for explaining the use of rotational symmetry of a trajectory in a modified example of the second embodiment. [Figure 12] FIG. 13 is a diagram for explaining a determination process according to a first modified example of the second embodiment. [Figure 13] FIG. 13 is a diagram for explaining a determination process according to a modified example of the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] Hereinafter, a driving state identification device and a power conversion device according to an embodiment will be described with reference to the drawings. In the following description, components having the same or similar functions are denoted by the same reference numerals.

[0008] In the specification, the term "connection" is not limited to physical connection, but also includes electrical connection.

[0009] (First embodiment) Fig. 1A is a configuration diagram showing an electric motor drive system 1 including a power conversion device 2 of an embodiment. Fig. 1A shows the electric motor drive system 1, the power conversion device 2, an electric motor 3, a mechanical load 4, a terminal device 5, and an AC power source PS.

[0010] The AC power source PS is a commercial power system or a generator, and supplies three-phase AC power to the power conversion device 2, for example.

[0011] The electric motor drive system 1 includes, for example, a power conversion device 2 and an electric motor 3.

[0012] The motor 3 is, for example, a three-phase AC motor (M) capable of variable speed control such as an induction motor. When three-phase AC power is supplied to its windings, the motor 3 generates a rotational driving force and outputs it to an output shaft, and the rotational driving force drives a mechanical load 4 connected to the output shaft. The motor 3 may be equipped with a rotational speed sensor 3A that detects the rotational speed of the shaft of the motor 3. The rotational speed sensor 3A outputs, for example, the detected rotational speed ωr of the shaft of the motor 3.

[0013] The power conversion device 2 generates three-phase AC power, and supplies the generated three-phase AC power to the windings of the electric motor 3.

[0014] For example, the power conversion device 2 includes a converter 20 (rectifier), a capacitor 30, a power inverter 50 (inverter), a control unit 60, a current detector 70, and a communication interface unit 80.

[0015] An AC power source PS is connected to the AC side of the forward converter 20, and a capacitor 30 and an inverter 50 are connected via a DC link to the DC side of the forward converter 20. The forward converter 20 converts AC power supplied from the AC power source PS into DC power, and the capacitor 30 smoothes the voltage of the DC power.

[0016] The inverter 50 is a three-phase inverter including a plurality of switching elements 50S such as IGBTs. The type of the switching elements 50S is not limited to IGBTs and may be other types. For example, the switching elements 50S of the inverter 50 are PWM (Pulse Width Modulation) controlled by the control unit 60. The inverter 50 converts DC power supplied from the forward converter 20 into AC power. The inverter 50 supplies the converted three-phase AC power to the windings of the motor 3 connected to the output of the inverter 50 to drive the motor. The phases of the three-phase AC power output by the inverter 50 are called U-phase, V-phase and W-phase.

[0017] FIG. 1B is a configuration diagram of an inverter 50 in a power converter 2 of an embodiment. As shown in FIG. 1B, the inverter 50 includes upper arms 50U, 50V, and 50W, and lower arms 50X, 50Y, and 50Z. The upper arm 50U and the lower arm 50X form a U-phase leg LU. The upper arm 50V and the lower arm 50Y form a V-phase leg LV. The upper arm 50W and the lower arm 50Z form a W-phase leg LW. Each of the upper arms and each of the lower arms includes one or more switching elements 50S. In the following description, a case where each of the upper arms includes one switching element 50S will be illustrated.

[0018] Returning to FIG. 1A, the current detector 70 is provided, for example, in the V phase and W phase of the load power line connecting the output of the inverter 50 and the electric motor 3, and detects the load currents Ivs and Iws flowing through the windings of the electric motor 3.

[0019] The communication interface unit 80 communicates with the terminal device 5 under the control of the control unit 60, for example.

[0020] The control unit 60 includes, for example, a rotational speed reference generating unit 61, an analysis processing unit 62, a detected speed processing unit 63, a speed control unit 64, a detected current processing unit 65, a coordinate conversion unit 66, a current control unit 67, an inverse coordinate conversion unit 68, and a PWM controller 69. The control unit 60 is an example of a driving state identification device.

[0021] The rotational speed reference generating unit 61, the detected speed processing unit 63, the speed control unit 64, the detected current processing unit 65, the coordinate conversion unit 66, the current control unit 67, the inverse coordinate conversion unit 68, and the PWM controller 69 work in cooperation with each other to control the inverter 50. The analysis processing unit 62 analyzes the operating state of the inverter 50. The control unit 60 may adjust the control of each of the above units based on the analysis result of the analysis processing unit 62 of the operating state of the inverter 50. The adjustment of the control of each of the above units includes, for example, stopping the operation of the electric motor 3, limiting the output, etc. Each unit will be described in detail below.

[0022] The detected speed processing unit 63 generates and outputs a rotation speed ω_fbk and a phase θ_fbk based on the rotation speed ωr of the shaft of the electric motor 3 detected by the rotation speed sensor 3A.

[0023] The rotational speed reference generating unit 61 generates a rotational speed reference ω_ref that defines the rotational speed of the electric motor 3 , and outputs the generated rotational speed reference ω_ref to a speed control unit 64 .

[0024] The speed control unit 64 generates a current reference Idq_ref based on the rotation speed reference ω_ref generated by the rotation speed reference generating unit 61 and the rotation speed ω_fbk output from the detected speed processing unit 63. The current reference Idq_ref is a vector value of the current reference Id_ref and the current reference Iq_ref in a rotor coordinate system having orthogonal dq axes. For example, the speed control unit 64 generates the current reference Idq_ref so that the difference between the rotation speed reference ω_ref and the rotation speed ω_fbk becomes 0 for each of the d-axis and q-axis components. As a result, the electric motor 3 is driven by the speed reference defined by the rotation speed reference ω_ref. The speed control unit 64 may perform field weakening control by a general method based on the rotation speed ω_fbk.

[0025] The detected current processing unit 65 outputs a current value Iuvw_fbk based on the load current detected by the current detector 70. The current value Iuvw_fbk indicates the phase currents Iu_fbk, Iv_fbk, Iw_fbk of the motor 3 as vector values ​​in a three-phase coordinate space having three axes corresponding to the U-phase, V-phase, and W-phase. A coordinate system (stator coordinate system) that defines this three-phase coordinate space is called a three-phase stationary coordinate system.

[0026] The coordinate conversion unit 66 converts the current value Iuvw_fbk in the three-phase stationary coordinate system into a rotor coordinate system that has dq axes and is defined as a two-phase coordinate system, using the phase θ_fbk, to generate a current value Idq_fbk. This is called dq conversion. The rotor coordinate system that has dq axes rotates to a position where the angle between the axis in the U-phase direction of the three-phase stationary coordinate system and the d-axis is equal to the phase θ_fbk.

[0027] The current control unit 67 generates a voltage reference Vdq_ref based on the current reference Idq_ref generated by the speed control unit 64 and the current value Idq_fbk output from the coordinate conversion unit 66 so that the difference between the components of each axis of the current reference Idq_ref and the current value Idq_fbk is zero.

[0028] The inverse coordinate transformation unit 68 transforms the voltage reference Vdq_ref generated by the current control unit 67 from the rotor coordinate system to a three-phase stationary coordinate system using the phase θ_fbk to generate the voltage reference Vuvw_ref. In other words, the inverse coordinate transformation unit 68 performs an inverse transformation (dq inverse transformation) of the above-mentioned dq transformation on the voltage reference Vdq_ref to generate the voltage reference Vuvw_ref.

[0029] The PWM controller 69 compares the voltage reference Vuvw_ref generated by the inverse coordinate conversion unit 68 with the carrier signal car of a predetermined frequency to generate a PWM signal for each arm constituting the inverter 50. The PWM controller 69 supplies the PWM signal for each arm to the inverter 50 to control the switching of the switching element 50S of each arm.

[0030] Next, an example of the PWM controller 69 will be described with reference to Fig. 1C. Fig. 1C is a configuration diagram of the PWM controller 69 according to an embodiment.

[0031] The PWM controller 69 includes, for example, a carrier signal generating unit 691, comparators 692U, 692V, and 692W, gate pulse generating units 693U, 693V, 693W, 693X, 693Y, and 693Z, and NOT circuits 694X, 694Y, and 694Z.

[0032] The carrier signal generating unit 691 generates a triangular wave with a predetermined period and outputs it as a PWM controlled carrier signal car. For example, this carrier signal car may be common to each phase.

[0033] First, the U phase will be described. The comparator 692U compares the amplitude of the U phase component (voltage reference Vu_ref) of the voltage reference Vuvw_ref with the amplitude of the carrier signal car as a reference. The gate pulse generating unit 693U generates a gate pulse for the upper arm 50U based on the determination result of the comparator 692. The gate pulse generating unit 693X generates a gate pulse for the lower arm 50X based on the determination result of the comparator 692 whose logic has been inverted by the NOT circuit 694X.

[0034] As with the U phase, for the V phase, the comparator 692V compares the amplitude of the V phase component (voltage reference Vv_ref) with the amplitude of the carrier signal car as a reference. A gate pulse generator 693V generates a gate pulse for the upper arm 50V based on the determination result of the comparator 692. A gate pulse generator 693Y generates a gate pulse for the lower arm 50Y based on the determination result of the comparator 692 whose logic has been inverted by a NOT circuit 694Y.

[0035] As with the U phase, for the W phase, the comparator 692W compares the amplitude of the W phase component (voltage reference Vw_ref) with the amplitude of the carrier signal car as a reference. The gate pulse generator 693W generates a gate pulse for the upper arm 50W based on the determination result of the comparator 692. The gate pulse generator 693Z generates a gate pulse for the lower arm 50Z based on the determination result of the comparator 692 whose logic has been inverted by the NOT circuit 694Z.

[0036] For example, the gate pulse generating unit 693U includes a delay circuit that delays the input pulse by a predetermined time, and an AND circuit. The gate pulse generating unit 693U secures the dead time by delaying the rising edge of the input positive pulse by the above-mentioned predetermined time using the delay circuit and shortening the period during which the upper arm 50U is turned ON using the AND circuit. The other gate pulse generating units also generate gate pulses with secured dead time in the same manner as the gate pulse generating unit 693U.

[0037] The PWM controller 69 ensures the balance of each phase by making the length of the dead time of each phase uniform. However, if the ON-state time of the switching element 50S of each phase becomes uneven due to some factor, the balance of each phase may be lost. The analysis processing unit 62 is configured to be able to detect such a state.

[0038] Next, the analysis processing unit 62 will be described with reference to Fig. 1D. Fig. 1D is a configuration diagram of the analysis processing unit 62 according to the embodiment.

[0039] The analysis processing unit 62 includes, for example, a storage unit 621 , a current value acquisition unit 622 , a coordinate conversion unit 623 , a mapping processing unit 624 , a state identification unit 625 , an output processing unit 626 , and a criterion generation unit 627 .

[0040] The storage unit 621 stores data such as current detection value data 621a of the current value I_fbk based on the load current detected by the current value acquisition unit 622, current reference value data 621b, mapping image data 621c, judgment criterion data 621d, and analysis processing programs. The storage unit 621 stores the above current detection value data as time-series data. Details of each of the above data will be described later.

[0041] The current value acquisition unit 622, the coordinate conversion unit 623, the mapping processing unit 624, the state identification unit 625, the output processing unit 626, and the judgment criterion generation unit 627 are each realized by, for example, a hardware processor such as a CPU (Central Processing Unit) 620 executing a program (software). In addition, some or all of these components may be realized by hardware (including circuitry) such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a GPU (Graphics Processing Unit), or may be realized by cooperation between software and hardware. The storage unit 621 is realized by, for example, a HDD (Hard Disk Drive), a flash memory, an EEPROM (Electrically Erasable Programmable Read Only Memory), a ROM (Read Only Memory), or a RAM (Random Access Memory).

[0042] The current value acquiring unit 622 acquires the current value Iuvw_fbk from the detected current processing unit 65, and adds the acquired data of the current value I_fbk as time-series data to current detection value data 621a in the storage unit 621. The current value acquiring unit 622 acquires the current reference Idq_ref from the speed control unit 64, and adds the acquired data of the current reference Idq_ref to current reference value data 621b in the storage unit 621 as time-series data.

[0043] Moreover, the current value acquiring unit 622 acquires a predetermined number of pieces of data on current values ​​Iuvw_fbk (detected values) corresponding to a predetermined period from the current detection value data 621a stored in the storage unit 621, and supplies the data to the coordinate conversion unit 623. The current value acquiring unit 622 acquires a predetermined number of pieces of data on current reference Idq_ref (command value) corresponding to the above-mentioned predetermined period from the current reference value data 621b stored in the storage unit 621, and supplies the data to the coordinate conversion unit 623. The current value acquiring unit 622 may set the above-mentioned data on current values ​​Iuvw_fbk and data on current values ​​Iuvw_fbk to be data for a common period.

[0044] The current value acquiring unit 622 may acquire the current value Iuvw_fbk from the detected current processing unit 65 and read the data of the current value Iuvw_fbk from the storage unit 621 in parallel. The current value acquiring unit 622 may acquire the current reference Idq_ref from the speed control unit 64 and read the data of the current reference Idq_ref from the storage unit 621 in parallel.

[0045] The coordinate converter 623 converts the current value Iuvw_fbk in the three-phase stationary coordinate system into a two-phase stationary coordinate system having αβ coordinate axes using the phase θ_offset to generate a current value Iab_fbk. For example, when the phase θ_offset is set to 0, the conversion process by the coordinate converter 623 is equivalent to the Clarke conversion.

[0046] The coordinate conversion unit 623 converts the current reference Idq_ref in the rotating coordinate system into the above-mentioned two-phase stationary coordinate system using a predetermined phase θ_offset to generate a current value Iab_fbk. The predetermined phase θ_offset is, for example, a multiple of 0° or 60°. The coordinate conversion unit 623 converts data of the current detection value into data of the αβ coordinate system using Clarke transformation for conversion from the above-mentioned three-phase stationary coordinate system to the αβ coordinate system.

[0047] The Clarke transformation of the current value Iuvw_fbk in the three-phase stationary coordinate system by the coordinate transformation unit 623 is a process of rotating the result of the Clarke transformation by the phase θ_offset around the origin common to the three-phase stationary coordinate system and the two-phase stationary coordinate system. This process is formulated in the following equation (1). "M" is the Clarke transformation matrix.

[0048] Iab_fbk = R(M[Iuvw_fbk], θ_offset) (1)

[0049] In addition, "R(·, θ_offset)" in the above formula (1) indicates a function that rotates the first variable (vector) according to the magnitude of the phase θ_offset. By the processing of this coordinate conversion unit 623, points on the three-phase coordinate plane indicated by the scalar values ​​for three phases are projected onto points on a plane including the αβ coordinate axes (referred to as the αβ coordinate plane) (see FIG. 2C).

[0050] The mapping processing unit 624, for example, obtains a predetermined number of data (detection values) of the current value Iuvw_fbk corresponding to a predetermined period from the current detection value data 621a stored in the storage unit 621, and projects it onto the αβ coordinate plane based on this. For example, the above-mentioned predetermined period is set to be longer than the time for the electric motor 3 to make one rotation, or to be equal in length to the time for one rotation. The mapping processing unit 624 generates an image on the αβ coordinate plane corresponding to that one rotation, and adds data of the generated image (referred to as image data) to the mapping image data 621c of the storage unit 621. For example, the data of the current value Iuvw_fbk (detection value of the current) mapped onto the αβ coordinate plane may include data corresponding to an angle range equal to or greater than a predetermined rotation angle in the mechanical angle of the electric motor 3. The above-mentioned angle range equal to or greater than the predetermined rotation angle may be at least 360° or more around the origin of the αβ coordinate.

[0051] More specifically, the following method may be implemented. For example, the mapping processor 624 projects the position indicated by the current value Iuvw_fbk of the time-series data onto the αβ coordinate plane. At this time, the mapping processor 624 adds time history information as time-series data to the position data of each point. Furthermore, based on the above result, the mapping processor 624 connects each point (points projected onto the αβ coordinate plane) in order using attribute information of the time-series data to generate a polygonal line trajectory (hereinafter, sometimes simply referred to as a trajectory) (see FIG. 2C).

[0052] The state identification unit 625 judges the image data stored in the mapping image data 621c of the storage unit 621 based on a predefined judgment rule. The state identification unit 625 causes the output processing unit 626 to output the judgment result. For example, the state identification unit 625 may add the judgment result as data in the storage unit 621 by the output processing unit 626, or may cause the output processing unit 626 to write the judgment result in an easily removable storage medium. The judgment result data added to the storage unit 621 can be referenced by each unit in the control unit 60. Each unit in the control unit 60 may perform necessary control based on the judgment result.

[0053] (Detection results under normal conditions) The detection result in a normal state will be described with reference to FIG. 2A to FIG. 2C. Fig. 2A is a diagram for explaining the relationship between a three-phase signal and a two-phase signal in a normal state of the embodiment. Fig. 2B is a diagram for explaining the relationship between a two-phase signal and a corresponding command value in a normal state of the embodiment. Fig. 2C is a diagram for explaining an image of a detection result in a normal state of the embodiment.

[0054] The waveforms SU, SV, and SW shown in (a) of FIG. 2A indicate three-phase signals (current value Iuvw_fbk) of the U, V, and W phases. Although the waveforms SU, SV, and SW contain fluctuations due to switching noise, the waveforms due to the main components are generally sinusoidal waves. The waveforms SA and SB shown in (b) of FIG. 2A indicate two-phase signals (current value Iab_fbk) obtained by αβ conversion of the three-phase signals (current value Iuvw_fbk). The waveforms SA and SB are waveforms with a phase difference of 90°. Although the waveforms SA and SB contain fluctuations due to switching noise, they are generally sinusoidal waves.

[0055] The waveforms SA and SB shown in (a) of Fig. 2B are the same as the waveforms SA and SB shown in (b) of Fig. 2A. The waveforms SAr and SBr shown in (b) of Fig. 2B are reference waveforms corresponding to the waveforms SA and SB shown in (a) of Fig. 2B. The waveforms SAr and SBr may be, for example, a current reference Iab_ref.

[0056] The image shown in FIG. 2C shows an αβ coordinate plane and a trajectory projected onto the plane. For example, the α axis is assigned to the horizontal axis, and the β axis is assigned to the vertical axis. The trajectories shown in FIG. 2C are trajectory Ln corresponding to the waveforms SA and SB shown in FIG. 2B (a), and trajectory Lr corresponding to the waveforms SAr and SBr shown in FIG. 2B (b). If the waveforms SAr and SBr are sine waves that are free of distortion and perpendicular to each other, the trajectory Lr based on these will be a perfect circle. In contrast, the trajectory Ln is based on the waveforms SA and SB, which contain switching noise fluctuations, and therefore is not a perfect circle. As shown in the figure, the trajectory Ln is a figure made by connecting fine line segments corresponding to the PWM control period. Depending on the angle from the α axis centered on the origin, the width away from a perfect circle changes, but symmetry with respect to the origin is generally maintained.

[0057] Next, referring to Figures 3A to 3C, a current waveform and its corresponding trajectory when a specific gate pulse is intentionally distorted will be described. Figures 3A to 3C are diagrams for explaining a current waveform and its corresponding trajectory when a specific gate pulse is intentionally distorted. This trajectory is a model of an abnormal state that occurs during the dead time of the gate pulse.

[0058] The waveforms SAu and SBu shown in FIG. 3A(a) are obtained when the dead time of the gate pulse for the upper arm 50U is intentionally lengthened to make the dead time of the gate pulse for the other five arms the same as that shown in FIG. 2B. For example, the dead time of the former is twice that of the latter. This state reproduces a state in which an abnormality occurs in the dead time of the upper arm 50U. The waveforms SAr and SBr shown in FIG. 3A(b) are the same as the waveforms SAr and SBr shown in FIG. 2B(b).

[0059] The configuration of the images shown in Figures 3B and 3C is the same as that of the image shown in Figure 2C. The trajectories drawn on them are different. Figures 3B and 3C show two examples in which an abnormality occurs in the dead time in different phases, and the differences between them will be discussed.

[0060] For example, the locus shown in FIG. 3B is the locus Lu corresponding to the waveforms SAu and SBu shown in FIG. 3A(a) and the locus Lr described above. Here, the locus Lr is considered to be a perfect circle. The locus Lu includes deviations from the locus Lr (perfect circle) due to minute fluctuations caused by fluctuations in switching noise. Furthermore, unlike the above-mentioned normal locus Ln, the center of the locus Lu is far from the origin and symmetry with respect to the origin is not maintained. In this case, it can be seen that the entire locus Lu is displaced in the negative direction of the α axis with respect to the locus Lr (perfect circle) to the extent that it can be read by the naked eye.

[0061] The loci shown in Fig. 3C are the locus Lz and the locus Lr described above. The locus Lz reproduces a state in which an abnormality occurs in the dead time of the lower arm 50Z. The entire locus Lz is displaced from the negative direction of the α axis to the negative direction of the β axis with respect to a perfect circle.

[0062] When the locus Lu shown in FIG. 3B is compared with the locus Lz shown in FIG. 3C, it can be seen that the directions in which the loci are displaced with respect to the perfect circle are different.

[0063] The determination process of the embodiment will be described with reference to Fig. 4 and Fig. 5. Fig. 4 is a flowchart related to the determination process of the embodiment. Fig. 5 is a flowchart related to the generation process of the determination criterion of the embodiment.

[0064] First, prior to the analysis process for detecting the state during the movement, which will be described later, the analysis processing unit 62 determines a judgment criterion for the analysis process (step S10) and adds it to the judgment criterion data 621d. After the judgment criterion is determined, the analysis processing unit 62 performs the analysis process for detecting the state during the movement by the coordinate conversion unit 623 and the mapping processing unit 624 (step S20). This analysis process includes a mapping process for the αβ coordinate plane.

[0065] The state identification unit 625 uses the above-mentioned criteria to determine whether or not an abnormal state has been detected based on the mapping result of the detection value data (step S30). If an abnormal state is detected, the output processing unit 626 outputs a detection result indicating that an abnormal state has been detected (step S40) and ends the series of processes. If an abnormal state is not detected, the analysis processing unit 62 similarly ends the series of processes, for example, by omitting output of the detection result of the abnormal state.

[0066] The process of generating the judgment criterion in step S10 above may be performed according to the procedure shown in FIG. 5. The magnitude of the detected current changes successively depending on the control state. Therefore, it is difficult to fix the judgment criterion. Therefore, in this embodiment, a method of optimizing the judgment criterion based on the detection result of the current will be described.

[0067] First, the current value acquisition unit 622 acquires current data (step S11). The coordinate conversion unit 623 performs coordinate conversion processing on the acquired current data (detection value data and command value data) (step S12). The mapping processing unit 624 performs mapping processing on the point indicated by the result of the coordinate conversion processing onto the coordinate plane for generating a criterion (step S13). The criterion generation unit 627 generates a criterion based on the result of the mapping processing in accordance with a predetermined type of criterion (step S14) and adds the criterion to the criterion data 621d. This completes the process of generating the criterion.

[0068] The generation of the judgment criteria will be described more specifically below. The judgment criteria include a criterion (called a first judgment criterion) that is not dependent on the phase of the electric motor 3, and a criterion (called a second judgment criterion) that utilizes the phase of the electric motor 3. These will be described in order.

[0069] FIG. 6 is a diagram for explaining the first criterion and a method for generating the first criterion according to the embodiment. For example, by using the first judgment criterion shown in FIG. 6 for identification, it is possible to detect a state in which symmetry with respect to the origin of the coordinate system is not maintained.

[0070] For example, assume that the trajectory indicating the control target is a circle with a predetermined radius r. The above judgment rule can be determined using two concentric circles (circle Lr1 and circle Lr2) with respect to the circle indicating the control target. Circle Lr1 has a radius r1 that is longer than the radius r. Circle Lr2 has a radius r2 that is shorter than the radius r. For example, a judgment rule may be used that judges that an abnormal state exists when the trajectory extends beyond the circumference of circle Lr1 or when the trajectory extends within the circumference of circle Lr2.

[0071] According to the above embodiment, this first criterion is used. The coordinate conversion unit 623 converts data on the detection value of the current flowing through the windings of the motor 3 from a three-phase stationary coordinate system having U-phase, V-phase and W-phase axes to an αβ coordinate system (two-phase stationary coordinate system). The mapping processing unit 624 maps the result of the αβ coordinate conversion onto an αβ coordinate plane (two-phase stationary coordinate plane). The state identification unit 625 identifies the driving state of the motor 3 using a predetermined criterion and the result mapped by the mapping processing unit 624. This makes it possible to identify that an operating state different from normal has occurred in the operating state of the multiple switching elements 50S.

[0072] (Modification of the first embodiment) A modified example of the first embodiment will be described. In the above embodiment, the detection of an abnormal state is described by identifying the trajectory of one revolution of the motor 3 using the first judgment criterion. In this modified example, an identification method (second judgment criterion) is used to detect an abnormal state by comparing an abnormality occurring in a specific phase of the motor 3 with other phases and identifying the trajectory. This example will be described below.

[0073] The second criterion and a method for generating the second criterion according to the embodiment will be described with reference to FIGS.

[0074] 7 is a diagram for explaining the phase relationship of the current (vector) of each phase of the electric motor 3 of the embodiment. If the axis pointing upward in this figure is assigned to the U phase of the electric motor 3, the currents (vectors) of each phase of the X, V, Y, W, and Z phases are arranged in 60° increments clockwise.

[0075] The characteristics appearing in the current flowing from the inverter 50 to each phase are due to the operation of the three phases and the six switching elements 50S, and these include characteristics related to the directions of six vectors. In other words, when the characteristics appearing on the trajectory due to an abnormality in each of the six switching elements 50S are rotated by 60 degrees, the trajectory has the same tendency as an abnormality occurring in another semiconductor switch. In other words, by rotating the obtained trajectory by 60 degrees and comparing it with the trajectories in the ranges of other phases, the difference from the state in the other phases can be identified.

[0076] For example, when determining whether an abnormal state has been detected in step S30 of Fig. 4 described above, the state identification unit 625 may identify the difference from the state of other phases by processing such as rotating the trajectory in the specific direction by 60° and comparing it with the trajectories of other phases. The specific direction may be determined, for example, as the U-phase direction, and a trajectory within a range of ±30° around the axis of the U-phase direction may be selected as a trajectory in a predetermined range. This is determined as a second determination criterion and used for comparison with the trajectories of other phases.

[0077] If there is no abnormality in each phase, the trajectory in the range defined by the second criterion has a high similarity to the trajectories of the other phases. If there is an abnormality in any of the phases, there will be a direction that is determined to have a low similarity to the trajectories of the other phases in the discrimination process using the second criterion.

[0078] The state identification unit 625 may determine whether or not an abnormal state has been detected based on the result of the identification process using the second determination criterion. If a difference is found in a specific phase in the comparison result of the trajectories (if there is a phase with low correlation), it may be determined that there is an abnormal state related to that phase.

[0079] 8 is a diagram for explaining a circuit module that executes a rotation process according to a modified example. For example, the state identification unit 625 includes a circuit module that executes the rotation process. The transformation formula of this circuit module is equivalent to that for so-called dq transformation, and is different in that the reference phase is a fixed value.

[0080] For example, FIG. 8(a) shows a circuit module for obtaining a trajectory based on the axis of a reference phase (e.g., U-phase). The current value Iuvw_fbk of each phase of a three-phase AC is supplied to the input of this circuit module. The current value Iab_fbk converted into orthogonal α-axis and β-axis components is output as the output of this circuit module. In the configuration shown in FIG. 8(a), the reference phase of this circuit module is fixed at 0°. In the configuration shown in FIG. 8(b), the reference phase of this circuit module is set to an angle of a multiple of 60° in 60° increments. By adjusting the reference phase of the circuit module in this way, it is possible to convert to a different angle. This is just one example, and the rotation process may be performed by an arithmetic method.

[0081] 9A and 9B are diagrams for explaining the results of the rotation process on the trajectory of the modified example. The diagrams (a1) to (f1) arranged in the left column of each row of Figures 9A and 9B show trajectories before rotation processing is performed on the trajectories, and show trajectories when simulating abnormalities in the U phase, X phase, V phase, Y phase, W phase, and Z phase, respectively. The diagrams (a2) to (f2) arranged in the right column of each row of Figures 9A and 9B show the results of rotation processing of a predetermined angle determined for each phase on the trajectories shown in each of the diagrams (a1) to (f1). For example, the above-mentioned predetermined angles are 0°, 60°, 120°, 180°, 240°, and 300° clockwise.

[0082] In each of the figures (a1) to (f1) above, the characteristic points that indicate an abnormal state are marked with circles to indicate their positions. The positions of the circles shown in each of the figures (a1) to (f1) above are in various directions with the origin as the reference point. In contrast, the positions of the circles shown in each of the figures (a2) to (f2) above are aligned in a common direction from the origin.

[0083] In the above embodiment, an example was shown in which analysis was performed using data in the αβ coordinate system obtained by Clarke transformation. However, the present invention is not limited to this, and data in the rotor coordinate system after a so-called dq transformation may be used depending on the purpose of analysis.

[0084] However, when using data in the αβ coordinate system, unlike the so-called dq transformation, a reference phase that indicates the direction of the rotating magnetic field is not required, and analysis can be performed without being affected by the accuracy of the reference phase. For example, the winding current obtained by PWMing the voltage reference after the inverse dq transformation using the reference phase can be affected by the accuracy of the reference phase that is generated sequentially. By using data in the αβ coordinate system, analysis can be performed without being affected by the accuracy of such a reference phase.

[0085] Second embodiment A second embodiment will be described. In the first embodiment, a case where a method for generating a criterion for an expected abnormality is determined in advance by a user and an abnormal state is detected using the criterion based on the method has been described. In the second embodiment, a case where a model optimized by machine learning is used to detect an abnormal state will be described.

[0086] The power conversion device 2 of this embodiment includes a state identifying unit 625A and a learning processing unit 627A, instead of the state identifying unit 625 and the criterion generating unit 627 in the power conversion device 2 of the first embodiment.

[0087] The state identification unit 625A of the embodiment detects features appearing in the trajectory by applying an image recognition technique. Fig. 10 is a diagram for explaining the determination process of the second embodiment.

[0088] For example, the state identification unit 625A includes a deep neural network (DNN) that performs two-dimensional image recognition processing. The DNN is an example of an artificial neural network.

[0089] The DNN in the state identification unit 625A shown in Fig. 10 has a configuration similar to that for image processing applied to the comparative example, such as recognizing handwritten characters. This DNN, for example, performs image recognition processing on the input image IMG0, making it possible to recognize and classify handwritten characters in the input image IMG0. Note that the adjustment of the parameters of this DNN differs between the case where it is applied to this embodiment and the case where it is applied to the recognition of handwritten characters.

[0090] For example, this DNN includes an input layer ILA, a pooling layer PLA, an intermediate layer ML, and an output layer OL. Each of the input layer ILA, the pooling layer PLA, the intermediate layer ML, and the output layer OL includes a plurality of artificial neurons (simply referred to as neurons).

[0091] The neurons in the input layer ILA are associated with each pixel that constitutes the input image IMG1, and the data of each pixel is used as input data. Each pixel shown in Fig. 10 has a square shape and is arranged in a lattice pattern.

[0092] For example, a pooling layer PLA is provided behind the input layer ILA. The pooling layer PLA performs pooling using a filter FLA while maintaining a two-dimensional structure corresponding to the input image IMG1. The filter FLA used for this pooling may be configured with a total of five pixels including the center and the pixels above, below, left and right, or may be configured with a total of nine pixels including the center and its periphery. The filter FLA shown in FIG. 10 is configured with a total of nine pixels. The number of pooling layers PLA is not limited to one, and may be appropriately provided in accordance with the number of types of filters FLA used. The characteristics of the filter FLA are patterned by the values ​​of pixels in the filter FLA. The number of types of filters FLA corresponds to the number of patterns that define different characteristics.

[0093] For example, the pooling layer PLA generates data FDA00 by pooling the data of the top left 9 pixels of the input image IMG1 with the filter FLA. The pooling layer PLA performs pooling on each range of data by shifting the range of the calculation target using the filter FLA. For example, at the end of processing the entire input image IMG1, the pooling layer PLA generates data FDAmn by pooling the data of the bottom right 9 pixels of the input image IMG1 with the filter FLA.

[0094] Each data calculated by the pooling layer PLA becomes input data for the next hidden layer ML. The hidden layer ML has multiple layers, such as hidden layers ML1, ML2, etc. The number of neurons in these layers and the number of layers in the hidden layer ML may be determined appropriately, and each layer in the hidden layer ML may be configured with a fully connected layer.

[0095] For example, the neurons in the front pooling layer PLA are partially connected to the neurons in the hidden layer ML1, which limits the connections to a certain range. By providing the pooling layer PLA, the effect of misalignment of the detected object in the image can be reduced.

[0096] Each piece of data calculated by the hidden layer ML becomes input data for the next output layer OL. The number of neurons in the output layer may be determined appropriately, and the hidden layer ML is composed of a fully connected layer.

[0097] As another example of the above configuration, it is possible to configure the system without providing a pooling layer. For example, in this case, data of each pixel is associated with neurons in the input layer IM arranged in one dimension. The neurons in the input layer IM form a fully connected layer connected to neurons in the subsequent hidden layer ML. In the following explanation, an example in which a pooling layer PL is provided will be described.

[0098] In the case of the comparative example, the character (number) drawn in the input image IMG0 is identified. For example, the binary image of the number 5 written by hand or the shade of the shade image is converted into a numerical value for each pixel and associated with each pixel. The trained DNN identifies the difference between the character "5" and other numbers and activates the output of the fifth neuron in the output layer OL.

[0099] The state identification unit 625A of this embodiment applies a configuration similar to that for handwritten character recognition as shown in FIG. 10, for example, by configuring the output layer with six neurons and identifying the characteristics of the trajectory in the image, it is possible to divide the results of the calculation into six phases and output them.

[0100] The results of classification by the DNN are not limited to classification for each of the six phases as described above. The DNN may be configured to calculate the presence or absence of one or more types of abnormal states, or the probability of reaching an abnormal state, based on the data of an image in which at least one trajectory is drawn.

[0101] For example, the trajectory drawn on the αβ coordinate plane shows different trends depending on the factors related to the abnormal state. The DNN of the state identification unit 625A can detect the presence or absence of an abnormal state or the increasing probability of reaching an abnormal state by detecting this trend.

[0102] In this way, by generating the judgment criteria using a machine learning technique, it becomes possible to detect abnormal states due to various factors. For example, the learning processing unit 627A generates the judgment criteria using a machine learning technique. As the teacher data, for example, a set of data in which the data (image data) of an image showing a trajectory drawn on an αβ coordinate plane is used as input data, and the information on the state shown by this trajectory is used as output data is used.

[0103] For example, the analysis may identify, based on the trajectory, one or more of the following events: The driving status of the motor 3 is classified into two types: "normal" and "abnormal." The probability of the driving state of the motor 3 being “normal” or “abnormal” is estimated. The driving status of the motor 3 is classified into two cases for each of the six phases: “normal” and “abnormal.” As the driving status of the motor 3, the probability that each of the six phases is “normal” or “abnormal” is estimated. As the driving state of Motor 3, the phase that is most likely to be “abnormal” is estimated among the six phases.

[0104] In this way, the configuration of the output layer OL of the DNN may be determined according to the above events. Note that the teacher data includes a plurality of pairs of input data, which is an image showing a trajectory, and output data related to the above events.

[0105] The learning processing unit 627A uses the above teacher data to optimize the parameters that determine the characteristics of the DNN (model) through machine learning, thereby determining the model to be used as the judgment criterion. General methods may be applied as the learning method of the DNN and the optimization method during machine learning. The parameters that determine the characteristics of the DNN are a collective term for weight values ​​for data input to each neuron, offsets to be added to the calculation results of each neuron, etc.

[0106] According to the above embodiment, abnormal states can be detected by applying machine learning techniques.

[0107] For example, when square unit cells are arranged in a honeycomb pattern on the αβ coordinate plane as described above, the state identification unit 625A detects that the position indicated by the result of the coordinate transformation is included within the above unit cell, and identifies the driving state of the electric motor 3 using the result of this detection.

[0108] The state identification unit 625A uses a learned model with identified parameters as a criterion for identifying the tendency of the driving state of the motor 3 to change between the first state and the second state. For example, a normal operating state is an example of the first state. An operating state in which an abnormal state occurs is an example of the second state. The above-mentioned second state includes a state in which the balance of the currents flowing through the multiple switching elements 50S is lost. The state identification unit 625A may use this model to identify whether an abnormal state has occurred in the driving state of the motor 3 or whether it is in a state in which an abnormal state may occur.

[0109] In this case, the learning processing unit 627A may use the first learning data indicating either the first state or the second state for machine learning of the model. The first learning data includes both the learning data indicating the first state and the learning data indicating the second state, so that efficient learning can be achieved.

[0110] This first learning data may include a pattern of drive signals for driving each of the multiple switching elements of the inverter 50 that drives the electric motor 3 so as to simulate the above-mentioned second state. Note that the features appearing in the trajectory drawn on the αβ coordinate plane include an event in which the trajectory is displaced in a specific direction as shown in the above-mentioned first embodiment. Events detected as events of the second state are not limited to this, and may include abnormal states caused by other factors.

[0111] (First modified example of the second embodiment) A first modified example of the second embodiment will be described. When the electric motor 3 is the object to be controlled, the locus drawn on the αβ coordinate plane has rotational symmetry with respect to the origin. The rotational symmetry of the locus is not non-directional but has a correlation with the directions of each phase, as shown in Figures 9A and 9B.

[0112] Therefore, a method for improving the efficiency of machine learning by utilizing the rotational symmetry of the trajectory is proposed. FIG. 11 is a diagram for explaining the use of the rotational symmetry of the trajectory in the modified example. FIG. 11(a) shows a case where the pixels are assigned to each pixel in the image IMG1 in FIG. 10 described above. In contrast, FIG. 11(b) illustrates a case where the lattice defining the arrangement of each pixel is rotated by a predetermined angle in a direction determined based on the characteristics of the rotational symmetry of the trajectory when the image IMG2 is imported into the DNN. For example, the image IMG1 includes a trajectory that characterizes an abnormal state that has occurred in the U phase. The image IMG2 includes a trajectory that characterizes an abnormal state that has occurred in the Z phase. In this way, by rotating the lattice in a predetermined direction and importing the image into the DNN, learning data that utilizes the characteristics of the rotational symmetry of the trajectory can be generated.

[0113] FIG. 12 is a diagram for explaining the determination process of the first modified example of the second embodiment. The DNN of the state identification unit 625A shown in Fig. 12 corresponds to that shown in Fig. 10. In the case of Fig. 12, the method of pixel division of image data used as input data for the DNN and the method of using this as learning data are different. This point will be mainly explained.

[0114] The learning processing unit 627A uses the images IMG1 and IMG2 as learning data for this DNN. The images IMG1 and IMG2 are examples of learning data, and may be combined with other learning data. The learning processing unit 627A uses the image IMG1 as learning data without performing a rotation process on it, and uses the result of performing a 60° rotation process on the image IMG2 as learning data.

[0115] For example, the learning processing unit 627A may generate image data (second learning data) by rotating and transforming the above image IMG2 (first learning data) according to a predetermined rule related to the αβ coordinate plane, and use the image data (second learning data) for machine learning of the above model. For example, the above predetermined rule may include the position of the center of rotation, a rotation angle associated with the phase in which an abnormal state occurs, and the like. More specifically, for example, in the case of image IMG2 in which an abnormality is found in the Z phase, the learning processing unit 627A rotates the lattice by 60° counterclockwise. The relationship between the phase and the angle may be determined in advance.

[0116] After the model is machine-learned as described above, the state identification unit 625A may identify the driving state of the electric motor 3 using the model.

[0117] According to this modification, the learning processing unit 627A can generate learning data that simulates an abnormality occurring in a specific direction by adding a rotation process that rotates an input image (e.g., image IMG2) to be identified by the DNN by a predetermined angle in the front stage of the DNN. The learning processing unit 627A may perform the learning process of such a method for one specific direction, or may perform it for each phase direction.

[0118] (Second modified example of the second embodiment) A second modified example of the second embodiment will be described. In the second embodiment, a case where an input image to a DNN is divided into quadrilateral (square) pixels will be described. In this modified example, a case where an input image to a DNN is divided into hexagonal (regular hexagonal) pixels will be described.

[0119] First, with reference to (c) and (d) of FIG. 11, a case where an input image of a DNN is divided into hexagonal pixels will be described. The images (IMG1, IMG2) shown in (c) and (d) of FIG. 11 are the same as the images shown in (a) and (b) of FIG. 11 above, but the shape of the lattice that divides them into pixels is different. In the cases of (a) and (b) of FIG. 11, it is a quadrilateral (square), but in (c) and (d) of FIG. 11, it is a hexagon (regular hexagon). When lattices (unit lattices) are densely arranged in this way, that is, when arranged in a honeycomb shape, the shape of the unit lattice is limited to a triangle, a square, or a regular hexagon.

[0120] For example, if the lattice is a quadrilateral (square), the angles at which the lattices overlap when rotated are every 90°, but if it is a regular hexagon, the angles at which the lattices overlap when rotated are every 60°. If the center of rotation is chosen appropriately, the lattice after rotating a regular hexagonal lattice by 60° will be the same as the lattice before rotation. When a quadrilateral (square) lattice is rotated by 60°, areas will arise where the lattices do not overlap, and so a certain amount of information will be lost when identifying whether or not a trajectory exists. This does not occur when a hexagonal lattice is rotated by 60°.

[0121] FIG. 13 is a diagram for explaining a determination process according to a modified example of the second embodiment. The power converter 2 in this modification includes a state identifying unit 625B instead of the state identifying unit 625A described above.

[0122] The state identification unit 625B of the embodiment may be the same as the state identification unit 625A described above, except for the shape of the pixels and the configuration of the filters. Accordingly, the input layer ILA, the pooling layer PLA, and the filter FLA are replaced by the input layer ILB, the pooling layer PLB, and the filter FLB, respectively. The shape of the pixels of the image related to the input layer ILB, the pooling layer PLB, and the filter FLB is hexagonal.

[0123] For example, the neurons of the input layer IMB are associated with hexagonal unit cells arranged in a honeycomb pattern. The filter FLB used in the pooling process of the pooling layer PLB is defined to include the first hexagonal unit cell arranged in a honeycomb pattern and six (multiple) second unit cells surrounding the first unit cell as objects of calculation. The pooling process by the pooling layer PLB is performed using the above filter FLB.

[0124] When regular hexagonal unit cells are arranged in a honeycomb pattern on the αβ coordinate plane as described above, the state identification unit 625B detects that the position indicated by the result of the coordinate transformation is included within the above unit cell, and identifies the driving state of the electric motor 3 using the result of this detection.

[0125] In the case of the inverter 50 using six switching elements 50S as described above, the directionality has a relatively strong meaning as described above. Since the characteristics related to this direction appear every 60 degrees, by utilizing this directionality, it is possible to treat 60 degrees as one unit.

[0126] According to at least one of the embodiments described above, the drive state identification device of the embodiment includes a coordinate conversion unit 623, a mapping processing unit 624, and a state identification unit 625. The coordinate conversion unit 623 converts data of the detection value of the current flowing through the windings of the three-phase AC motor from a three-phase stationary coordinate system to a two-phase stationary coordinate system. The mapping processing unit 624 maps the result of the coordinate conversion onto a two-phase stationary coordinate plane. The state identification unit 625 identifies the drive state of the motor 3 using a judgment criterion and the mapped result. This makes it possible to identify that an operating state different from normal has occurred in the operating state of a plurality of switching elements.

[0127] Although some 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 in the scope of the invention and its equivalents described in the claims, as well as in the scope and spirit of the invention. For example, instead of the above DNN, a convolutional neural network (CNN) having a convolutional layer that performs convolution processing before the pooling layer may be used. [Explanation of symbols]

[0128] 1...motor drive system, 2...power conversion device, 3...electric motor, 4...mechanical load, 50 inverse conversion device, 60...control unit, 61...rotational speed reference generation unit, 62...analysis processing unit, 64...speed control unit, 65...detected current processing unit, 69...PWM controller, 70 current detector, 621...storage unit, 622...current value acquisition unit, 623...coordinate conversion unit, 624...mapping processing unit, 625...state identification unit, 626...output processing unit, 627...criterion generation unit, 621a...current detection value data, 621b...current reference value data, 621c...mapping image data, 621d...criterion data

Claims

1. A current command value for adjusting a current flowing through a winding of a three-phase AC motor by controlling a switching element of a power converter and a detection value of a current flowing through the winding of the motor are represented by data in a stationary coordinate system; a state identification unit that identifies the occurrence of an operating state of the switching element that is different from a normal operating state based on a difference between a trajectory of the command value of the current mapped on a stationary coordinate plane and a trajectory of the detected value of the current; A driving state identification device comprising:

2. The data of the current detection value mapped onto the stationary coordinate plane includes data of an angle range of at least 360° or more in the mechanical angle of the electric motor. The driving condition identification device according to claim 1 .

3. The switching elements of the power converter are controlled by vector control to detect the current flowing through the windings of a three-phase AC motor and express it as a vector in a stationary coordinate system. a state identification unit which compares a result of rotating a vector of a specific phase of a three-phase AC by a predetermined angle with the direction of the vector of another phase on a stationary coordinate plane, and identifies that an operating state different from normal has occurred in the operating state of the switching element when a difference between the direction of the result of rotating the vector of the specific phase by the predetermined angle and the direction of the vector of another phase falls outside a predetermined range. A driving state identification device comprising:

4. the predetermined angle is 30 degrees, The predetermined range is 60 degrees. The driving condition identification device according to claim 3 .

5. The switching elements of the power converter are controlled by vector control to detect the current flowing through the windings of a three-phase AC motor and display it as data in a stationary coordinate system. a learning processing unit that determines parameters of a model used to identify an operating state of the switching element by learning using a trajectory of the detection value of the current mapped onto a stationary coordinate plane; a state identification unit that identifies that an operating state different from a normal operating state has occurred in the operating state of the switching element by using the model in which the parameters have been identified and a trajectory of the detection value of the current; A driving state identification device comprising:

6. The learning processing unit: a first learning data indicating an operation state of the switching element including an operation state different from the normal operation state is used for machine learning of the model; The driving condition identification device according to claim 5 .

7. The learning processing unit: generating second learning data by performing rotational transformation on the first learning data in accordance with a predetermined rule related to a plane of an αβ coordinate system (referred to as an αβ coordinate plane), which is a two-phase stationary coordinate system, and using the second learning data for machine learning of the model; The driving condition identification device according to claim 6.

8. The state identification unit is identifying a drive state of the electric motor using the model; The driving condition identification device according to claim 7.

9. A regular hexagonal unit cell is arranged in a honeycomb pattern on the αβ coordinate plane, The state identification unit is It is detected that the position indicated by the result of the coordinate transformation into the αβ coordinate system is included in the unit lattice, and the result of the detection is used to identify the occurrence of an operating state different from normal in an operating state of a switching element that adjusts a current flowing through a winding of the motor. The driving condition identification device according to claim 7.

10. The neurons in an input layer of the artificial neural network constituting the model are associated with the hexagonal unit cells arranged in the honeycomb pattern, The learning processing unit: performing a pooling process using a filter defined so as to include the hexagonal first unit lattice arranged in a honeycomb pattern and a plurality of second unit lattices located at positions surrounding the first unit lattice as objects of calculation; The driving condition identification device according to claim 9.

11. The power converter includes a plurality of switching elements, and drives the motor by passing currents through the windings of the motor, The operating state different from the normal state includes a state in which the balance of currents flowing through the multiple switching elements is lost. The driving condition identification device according to claim 5 or 6.

12. The first learning data includes a pattern of a drive signal for driving each of a plurality of switching elements of an inverter that drives the electric motor so as to simulate an operating state different from the normal operating state. The driving condition identification device according to claim 6.

13. A current command value for adjusting a current flowing through a winding of a three-phase AC motor by controlling a switching element of a power converter and a detection value of a current flowing through the winding of the motor are represented by data in a stationary coordinate system; a state identification unit that identifies that an operating state different from a normal operating state has occurred in the switching element based on a difference between a trajectory of the current command value and a trajectory of the current detection value mapped on a stationary coordinate plane; an inverter that drives the electric motor based on the identification result; A power conversion device comprising:

14. The switching elements of the power converter are controlled by vector control to detect the current flowing through the windings of a three-phase AC motor and express it as a vector in a stationary coordinate system. a state identification unit which compares a result of rotating a vector of a specific phase of a three-phase AC by a predetermined angle with the direction of the vector of another phase on a stationary coordinate plane, and identifies that an operating state different from normal has occurred in the operating state of the switching element when a difference between the direction of the result of rotating the vector of the specific phase by the predetermined angle and the direction of the vector of another phase falls outside a predetermined range; an inverter that drives the electric motor based on the identification result; A power conversion device comprising:

15. The switching elements of the power converter are controlled by vector control to detect the current flowing through the windings of a three-phase AC motor and display it as data in a stationary coordinate system. a learning processing unit that determines parameters of a model used to identify an operating state of the switching element by learning using a trajectory of the detection value of the current mapped onto a stationary coordinate plane; a state identification unit that identifies that an operating state different from a normal operating state has occurred in the operating state of the switching element by using the model in which the parameters have been identified and a trajectory of the detection value of the current; an inverter that drives the electric motor based on the identification result; A power conversion device comprising:

Citation Information

Patent Citations

  • Electric motor control system

    JP2011050214A

  • Voltage abnormality detection device and voltage abnormality detection method

    JP2019158858A

  • Electric power conversion device, rotating machine system, and diagnostic method

    JP2020114084A

  • Power conversion device, and method of determining faulty phase

    JP2022107438A